diff --git a/integration-tests/models/__snapshots__/test_idefics/test_idefics.json b/integration-tests/models/__snapshots__/test_idefics/test_idefics.json
new file mode 100644
index 00000000..1df5761d
--- /dev/null
+++ b/integration-tests/models/__snapshots__/test_idefics/test_idefics.json
@@ -0,0 +1,168 @@
+{
+  "details": {
+    "best_of_sequences": null,
+    "finish_reason": "length",
+    "generated_tokens": 10,
+    "prefill": [
+      {
+        "id": 1,
+        "logprob": null,
+        "text": "<s>"
+      },
+      {
+        "id": 4911,
+        "logprob": -5.3632812,
+        "text": "User"
+      },
+      {
+        "id": 29901,
+        "logprob": -0.00762558,
+        "text": ":"
+      },
+      {
+        "id": 32000,
+        "logprob": -0.7739258,
+        "text": "<fake_token_around_image>"
+      },
+      {
+        "id": 32001,
+        "logprob": -9.775162e-05,
+        "text": "<image>"
+      },
+      {
+        "id": 32000,
+        "logprob": -1.1920929e-07,
+        "text": "<fake_token_around_image>"
+      },
+      {
+        "id": 1815,
+        "logprob": -4.4140625,
+        "text": "Can"
+      },
+      {
+        "id": 366,
+        "logprob": -0.01436615,
+        "text": "you"
+      },
+      {
+        "id": 2649,
+        "logprob": -4.9414062,
+        "text": "tell"
+      },
+      {
+        "id": 592,
+        "logprob": -0.3005371,
+        "text": "me"
+      },
+      {
+        "id": 263,
+        "logprob": -3.5703125,
+        "text": "a"
+      },
+      {
+        "id": 1407,
+        "logprob": -9.4296875,
+        "text": "very"
+      },
+      {
+        "id": 3273,
+        "logprob": -1.9111328,
+        "text": "short"
+      },
+      {
+        "id": 5828,
+        "logprob": -0.28881836,
+        "text": "story"
+      },
+      {
+        "id": 2729,
+        "logprob": -3.4179688,
+        "text": "based"
+      },
+      {
+        "id": 373,
+        "logprob": -0.00056886673,
+        "text": "on"
+      },
+      {
+        "id": 278,
+        "logprob": -0.14123535,
+        "text": "the"
+      },
+      {
+        "id": 1967,
+        "logprob": -0.053985596,
+        "text": "image"
+      },
+      {
+        "id": 29973,
+        "logprob": -0.15771484,
+        "text": "?"
+      }
+    ],
+    "seed": null,
+    "tokens": [
+      {
+        "id": 32002,
+        "logprob": -0.004295349,
+        "special": true,
+        "text": "<end_of_utterance>"
+      },
+      {
+        "id": 29871,
+        "logprob": -7.43866e-05,
+        "special": false,
+        "text": " "
+      },
+      {
+        "id": 13,
+        "logprob": -2.3126602e-05,
+        "special": false,
+        "text": "\n"
+      },
+      {
+        "id": 7900,
+        "logprob": -3.9339066e-06,
+        "special": false,
+        "text": "Ass"
+      },
+      {
+        "id": 22137,
+        "logprob": 0.0,
+        "special": false,
+        "text": "istant"
+      },
+      {
+        "id": 29901,
+        "logprob": -2.6226044e-06,
+        "special": false,
+        "text": ":"
+      },
+      {
+        "id": 319,
+        "logprob": -0.87841797,
+        "special": false,
+        "text": " A"
+      },
+      {
+        "id": 521,
+        "logprob": -1.3837891,
+        "special": false,
+        "text": " ch"
+      },
+      {
+        "id": 21475,
+        "logprob": -0.00051641464,
+        "special": false,
+        "text": "icken"
+      },
+      {
+        "id": 338,
+        "logprob": -1.1435547,
+        "special": false,
+        "text": " is"
+      }
+    ]
+  },
+  "generated_text": "\nAssistant: A chicken is"
+}
diff --git a/integration-tests/models/__snapshots__/test_idefics/test_idefics_load.json b/integration-tests/models/__snapshots__/test_idefics/test_idefics_load.json
new file mode 100644
index 00000000..e0dfe96c
--- /dev/null
+++ b/integration-tests/models/__snapshots__/test_idefics/test_idefics_load.json
@@ -0,0 +1,674 @@
+[
+  {
+    "details": {
+      "best_of_sequences": null,
+      "finish_reason": "length",
+      "generated_tokens": 10,
+      "prefill": [
+        {
+          "id": 1,
+          "logprob": null,
+          "text": "<s>"
+        },
+        {
+          "id": 4911,
+          "logprob": -5.3476562,
+          "text": "User"
+        },
+        {
+          "id": 29901,
+          "logprob": -0.0075531006,
+          "text": ":"
+        },
+        {
+          "id": 32000,
+          "logprob": -0.7729492,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 32001,
+          "logprob": -9.787083e-05,
+          "text": "<image>"
+        },
+        {
+          "id": 32000,
+          "logprob": -1.1920929e-07,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 1815,
+          "logprob": -4.4296875,
+          "text": "Can"
+        },
+        {
+          "id": 366,
+          "logprob": -0.01424408,
+          "text": "you"
+        },
+        {
+          "id": 2649,
+          "logprob": -4.9335938,
+          "text": "tell"
+        },
+        {
+          "id": 592,
+          "logprob": -0.2993164,
+          "text": "me"
+        },
+        {
+          "id": 263,
+          "logprob": -3.5664062,
+          "text": "a"
+        },
+        {
+          "id": 1407,
+          "logprob": -9.4453125,
+          "text": "very"
+        },
+        {
+          "id": 3273,
+          "logprob": -1.9306641,
+          "text": "short"
+        },
+        {
+          "id": 5828,
+          "logprob": -0.2836914,
+          "text": "story"
+        },
+        {
+          "id": 2729,
+          "logprob": -3.4179688,
+          "text": "based"
+        },
+        {
+          "id": 373,
+          "logprob": -0.00056934357,
+          "text": "on"
+        },
+        {
+          "id": 278,
+          "logprob": -0.13928223,
+          "text": "the"
+        },
+        {
+          "id": 1967,
+          "logprob": -0.05355835,
+          "text": "image"
+        },
+        {
+          "id": 29973,
+          "logprob": -0.15771484,
+          "text": "?"
+        }
+      ],
+      "seed": null,
+      "tokens": [
+        {
+          "id": 32002,
+          "logprob": -0.004333496,
+          "special": true,
+          "text": "<end_of_utterance>"
+        },
+        {
+          "id": 29871,
+          "logprob": -7.426739e-05,
+          "special": false,
+          "text": " "
+        },
+        {
+          "id": 13,
+          "logprob": -2.348423e-05,
+          "special": false,
+          "text": "\n"
+        },
+        {
+          "id": 7900,
+          "logprob": -3.9339066e-06,
+          "special": false,
+          "text": "Ass"
+        },
+        {
+          "id": 22137,
+          "logprob": 0.0,
+          "special": false,
+          "text": "istant"
+        },
+        {
+          "id": 29901,
+          "logprob": -2.861023e-06,
+          "special": false,
+          "text": ":"
+        },
+        {
+          "id": 319,
+          "logprob": -0.8828125,
+          "special": false,
+          "text": " A"
+        },
+        {
+          "id": 521,
+          "logprob": -1.3759766,
+          "special": false,
+          "text": " ch"
+        },
+        {
+          "id": 21475,
+          "logprob": -0.0005083084,
+          "special": false,
+          "text": "icken"
+        },
+        {
+          "id": 338,
+          "logprob": -1.1367188,
+          "special": false,
+          "text": " is"
+        }
+      ]
+    },
+    "generated_text": "\nAssistant: A chicken is"
+  },
+  {
+    "details": {
+      "best_of_sequences": null,
+      "finish_reason": "length",
+      "generated_tokens": 10,
+      "prefill": [
+        {
+          "id": 1,
+          "logprob": null,
+          "text": "<s>"
+        },
+        {
+          "id": 4911,
+          "logprob": -5.3476562,
+          "text": "User"
+        },
+        {
+          "id": 29901,
+          "logprob": -0.0075531006,
+          "text": ":"
+        },
+        {
+          "id": 32000,
+          "logprob": -0.7729492,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 32001,
+          "logprob": -9.787083e-05,
+          "text": "<image>"
+        },
+        {
+          "id": 32000,
+          "logprob": -1.1920929e-07,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 1815,
+          "logprob": -4.4296875,
+          "text": "Can"
+        },
+        {
+          "id": 366,
+          "logprob": -0.01423645,
+          "text": "you"
+        },
+        {
+          "id": 2649,
+          "logprob": -4.9335938,
+          "text": "tell"
+        },
+        {
+          "id": 592,
+          "logprob": -0.2993164,
+          "text": "me"
+        },
+        {
+          "id": 263,
+          "logprob": -3.5664062,
+          "text": "a"
+        },
+        {
+          "id": 1407,
+          "logprob": -9.4453125,
+          "text": "very"
+        },
+        {
+          "id": 3273,
+          "logprob": -1.9306641,
+          "text": "short"
+        },
+        {
+          "id": 5828,
+          "logprob": -0.2836914,
+          "text": "story"
+        },
+        {
+          "id": 2729,
+          "logprob": -3.4179688,
+          "text": "based"
+        },
+        {
+          "id": 373,
+          "logprob": -0.00056934357,
+          "text": "on"
+        },
+        {
+          "id": 278,
+          "logprob": -0.13928223,
+          "text": "the"
+        },
+        {
+          "id": 1967,
+          "logprob": -0.05355835,
+          "text": "image"
+        },
+        {
+          "id": 29973,
+          "logprob": -0.15771484,
+          "text": "?"
+        }
+      ],
+      "seed": null,
+      "tokens": [
+        {
+          "id": 32002,
+          "logprob": -0.004333496,
+          "special": true,
+          "text": "<end_of_utterance>"
+        },
+        {
+          "id": 29871,
+          "logprob": -7.4505806e-05,
+          "special": false,
+          "text": " "
+        },
+        {
+          "id": 13,
+          "logprob": -2.3722649e-05,
+          "special": false,
+          "text": "\n"
+        },
+        {
+          "id": 7900,
+          "logprob": -3.9339066e-06,
+          "special": false,
+          "text": "Ass"
+        },
+        {
+          "id": 22137,
+          "logprob": 0.0,
+          "special": false,
+          "text": "istant"
+        },
+        {
+          "id": 29901,
+          "logprob": -2.861023e-06,
+          "special": false,
+          "text": ":"
+        },
+        {
+          "id": 319,
+          "logprob": -0.8828125,
+          "special": false,
+          "text": " A"
+        },
+        {
+          "id": 521,
+          "logprob": -1.3759766,
+          "special": false,
+          "text": " ch"
+        },
+        {
+          "id": 21475,
+          "logprob": -0.00050878525,
+          "special": false,
+          "text": "icken"
+        },
+        {
+          "id": 338,
+          "logprob": -1.1367188,
+          "special": false,
+          "text": " is"
+        }
+      ]
+    },
+    "generated_text": "\nAssistant: A chicken is"
+  },
+  {
+    "details": {
+      "best_of_sequences": null,
+      "finish_reason": "length",
+      "generated_tokens": 10,
+      "prefill": [
+        {
+          "id": 1,
+          "logprob": null,
+          "text": "<s>"
+        },
+        {
+          "id": 4911,
+          "logprob": -5.3476562,
+          "text": "User"
+        },
+        {
+          "id": 29901,
+          "logprob": -0.0075531006,
+          "text": ":"
+        },
+        {
+          "id": 32000,
+          "logprob": -0.7729492,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 32001,
+          "logprob": -9.775162e-05,
+          "text": "<image>"
+        },
+        {
+          "id": 32000,
+          "logprob": -1.1920929e-07,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 1815,
+          "logprob": -4.4296875,
+          "text": "Can"
+        },
+        {
+          "id": 366,
+          "logprob": -0.01424408,
+          "text": "you"
+        },
+        {
+          "id": 2649,
+          "logprob": -4.9335938,
+          "text": "tell"
+        },
+        {
+          "id": 592,
+          "logprob": -0.2993164,
+          "text": "me"
+        },
+        {
+          "id": 263,
+          "logprob": -3.5664062,
+          "text": "a"
+        },
+        {
+          "id": 1407,
+          "logprob": -9.4453125,
+          "text": "very"
+        },
+        {
+          "id": 3273,
+          "logprob": -1.9306641,
+          "text": "short"
+        },
+        {
+          "id": 5828,
+          "logprob": -0.2836914,
+          "text": "story"
+        },
+        {
+          "id": 2729,
+          "logprob": -3.4179688,
+          "text": "based"
+        },
+        {
+          "id": 373,
+          "logprob": -0.00056934357,
+          "text": "on"
+        },
+        {
+          "id": 278,
+          "logprob": -0.13928223,
+          "text": "the"
+        },
+        {
+          "id": 1967,
+          "logprob": -0.05355835,
+          "text": "image"
+        },
+        {
+          "id": 29973,
+          "logprob": -0.15771484,
+          "text": "?"
+        }
+      ],
+      "seed": null,
+      "tokens": [
+        {
+          "id": 32002,
+          "logprob": -0.004333496,
+          "special": true,
+          "text": "<end_of_utterance>"
+        },
+        {
+          "id": 29871,
+          "logprob": -7.43866e-05,
+          "special": false,
+          "text": " "
+        },
+        {
+          "id": 13,
+          "logprob": -2.360344e-05,
+          "special": false,
+          "text": "\n"
+        },
+        {
+          "id": 7900,
+          "logprob": -3.9339066e-06,
+          "special": false,
+          "text": "Ass"
+        },
+        {
+          "id": 22137,
+          "logprob": 0.0,
+          "special": false,
+          "text": "istant"
+        },
+        {
+          "id": 29901,
+          "logprob": -2.7418137e-06,
+          "special": false,
+          "text": ":"
+        },
+        {
+          "id": 319,
+          "logprob": -0.8828125,
+          "special": false,
+          "text": " A"
+        },
+        {
+          "id": 521,
+          "logprob": -1.3759766,
+          "special": false,
+          "text": " ch"
+        },
+        {
+          "id": 21475,
+          "logprob": -0.00050878525,
+          "special": false,
+          "text": "icken"
+        },
+        {
+          "id": 338,
+          "logprob": -1.1367188,
+          "special": false,
+          "text": " is"
+        }
+      ]
+    },
+    "generated_text": "\nAssistant: A chicken is"
+  },
+  {
+    "details": {
+      "best_of_sequences": null,
+      "finish_reason": "length",
+      "generated_tokens": 10,
+      "prefill": [
+        {
+          "id": 1,
+          "logprob": null,
+          "text": "<s>"
+        },
+        {
+          "id": 4911,
+          "logprob": -5.3632812,
+          "text": "User"
+        },
+        {
+          "id": 29901,
+          "logprob": -0.00762558,
+          "text": ":"
+        },
+        {
+          "id": 32000,
+          "logprob": -0.7739258,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 32001,
+          "logprob": -9.775162e-05,
+          "text": "<image>"
+        },
+        {
+          "id": 32000,
+          "logprob": -1.1920929e-07,
+          "text": "<fake_token_around_image>"
+        },
+        {
+          "id": 1815,
+          "logprob": -4.4140625,
+          "text": "Can"
+        },
+        {
+          "id": 366,
+          "logprob": -0.01436615,
+          "text": "you"
+        },
+        {
+          "id": 2649,
+          "logprob": -4.9414062,
+          "text": "tell"
+        },
+        {
+          "id": 592,
+          "logprob": -0.3005371,
+          "text": "me"
+        },
+        {
+          "id": 263,
+          "logprob": -3.5703125,
+          "text": "a"
+        },
+        {
+          "id": 1407,
+          "logprob": -9.4296875,
+          "text": "very"
+        },
+        {
+          "id": 3273,
+          "logprob": -1.9111328,
+          "text": "short"
+        },
+        {
+          "id": 5828,
+          "logprob": -0.28881836,
+          "text": "story"
+        },
+        {
+          "id": 2729,
+          "logprob": -3.4179688,
+          "text": "based"
+        },
+        {
+          "id": 373,
+          "logprob": -0.00056886673,
+          "text": "on"
+        },
+        {
+          "id": 278,
+          "logprob": -0.14123535,
+          "text": "the"
+        },
+        {
+          "id": 1967,
+          "logprob": -0.053985596,
+          "text": "image"
+        },
+        {
+          "id": 29973,
+          "logprob": -0.15771484,
+          "text": "?"
+        }
+      ],
+      "seed": null,
+      "tokens": [
+        {
+          "id": 32002,
+          "logprob": -0.004295349,
+          "special": true,
+          "text": "<end_of_utterance>"
+        },
+        {
+          "id": 29871,
+          "logprob": -7.43866e-05,
+          "special": false,
+          "text": " "
+        },
+        {
+          "id": 13,
+          "logprob": -2.3126602e-05,
+          "special": false,
+          "text": "\n"
+        },
+        {
+          "id": 7900,
+          "logprob": -3.9339066e-06,
+          "special": false,
+          "text": "Ass"
+        },
+        {
+          "id": 22137,
+          "logprob": 0.0,
+          "special": false,
+          "text": "istant"
+        },
+        {
+          "id": 29901,
+          "logprob": -2.6226044e-06,
+          "special": false,
+          "text": ":"
+        },
+        {
+          "id": 319,
+          "logprob": -0.87841797,
+          "special": false,
+          "text": " A"
+        },
+        {
+          "id": 521,
+          "logprob": -1.3837891,
+          "special": false,
+          "text": " ch"
+        },
+        {
+          "id": 21475,
+          "logprob": -0.00051641464,
+          "special": false,
+          "text": "icken"
+        },
+        {
+          "id": 338,
+          "logprob": -1.1435547,
+          "special": false,
+          "text": " is"
+        }
+      ]
+    },
+    "generated_text": "\nAssistant: A chicken is"
+  }
+]
diff --git a/integration-tests/models/test_idefics.py b/integration-tests/models/test_idefics.py
new file mode 100644
index 00000000..5659dd5c
--- /dev/null
+++ b/integration-tests/models/test_idefics.py
@@ -0,0 +1,46 @@
+import pytest
+
+
+@pytest.fixture(scope="module")
+def idefics_handle(launcher):
+    with launcher(
+        "HuggingFaceM4/idefics-9b-instruct", num_shard=2
+    ) as handle:
+        yield handle
+
+
+@pytest.fixture(scope="module")
+async def idefics(idefics_handle):
+    await idefics_handle.health(300)
+    return idefics_handle.client
+
+
+@pytest.mark.asyncio
+async def test_idefics(idefics, response_snapshot):
+    response = await idefics.generate(
+        "User:![](https://temp-5681.s3.us-west-2.amazonaws.com/chicken_on_money.png)Can you tell me a very short story based on the image?",
+        max_new_tokens=10,
+        decoder_input_details=True,
+    )
+
+    assert response.details.generated_tokens == 10
+    assert response == response_snapshot
+
+
+@pytest.mark.asyncio
+async def test_idefics_load(idefics, generate_load, response_snapshot):
+    responses = await generate_load(
+        idefics,
+        "User:![](https://temp-5681.s3.us-west-2.amazonaws.com/chicken_on_money.png)Can you tell me a very short story based on the image?",
+        max_new_tokens=10,
+        n=4,
+    )
+
+    generated_texts = [r.generated_text for r in responses]
+
+    assert len(generated_texts) == 4
+    assert generated_texts, all(
+        [text == generated_texts[0] for text in generated_texts]
+    )
+
+    assert responses == response_snapshot
diff --git a/server/poetry.lock b/server/poetry.lock
index 4a4e4528..aae7b7a5 100644
--- a/server/poetry.lock
+++ b/server/poetry.lock
@@ -152,13 +152,13 @@ frozenlist = ">=1.1.0"
 
 [[package]]
 name = "async-timeout"
-version = "4.0.2"
+version = "4.0.3"
 description = "Timeout context manager for asyncio programs"
 optional = true
-python-versions = ">=3.6"
+python-versions = ">=3.7"
 files = [
-    {file = "async-timeout-4.0.2.tar.gz", hash = "sha256:2163e1640ddb52b7a8c80d0a67a08587e5d245cc9c553a74a847056bc2976b15"},
-    {file = "async_timeout-4.0.2-py3-none-any.whl", hash = "sha256:8ca1e4fcf50d07413d66d1a5e416e42cfdf5851c981d679a09851a6853383b3c"},
+    {file = "async-timeout-4.0.3.tar.gz", hash = "sha256:4640d96be84d82d02ed59ea2b7105a0f7b33abe8703703cd0ab0bf87c427522f"},
+    {file = "async_timeout-4.0.3-py3-none-any.whl", hash = "sha256:7405140ff1230c310e51dc27b3145b9092d659ce68ff733fb0cefe3ee42be028"},
 ]
 
 [[package]]
@@ -310,35 +310,6 @@ files = [
 [package.dependencies]
 colorama = {version = "*", markers = "platform_system == \"Windows\""}
 
-[[package]]
-name = "cmake"
-version = "3.27.0"
-description = "CMake is an open-source, cross-platform family of tools designed to build, test and package software"
-optional = false
-python-versions = "*"
-files = [
-    {file = "cmake-3.27.0-py2.py3-none-macosx_10_10_universal2.macosx_10_10_x86_64.macosx_11_0_arm64.macosx_11_0_universal2.whl", hash = "sha256:9ccab4cd93578d3c2df32e66b44b313b75a7484032645040431dc06a583ca4aa"},
-    {file = "cmake-3.27.0-py2.py3-none-manylinux2010_i686.manylinux_2_12_i686.whl", hash = "sha256:199bfaefb752e82d8067aeee5d6a6e0414fe0d60e9a3fd08e95d537a97e0db16"},
-    {file = "cmake-3.27.0-py2.py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:8745eff805f36762d3e8e904698b853cb4a9da8b4b07d1c12bcd1e1a6c4a1709"},
-    {file = "cmake-3.27.0-py2.py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:58a3f39d3d1bc897f05e531bfa676246a2b25d424c6a47e4b6bbc193fb560db7"},
-    {file = "cmake-3.27.0-py2.py3-none-manylinux2014_i686.manylinux_2_17_i686.whl", hash = "sha256:b470ccd3f86cf19a63f6b221c9cceebcc58e32d3787d0d5f9f43d1d91a095090"},
-    {file = "cmake-3.27.0-py2.py3-none-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:35a8d397ce883e93b5e6561e2803ce9470df52283862264093c1078530f98189"},
-    {file = "cmake-3.27.0-py2.py3-none-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:1f38d87b2c65763a0113f4a6c652e6f4b5adf90b384c1e1d69e4f8a3104a57d6"},
-    {file = "cmake-3.27.0-py2.py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:b9d5811954dcedcaa6c915c4a9bb6d64b55ac189e9cbc74be726307d9d084804"},
-    {file = "cmake-3.27.0-py2.py3-none-musllinux_1_1_aarch64.whl", hash = "sha256:073e4f196d0888216e6794c08cd984ddabc108c0e4e66f48fbd7610d1e6d726d"},
-    {file = "cmake-3.27.0-py2.py3-none-musllinux_1_1_i686.whl", hash = "sha256:e58e48643903e6fad76274337f9a4d3c575b8e21cd05c6214780b2c98bb0c706"},
-    {file = "cmake-3.27.0-py2.py3-none-musllinux_1_1_ppc64le.whl", hash = "sha256:9740ed9f61a3bd8708a41cadd5c057c04f38e5b89bd773e369df2e210a1c55a3"},
-    {file = "cmake-3.27.0-py2.py3-none-musllinux_1_1_s390x.whl", hash = "sha256:1b3189171665f5c8d748ae7fe10a29fff1ebeedeaef57b16f1ea54b1ec0fe514"},
-    {file = "cmake-3.27.0-py2.py3-none-musllinux_1_1_x86_64.whl", hash = "sha256:c4c968c188e7518deb463a14e64f3a19f242c9dcf7f24e1dbcc1419690cd54e0"},
-    {file = "cmake-3.27.0-py2.py3-none-win32.whl", hash = "sha256:5561aca62b65aac844f3931e74cfeb696e4534de145e3307bf942e735736541e"},
-    {file = "cmake-3.27.0-py2.py3-none-win_amd64.whl", hash = "sha256:48be3afe62c9513a49be007896a4058fafec512cb1f269a50126da30aacad97f"},
-    {file = "cmake-3.27.0-py2.py3-none-win_arm64.whl", hash = "sha256:6f46a170b0c9c552d52da4346534570f818195dfc4f1d0c03264e24cc348fc60"},
-    {file = "cmake-3.27.0.tar.gz", hash = "sha256:d03f0a76a2b96805044ad1178b92aeeb5f695caa6776a32522bb5c430a55b4e8"},
-]
-
-[package.extras]
-test = ["coverage (>=4.2)", "flake8 (>=3.0.4)", "path.py (>=11.5.0)", "pytest (>=3.0.3)", "pytest-cov (>=2.4.0)", "pytest-runner (>=2.9)", "pytest-virtualenv (>=1.7.0)", "scikit-build (>=0.10.0)", "setuptools (>=28.0.0)", "virtualenv (>=15.0.3)", "wheel"]
-
 [[package]]
 name = "colorama"
 version = "0.4.6"
@@ -352,13 +323,13 @@ files = [
 
 [[package]]
 name = "datasets"
-version = "2.14.3"
+version = "2.14.4"
 description = "HuggingFace community-driven open-source library of datasets"
 optional = true
 python-versions = ">=3.8.0"
 files = [
-    {file = "datasets-2.14.3-py3-none-any.whl", hash = "sha256:8da5868e55e7c1f0bf3356913a14a9926fe5eca66663691886b95bc9c9b065f0"},
-    {file = "datasets-2.14.3.tar.gz", hash = "sha256:5fb20465a990cf7946f961884ba46fec19cdc95b1cd2cad2a7b87c8d183b02e1"},
+    {file = "datasets-2.14.4-py3-none-any.whl", hash = "sha256:29336bd316a7d827ccd4da2236596279b20ca2ac78f64c04c9483da7cbc2459b"},
+    {file = "datasets-2.14.4.tar.gz", hash = "sha256:ef29c2b5841de488cd343cfc26ab979bff77efa4d2285af51f1ad7db5c46a83b"},
 ]
 
 [package.dependencies]
@@ -436,13 +407,13 @@ files = [
 
 [[package]]
 name = "exceptiongroup"
-version = "1.1.2"
+version = "1.1.3"
 description = "Backport of PEP 654 (exception groups)"
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "exceptiongroup-1.1.2-py3-none-any.whl", hash = "sha256:e346e69d186172ca7cf029c8c1d16235aa0e04035e5750b4b95039e65204328f"},
-    {file = "exceptiongroup-1.1.2.tar.gz", hash = "sha256:12c3e887d6485d16943a309616de20ae5582633e0a2eda17f4e10fd61c1e8af5"},
+    {file = "exceptiongroup-1.1.3-py3-none-any.whl", hash = "sha256:343280667a4585d195ca1cf9cef84a4e178c4b6cf2274caef9859782b567d5e3"},
+    {file = "exceptiongroup-1.1.3.tar.gz", hash = "sha256:097acd85d473d75af5bb98e41b61ff7fe35efe6675e4f9370ec6ec5126d160e9"},
 ]
 
 [package.extras]
@@ -608,148 +579,148 @@ testing = ["protobuf (>=4.21.9)"]
 
 [[package]]
 name = "grpcio"
-version = "1.56.2"
+version = "1.57.0"
 description = "HTTP/2-based RPC framework"
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "grpcio-1.56.2-cp310-cp310-linux_armv7l.whl", hash = "sha256:bf0b9959e673505ee5869950642428046edb91f99942607c2ecf635f8a4b31c9"},
-    {file = "grpcio-1.56.2-cp310-cp310-macosx_12_0_universal2.whl", hash = "sha256:5144feb20fe76e73e60c7d73ec3bf54f320247d1ebe737d10672480371878b48"},
-    {file = "grpcio-1.56.2-cp310-cp310-manylinux_2_17_aarch64.whl", hash = "sha256:a72797549935c9e0b9bc1def1768c8b5a709538fa6ab0678e671aec47ebfd55e"},
-    {file = "grpcio-1.56.2-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c3f3237a57e42f79f1e560726576aedb3a7ef931f4e3accb84ebf6acc485d316"},
-    {file = "grpcio-1.56.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:900bc0096c2ca2d53f2e5cebf98293a7c32f532c4aeb926345e9747452233950"},
-    {file = "grpcio-1.56.2-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:97e0efaebbfd222bcaac2f1735c010c1d3b167112d9d237daebbeedaaccf3d1d"},
-    {file = "grpcio-1.56.2-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:c0c85c5cbe8b30a32fa6d802588d55ffabf720e985abe9590c7c886919d875d4"},
-    {file = "grpcio-1.56.2-cp310-cp310-win32.whl", hash = "sha256:06e84ad9ae7668a109e970c7411e7992751a116494cba7c4fb877656527f9a57"},
-    {file = "grpcio-1.56.2-cp310-cp310-win_amd64.whl", hash = "sha256:10954662f77dc36c9a1fb5cc4a537f746580d6b5734803be1e587252682cda8d"},
-    {file = "grpcio-1.56.2-cp311-cp311-linux_armv7l.whl", hash = "sha256:c435f5ce1705de48e08fcbcfaf8aee660d199c90536e3e06f2016af7d6a938dd"},
-    {file = "grpcio-1.56.2-cp311-cp311-macosx_10_10_universal2.whl", hash = "sha256:6108e5933eb8c22cd3646e72d5b54772c29f57482fd4c41a0640aab99eb5071d"},
-    {file = "grpcio-1.56.2-cp311-cp311-manylinux_2_17_aarch64.whl", hash = "sha256:8391cea5ce72f4a12368afd17799474015d5d3dc00c936a907eb7c7eaaea98a5"},
-    {file = "grpcio-1.56.2-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:750de923b456ca8c0f1354d6befca45d1f3b3a789e76efc16741bd4132752d95"},
-    {file = "grpcio-1.56.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fda2783c12f553cdca11c08e5af6eecbd717280dc8fbe28a110897af1c15a88c"},
-    {file = "grpcio-1.56.2-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:9e04d4e4cfafa7c5264e535b5d28e786f0571bea609c3f0aaab13e891e933e9c"},
-    {file = "grpcio-1.56.2-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:89a49cc5ad08a38b6141af17e00d1dd482dc927c7605bc77af457b5a0fca807c"},
-    {file = "grpcio-1.56.2-cp311-cp311-win32.whl", hash = "sha256:6a007a541dff984264981fbafeb052bfe361db63578948d857907df9488d8774"},
-    {file = "grpcio-1.56.2-cp311-cp311-win_amd64.whl", hash = "sha256:af4063ef2b11b96d949dccbc5a987272f38d55c23c4c01841ea65a517906397f"},
-    {file = "grpcio-1.56.2-cp37-cp37m-linux_armv7l.whl", hash = "sha256:a6ff459dac39541e6a2763a4439c4ca6bc9ecb4acc05a99b79246751f9894756"},
-    {file = "grpcio-1.56.2-cp37-cp37m-macosx_10_10_universal2.whl", hash = "sha256:f20fd21f7538f8107451156dd1fe203300b79a9ddceba1ee0ac8132521a008ed"},
-    {file = "grpcio-1.56.2-cp37-cp37m-manylinux_2_17_aarch64.whl", hash = "sha256:d1fbad1f9077372b6587ec589c1fc120b417b6c8ad72d3e3cc86bbbd0a3cee93"},
-    {file = "grpcio-1.56.2-cp37-cp37m-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:6ee26e9dfb3996aff7c870f09dc7ad44a5f6732b8bdb5a5f9905737ac6fd4ef1"},
-    {file = "grpcio-1.56.2-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a4c60abd950d6de3e4f1ddbc318075654d275c29c846ab6a043d6ed2c52e4c8c"},
-    {file = "grpcio-1.56.2-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:1c31e52a04e62c8577a7bf772b3e7bed4df9c9e0dd90f92b6ffa07c16cab63c9"},
-    {file = "grpcio-1.56.2-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:345356b307cce5d14355e8e055b4ca5f99bc857c33a3dc1ddbc544fca9cd0475"},
-    {file = "grpcio-1.56.2-cp37-cp37m-win_amd64.whl", hash = "sha256:42e63904ee37ae46aa23de50dac8b145b3596f43598fa33fe1098ab2cbda6ff5"},
-    {file = "grpcio-1.56.2-cp38-cp38-linux_armv7l.whl", hash = "sha256:7c5ede2e2558f088c49a1ddda19080e4c23fb5d171de80a726b61b567e3766ed"},
-    {file = "grpcio-1.56.2-cp38-cp38-macosx_10_10_universal2.whl", hash = "sha256:33971197c47965cc1d97d78d842163c283e998223b151bab0499b951fd2c0b12"},
-    {file = "grpcio-1.56.2-cp38-cp38-manylinux_2_17_aarch64.whl", hash = "sha256:d39f5d4af48c138cb146763eda14eb7d8b3ccbbec9fe86fb724cd16e0e914c64"},
-    {file = "grpcio-1.56.2-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:ded637176addc1d3eef35331c39acc598bac550d213f0a1bedabfceaa2244c87"},
-    {file = "grpcio-1.56.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c90da4b124647547a68cf2f197174ada30c7bb9523cb976665dfd26a9963d328"},
-    {file = "grpcio-1.56.2-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:3ccb621749a81dc7755243665a70ce45536ec413ef5818e013fe8dfbf5aa497b"},
-    {file = "grpcio-1.56.2-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:4eb37dd8dd1aa40d601212afa27ca5be255ba792e2e0b24d67b8af5e012cdb7d"},
-    {file = "grpcio-1.56.2-cp38-cp38-win32.whl", hash = "sha256:ddb4a6061933bd9332b74eac0da25f17f32afa7145a33a0f9711ad74f924b1b8"},
-    {file = "grpcio-1.56.2-cp38-cp38-win_amd64.whl", hash = "sha256:8940d6de7068af018dfa9a959a3510e9b7b543f4c405e88463a1cbaa3b2b379a"},
-    {file = "grpcio-1.56.2-cp39-cp39-linux_armv7l.whl", hash = "sha256:51173e8fa6d9a2d85c14426bdee5f5c4a0654fd5fddcc21fe9d09ab0f6eb8b35"},
-    {file = "grpcio-1.56.2-cp39-cp39-macosx_10_10_universal2.whl", hash = "sha256:373b48f210f43327a41e397391715cd11cfce9ded2fe76a5068f9bacf91cc226"},
-    {file = "grpcio-1.56.2-cp39-cp39-manylinux_2_17_aarch64.whl", hash = "sha256:42a3bbb2bc07aef72a7d97e71aabecaf3e4eb616d39e5211e2cfe3689de860ca"},
-    {file = "grpcio-1.56.2-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:5344be476ac37eb9c9ad09c22f4ea193c1316bf074f1daf85bddb1b31fda5116"},
-    {file = "grpcio-1.56.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c3fa3ab0fb200a2c66493828ed06ccd1a94b12eddbfb985e7fd3e5723ff156c6"},
-    {file = "grpcio-1.56.2-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:b975b85d1d5efc36cf8b237c5f3849b64d1ba33d6282f5e991f28751317504a1"},
-    {file = "grpcio-1.56.2-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:cbdf2c498e077282cd427cfd88bdce4668019791deef0be8155385ab2ba7837f"},
-    {file = "grpcio-1.56.2-cp39-cp39-win32.whl", hash = "sha256:139f66656a762572ae718fa0d1f2dce47c05e9fbf7a16acd704c354405b97df9"},
-    {file = "grpcio-1.56.2-cp39-cp39-win_amd64.whl", hash = "sha256:830215173ad45d670140ff99aac3b461f9be9a6b11bee1a17265aaaa746a641a"},
-    {file = "grpcio-1.56.2.tar.gz", hash = "sha256:0ff789ae7d8ddd76d2ac02e7d13bfef6fc4928ac01e1dcaa182be51b6bcc0aaa"},
+    {file = "grpcio-1.57.0-cp310-cp310-linux_armv7l.whl", hash = "sha256:092fa155b945015754bdf988be47793c377b52b88d546e45c6a9f9579ac7f7b6"},
+    {file = "grpcio-1.57.0-cp310-cp310-macosx_12_0_universal2.whl", hash = "sha256:2f7349786da979a94690cc5c2b804cab4e8774a3cf59be40d037c4342c906649"},
+    {file = "grpcio-1.57.0-cp310-cp310-manylinux_2_17_aarch64.whl", hash = "sha256:82640e57fb86ea1d71ea9ab54f7e942502cf98a429a200b2e743d8672171734f"},
+    {file = "grpcio-1.57.0-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:40b72effd4c789de94ce1be2b5f88d7b9b5f7379fe9645f198854112a6567d9a"},
+    {file = "grpcio-1.57.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2f708a6a17868ad8bf586598bee69abded4996b18adf26fd2d91191383b79019"},
+    {file = "grpcio-1.57.0-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:60fe15288a0a65d5c1cb5b4a62b1850d07336e3ba728257a810317be14f0c527"},
+    {file = "grpcio-1.57.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:6907b1cf8bb29b058081d2aad677b15757a44ef2d4d8d9130271d2ad5e33efca"},
+    {file = "grpcio-1.57.0-cp310-cp310-win32.whl", hash = "sha256:57b183e8b252825c4dd29114d6c13559be95387aafc10a7be645462a0fc98bbb"},
+    {file = "grpcio-1.57.0-cp310-cp310-win_amd64.whl", hash = "sha256:7b400807fa749a9eb286e2cd893e501b110b4d356a218426cb9c825a0474ca56"},
+    {file = "grpcio-1.57.0-cp311-cp311-linux_armv7l.whl", hash = "sha256:c6ebecfb7a31385393203eb04ed8b6a08f5002f53df3d59e5e795edb80999652"},
+    {file = "grpcio-1.57.0-cp311-cp311-macosx_10_10_universal2.whl", hash = "sha256:00258cbe3f5188629828363ae8ff78477ce976a6f63fb2bb5e90088396faa82e"},
+    {file = "grpcio-1.57.0-cp311-cp311-manylinux_2_17_aarch64.whl", hash = "sha256:23e7d8849a0e58b806253fd206ac105b328171e01b8f18c7d5922274958cc87e"},
+    {file = "grpcio-1.57.0-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:5371bcd861e679d63b8274f73ac281751d34bd54eccdbfcd6aa00e692a82cd7b"},
+    {file = "grpcio-1.57.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:aed90d93b731929e742967e236f842a4a2174dc5db077c8f9ad2c5996f89f63e"},
+    {file = "grpcio-1.57.0-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:fe752639919aad9ffb0dee0d87f29a6467d1ef764f13c4644d212a9a853a078d"},
+    {file = "grpcio-1.57.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:fada6b07ec4f0befe05218181f4b85176f11d531911b64c715d1875c4736d73a"},
+    {file = "grpcio-1.57.0-cp311-cp311-win32.whl", hash = "sha256:bb396952cfa7ad2f01061fbc7dc1ad91dd9d69243bcb8110cf4e36924785a0fe"},
+    {file = "grpcio-1.57.0-cp311-cp311-win_amd64.whl", hash = "sha256:e503cb45ed12b924b5b988ba9576dc9949b2f5283b8e33b21dcb6be74a7c58d0"},
+    {file = "grpcio-1.57.0-cp37-cp37m-linux_armv7l.whl", hash = "sha256:fd173b4cf02b20f60860dc2ffe30115c18972d7d6d2d69df97ac38dee03be5bf"},
+    {file = "grpcio-1.57.0-cp37-cp37m-macosx_10_10_universal2.whl", hash = "sha256:d7f8df114d6b4cf5a916b98389aeaf1e3132035420a88beea4e3d977e5f267a5"},
+    {file = "grpcio-1.57.0-cp37-cp37m-manylinux_2_17_aarch64.whl", hash = "sha256:76c44efa4ede1f42a9d5b2fed1fe9377e73a109bef8675fb0728eb80b0b8e8f2"},
+    {file = "grpcio-1.57.0-cp37-cp37m-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:4faea2cfdf762a664ab90589b66f416274887641ae17817de510b8178356bf73"},
+    {file = "grpcio-1.57.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c60b83c43faeb6d0a9831f0351d7787a0753f5087cc6fa218d78fdf38e5acef0"},
+    {file = "grpcio-1.57.0-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:b363bbb5253e5f9c23d8a0a034dfdf1b7c9e7f12e602fc788c435171e96daccc"},
+    {file = "grpcio-1.57.0-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:f1fb0fd4a1e9b11ac21c30c169d169ef434c6e9344ee0ab27cfa6f605f6387b2"},
+    {file = "grpcio-1.57.0-cp37-cp37m-win_amd64.whl", hash = "sha256:34950353539e7d93f61c6796a007c705d663f3be41166358e3d88c45760c7d98"},
+    {file = "grpcio-1.57.0-cp38-cp38-linux_armv7l.whl", hash = "sha256:871f9999e0211f9551f368612460442a5436d9444606184652117d6a688c9f51"},
+    {file = "grpcio-1.57.0-cp38-cp38-macosx_10_10_universal2.whl", hash = "sha256:a8a8e560e8dbbdf29288872e91efd22af71e88b0e5736b0daf7773c1fecd99f0"},
+    {file = "grpcio-1.57.0-cp38-cp38-manylinux_2_17_aarch64.whl", hash = "sha256:2313b124e475aa9017a9844bdc5eafb2d5abdda9d456af16fc4535408c7d6da6"},
+    {file = "grpcio-1.57.0-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b4098b6b638d9e0ca839a81656a2fd4bc26c9486ea707e8b1437d6f9d61c3941"},
+    {file = "grpcio-1.57.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:5e5b58e32ae14658085c16986d11e99abd002ddbf51c8daae8a0671fffb3467f"},
+    {file = "grpcio-1.57.0-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:0f80bf37f09e1caba6a8063e56e2b87fa335add314cf2b78ebf7cb45aa7e3d06"},
+    {file = "grpcio-1.57.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:5b7a4ce8f862fe32b2a10b57752cf3169f5fe2915acfe7e6a1e155db3da99e79"},
+    {file = "grpcio-1.57.0-cp38-cp38-win32.whl", hash = "sha256:9338bacf172e942e62e5889b6364e56657fbf8ac68062e8b25c48843e7b202bb"},
+    {file = "grpcio-1.57.0-cp38-cp38-win_amd64.whl", hash = "sha256:e1cb52fa2d67d7f7fab310b600f22ce1ff04d562d46e9e0ac3e3403c2bb4cc16"},
+    {file = "grpcio-1.57.0-cp39-cp39-linux_armv7l.whl", hash = "sha256:fee387d2fab144e8a34e0e9c5ca0f45c9376b99de45628265cfa9886b1dbe62b"},
+    {file = "grpcio-1.57.0-cp39-cp39-macosx_10_10_universal2.whl", hash = "sha256:b53333627283e7241fcc217323f225c37783b5f0472316edcaa4479a213abfa6"},
+    {file = "grpcio-1.57.0-cp39-cp39-manylinux_2_17_aarch64.whl", hash = "sha256:f19ac6ac0a256cf77d3cc926ef0b4e64a9725cc612f97228cd5dc4bd9dbab03b"},
+    {file = "grpcio-1.57.0-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e3fdf04e402f12e1de8074458549337febb3b45f21076cc02ef4ff786aff687e"},
+    {file = "grpcio-1.57.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:5613a2fecc82f95d6c51d15b9a72705553aa0d7c932fad7aed7afb51dc982ee5"},
+    {file = "grpcio-1.57.0-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:b670c2faa92124b7397b42303e4d8eb64a4cd0b7a77e35a9e865a55d61c57ef9"},
+    {file = "grpcio-1.57.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:7a635589201b18510ff988161b7b573f50c6a48fae9cb567657920ca82022b37"},
+    {file = "grpcio-1.57.0-cp39-cp39-win32.whl", hash = "sha256:d78d8b86fcdfa1e4c21f8896614b6cc7ee01a2a758ec0c4382d662f2a62cf766"},
+    {file = "grpcio-1.57.0-cp39-cp39-win_amd64.whl", hash = "sha256:20ec6fc4ad47d1b6e12deec5045ec3cd5402d9a1597f738263e98f490fe07056"},
+    {file = "grpcio-1.57.0.tar.gz", hash = "sha256:4b089f7ad1eb00a104078bab8015b0ed0ebcb3b589e527ab009c53893fd4e613"},
 ]
 
 [package.extras]
-protobuf = ["grpcio-tools (>=1.56.2)"]
+protobuf = ["grpcio-tools (>=1.57.0)"]
 
 [[package]]
 name = "grpcio-reflection"
-version = "1.56.2"
+version = "1.57.0"
 description = "Standard Protobuf Reflection Service for gRPC"
 optional = false
 python-versions = ">=3.6"
 files = [
-    {file = "grpcio-reflection-1.56.2.tar.gz", hash = "sha256:74a81766af639ab8f1b7f59531dc814640f4a1bcf012dc06ce6be205b50a394c"},
-    {file = "grpcio_reflection-1.56.2-py3-none-any.whl", hash = "sha256:004bcc3d4a3dcd89bf83253e4c08ca032fc7ca862f7532ea09ebdf08801fc193"},
+    {file = "grpcio-reflection-1.57.0.tar.gz", hash = "sha256:8f63a18729cba995a172f8325235f5094cb066febec75f9a3b1b2e28328aa166"},
+    {file = "grpcio_reflection-1.57.0-py3-none-any.whl", hash = "sha256:d7deb8587f9d0095fb5d367c2aa5ce1380e3f23b0f8bca6c00bc404c5429cb6a"},
 ]
 
 [package.dependencies]
-grpcio = ">=1.56.2"
+grpcio = ">=1.57.0"
 protobuf = ">=4.21.6"
 
 [[package]]
 name = "grpcio-status"
-version = "1.56.2"
+version = "1.57.0"
 description = "Status proto mapping for gRPC"
 optional = false
 python-versions = ">=3.6"
 files = [
-    {file = "grpcio-status-1.56.2.tar.gz", hash = "sha256:a046b2c0118df4a5687f4585cca9d3c3bae5c498c4dff055dcb43fb06a1180c8"},
-    {file = "grpcio_status-1.56.2-py3-none-any.whl", hash = "sha256:63f3842867735f59f5d70e723abffd2e8501a6bcd915612a1119e52f10614782"},
+    {file = "grpcio-status-1.57.0.tar.gz", hash = "sha256:b098da99df1eebe58337f8f78e50df990273ccacc1226fddeb47c590e3df9e02"},
+    {file = "grpcio_status-1.57.0-py3-none-any.whl", hash = "sha256:15d6af055914ebbc4ed17e55ebfb8e6bb17a45a57fea32e6af19978fb7844690"},
 ]
 
 [package.dependencies]
 googleapis-common-protos = ">=1.5.5"
-grpcio = ">=1.56.2"
+grpcio = ">=1.57.0"
 protobuf = ">=4.21.6"
 
 [[package]]
 name = "grpcio-tools"
-version = "1.56.2"
+version = "1.57.0"
 description = "Protobuf code generator for gRPC"
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "grpcio-tools-1.56.2.tar.gz", hash = "sha256:82af2f4040084141a732f0ef1ecf3f14fdf629923d74d850415e4d09a077e77a"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-linux_armv7l.whl", hash = "sha256:42272376e9a5a1c631863cda056c143c98d21e5b670db5c8c5b7ed0ba3a1a6eb"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-macosx_12_0_universal2.whl", hash = "sha256:a8735d7aa34be99dddfbd476eff6005e684bb2c893c0f62a5811528b84c5b371"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-manylinux_2_17_aarch64.whl", hash = "sha256:68ef3aa7509e5e7a6e7c0ecc183e28118e73da4bef0fc77079648601ce35e58f"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:380985b8d95ea2469e103945bd83a815d1213e370f580631fdd5a3dbaa17e446"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6bfb375eb4f1946d68b8bc7b963c756defa31aa573a35c152a7233d06c0ad6ad"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:13388a22fcba9a1a87f217130a1a01365716af74bd5d0a8a54fc383b8e048ef2"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:7a26160bc0ea5b464715789d4d2a66f01816271677673d65da39bac65b9ea838"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-win32.whl", hash = "sha256:ff16dd0b086e75f574dbc122e018a44dbd1c6dae3f3621ea99e8e5a6b2706e12"},
-    {file = "grpcio_tools-1.56.2-cp310-cp310-win_amd64.whl", hash = "sha256:2037109c1ce253a8e013c9e3ad3722e887d28a1807acdeb1a51b295c8200137b"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-linux_armv7l.whl", hash = "sha256:6dc43300189a69807857c52a3d782e9d3bbfb1cb72dcb27b4043c25161919601"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-macosx_10_10_universal2.whl", hash = "sha256:e7009623635ebcd3dd7fe974883fc2d9a3ff0fcef419bfc0a2da8071b372d9f5"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-manylinux_2_17_aarch64.whl", hash = "sha256:c7ca2272022f90b73efe900244aaebe9dd7cf3b379e99e08a88984e2fdd229c2"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:493775d17ea09cea6047ba81e4d3f0eb82e34d2fbd3b96e43f72b44ce74726ee"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:41af279cf5359b123138236c0980440f4cb4d3d18f03b5c1c314cc1512048351"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:857d72e991d449ec4d2f8337e5e24ddf77b4539965f5cabc84d4b63585832982"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:c0640728d63c9fa56e9a1679943ae4e33ad43a10802dd7a93255870731f44d07"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-win32.whl", hash = "sha256:355204d1b33c7a19e7d69afda411e6595d39ba1e9cbf561770ac1d5403296554"},
-    {file = "grpcio_tools-1.56.2-cp311-cp311-win_amd64.whl", hash = "sha256:ea5d108d28b4cd2e28539241c6aee96bda83086d8888c36785d9f84ea690d896"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-linux_armv7l.whl", hash = "sha256:28444615b7a76b3d9267f81d1487fcad21a581d00564164d9e25ccc28635a811"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-macosx_10_10_universal2.whl", hash = "sha256:45d8b5ad6716848d5b68d9cee29a1a9c5c4baa1824ec5b92d9e35acedddba076"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-manylinux_2_17_aarch64.whl", hash = "sha256:31d1183d28ffc8da242333cb9f683f5093941da80dd5281db0fa93077aecb518"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0059dfc9bea8f7bca69c15ca62c88904c4f907fde1137e0743b5eee054661873"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:24fc857252181c9950ed2d8cee3df5bd0f42861c4ad0db2a57400186827f96e5"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:3a74a5e4fc8121a51401665f96f9a70aee50a2f1221e4a199e67b3b8f55881e8"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:bec47db5d8b5c3b2a44afdbc3f3bf306e34279289a206d20222824381ca3cb13"},
-    {file = "grpcio_tools-1.56.2-cp37-cp37m-win_amd64.whl", hash = "sha256:c0dbaac63a25c088f864295f394230eeb7be48dac2264433fda2603f86c36b25"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-linux_armv7l.whl", hash = "sha256:0a4f9cce5a16613b6d3123c89f9d50e0d13b466799af17bc723dc7d2901a54e4"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-macosx_10_10_universal2.whl", hash = "sha256:ea5fc1b49514b44a3e5a45156c025002f172ade4c509e58c51967865c7c6fa45"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-manylinux_2_17_aarch64.whl", hash = "sha256:54da410124547bacb97a54546c1a95f1af0125e48edc8b5679412ef8b2844f81"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:5223668649172d879ee780253b8e4a79144c56a3cc1bb021847f583508c2b0be"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:483256d5f5be6a77b24d8a5f06ca152d1571c62bf5c738834da61107c7563afe"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:1f334718eb796799bfadbac5567456fb745cee8c7b438c93b74d1ce676c6ad07"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:26751f69cbbc8ea19cf0657b7d109a6db7df81f80caf16380ebcd20eea27652c"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-win32.whl", hash = "sha256:4056ff13e30813d42a30ce1cdfeaeb6bbee915515c161c1df896dac3143ae643"},
-    {file = "grpcio_tools-1.56.2-cp38-cp38-win_amd64.whl", hash = "sha256:878b9269ceb0dd934b61697a9dd9a5c3e9552521e8f46ab32cf4d72a223f7b6c"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-linux_armv7l.whl", hash = "sha256:216e86d3a6ccc31b27fa4c12491981a0a39d4787d2358b6df05baffa40084494"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-macosx_10_10_universal2.whl", hash = "sha256:778224fcbc1cc7eaf222ce94676afbac8d72b4f84cf4239e30b01d2450a46126"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-manylinux_2_17_aarch64.whl", hash = "sha256:14120fb2c6f7894fac5b689934368c692ec50f50a320e8073277ab7778fd612f"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:014da3ed176beb2b1c8430ccc34c8fe962cdd5480e56fb4ab9de60e60c315f3f"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8febb4f90b8fab3179f5bdaa159f1d2a20523ea17ec0d66bdec7732f9532de91"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:ffae7df3318266614f7aa440acb2098c064b6b5ae061fc22125092386349e526"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:7d86e24eb6e3973c55e9c74412ff755d1b9d15518c4eaf95676acff49d0162a2"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-win32.whl", hash = "sha256:506d00a86950adf4017395551a4547c0b7fcefa90e4c220135fc3e34e31be81b"},
-    {file = "grpcio_tools-1.56.2-cp39-cp39-win_amd64.whl", hash = "sha256:8da04f033b8f4c597e8fc990e2f626bad2b269227bdd554592ea618f624f1aa9"},
+    {file = "grpcio-tools-1.57.0.tar.gz", hash = "sha256:2f16130d869ce27ecd623194547b649dd657333ec7e8644cc571c645781a9b85"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-linux_armv7l.whl", hash = "sha256:4fb8a8468031f858381a576078924af364a08833d8f8f3237018252c4573a802"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-macosx_12_0_universal2.whl", hash = "sha256:35bf0dad8a3562043345236c26d0053a856fb06c04d7da652f2ded914e508ae7"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-manylinux_2_17_aarch64.whl", hash = "sha256:ec9aab2fb6783c7fc54bc28f58eb75f1ca77594e6b0fd5e5e7a8114a95169fe0"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0cf5fc0a1c23f8ea34b408b72fb0e90eec0f404ad4dba98e8f6da3c9ce34e2ed"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:26e69d08a515554e0cfe1ec4d31568836f4b17f0ff82294f957f629388629eb9"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:c39a3656576b6fdaaf28abe0467f7a7231df4230c1bee132322dbc3209419e7f"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:f64f8ab22d27d4a5693310748d35a696061c3b5c7b8c4fb4ab3b4bc1068b6b56"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-win32.whl", hash = "sha256:d2a134756f4db34759a5cc7f7e43f7eb87540b68d1cca62925593c6fb93924f7"},
+    {file = "grpcio_tools-1.57.0-cp310-cp310-win_amd64.whl", hash = "sha256:9a3d60fb8d46ede26c1907c146561b3a9caa20a7aff961bc661ef8226f85a2e9"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-linux_armv7l.whl", hash = "sha256:aac98ecad8f7bd4301855669d42a5d97ef7bb34bec2b1e74c7a0641d47e313cf"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-macosx_10_10_universal2.whl", hash = "sha256:cdd020cb68b51462983b7c2dfbc3eb6ede032b8bf438d4554df0c3f08ce35c76"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-manylinux_2_17_aarch64.whl", hash = "sha256:f54081b08419a39221cd646363b5708857c696b3ad4784f1dcf310891e33a5f7"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:ed85a0291fff45b67f2557fe7f117d3bc7af8b54b8619d27bf374b5c8b7e3ca2"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e868cd6feb3ef07d4b35be104fe1fd0657db05259ff8f8ec5e08f4f89ca1191d"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:dfb6f6120587b8e228a3cae5ee4985b5bdc18501bad05c49df61965dfc9d70a9"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:4a7ad7f328e28fc97c356d0f10fb10d8b5151bb65aa7cf14bf8084513f0b7306"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-win32.whl", hash = "sha256:9867f2817b1a0c93c523f89ac6c9d8625548af4620a7ce438bf5a76e23327284"},
+    {file = "grpcio_tools-1.57.0-cp311-cp311-win_amd64.whl", hash = "sha256:1f9e917a9f18087f6c14b4d4508fb94fca5c2f96852363a89232fb9b2124ac1f"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-linux_armv7l.whl", hash = "sha256:9f2aefa8a37bd2c4db1a3f1aca11377e2766214520fb70e67071f4ff8d8b0fa5"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-macosx_10_10_universal2.whl", hash = "sha256:850cbda0ec5d24c39e7215ede410276040692ca45d105fbbeada407fa03f0ac0"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-manylinux_2_17_aarch64.whl", hash = "sha256:6fa52972c9647876ea35f6dc2b51002a74ed900ec7894586cbb2fe76f64f99de"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:76c0eea89d7542719594e50e2283f51a072978b953e8b3e9fd7c59a2c762d4c1"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f3da5240211252fc70a6451fe00c143e2ab2f7bfc2445695ad2ed056b8e48d96"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:a0256f8786ac9e4db618a1aa492bb3472569a0946fd3ee862ffe23196323da55"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:c026bdf5c1366ce88b7bbe2d8207374d675afd3fd911f60752103de3da4a41d2"},
+    {file = "grpcio_tools-1.57.0-cp37-cp37m-win_amd64.whl", hash = "sha256:9053c2f655589545be08b9d6a673e92970173a4bf11a4b9f18cd6e9af626b587"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-linux_armv7l.whl", hash = "sha256:81ec4dbb696e095057b2528d11a8da04be6bbe2b967fa07d4ea9ba6354338cbf"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-macosx_10_10_universal2.whl", hash = "sha256:495e2946406963e0b9f063f76d5af0f2a19517dac2b367b5b044432ac9194296"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-manylinux_2_17_aarch64.whl", hash = "sha256:7b46fc6aa8eb7edd18cafcd21fd98703cb6c09e46b507de335fca7f0161dfccb"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:fb81ff861692111fa81bd85f64584e624cb4013bd66fbce8a209b8893f5ce398"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8a42dc220eb5305f470855c9284f4c8e85ae59d6d742cd07946b0cbe5e9ca186"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:90d10d9038ba46a595a223a34f136c9230e3d6d7abc2433dbf0e1c31939d3a8b"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:5bc3e6d338aefb052e19cedabe00452be46d0c10a4ed29ee77abb00402e438fe"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-win32.whl", hash = "sha256:34b36217b17b5bea674a414229913e1fd80ede328be51e1b531fcc62abd393b0"},
+    {file = "grpcio_tools-1.57.0-cp38-cp38-win_amd64.whl", hash = "sha256:dbde4004a0688400036342ff73e3706e8940483e2871547b1354d59e93a38277"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-linux_armv7l.whl", hash = "sha256:784574709b9690dc28696617ea69352e2132352fdfc9bc89afa8e39f99ae538e"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-macosx_10_10_universal2.whl", hash = "sha256:85ac4e62eb44428cde025fd9ab7554002315fc7880f791c553fc5a0015cc9931"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-manylinux_2_17_aarch64.whl", hash = "sha256:dc771d4db5701f280957bbcee91745e0686d00ed1c6aa7e05ba30a58b02d70a1"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:f3ac06703c412f8167a9062eaf6099409967e33bf98fa5b02be4b4689b6bdf39"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:02d78c034109f46032c7217260066d49d41e6bcaf588fa28fa40fe2f83445347"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:2db25f15ed44327f2e02d0c4fe741ac966f9500e407047d8a7c7fccf2df65616"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:2b417c97936d94874a3ce7ed8deab910f2233e3612134507cfee4af8735c38a6"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-win32.whl", hash = "sha256:f717cce5093e6b6049d9ea6d12fdf3658efdb1a80772f7737db1f8510b876df6"},
+    {file = "grpcio_tools-1.57.0-cp39-cp39-win_amd64.whl", hash = "sha256:1c0e8a1a32973a5d59fbcc19232f925e5c48116e9411f788033a31c5ca5130b4"},
 ]
 
 [package.dependencies]
-grpcio = ">=1.56.2"
+grpcio = ">=1.57.0"
 protobuf = ">=4.21.6,<5.0dev"
 setuptools = "*"
 
@@ -851,16 +822,6 @@ MarkupSafe = ">=2.0"
 [package.extras]
 i18n = ["Babel (>=2.7)"]
 
-[[package]]
-name = "lit"
-version = "16.0.6"
-description = "A Software Testing Tool"
-optional = false
-python-versions = "*"
-files = [
-    {file = "lit-16.0.6.tar.gz", hash = "sha256:84623c9c23b6b14763d637f4e63e6b721b3446ada40bf7001d8fee70b8e77a9a"},
-]
-
 [[package]]
 name = "loguru"
 version = "0.6.0"
@@ -1377,6 +1338,75 @@ docs-specific = ["hf-doc-builder"]
 quality = ["black (>=22.0,<23.0)", "ruff (>=0.0.241)", "urllib3 (<=2.0.0)"]
 test = ["black (>=22.0,<23.0)", "datasets", "diffusers", "hf-doc-builder", "parameterized", "pytest", "pytest-cov", "pytest-xdist", "ruff (>=0.0.241)", "urllib3 (<=2.0.0)"]
 
+[[package]]
+name = "pillow"
+version = "10.0.0"
+description = "Python Imaging Library (Fork)"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "Pillow-10.0.0-cp310-cp310-macosx_10_10_x86_64.whl", hash = "sha256:1f62406a884ae75fb2f818694469519fb685cc7eaff05d3451a9ebe55c646891"},
+    {file = "Pillow-10.0.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:d5db32e2a6ccbb3d34d87c87b432959e0db29755727afb37290e10f6e8e62614"},
+    {file = "Pillow-10.0.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:edf4392b77bdc81f36e92d3a07a5cd072f90253197f4a52a55a8cec48a12483b"},
+    {file = "Pillow-10.0.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:520f2a520dc040512699f20fa1c363eed506e94248d71f85412b625026f6142c"},
+    {file = "Pillow-10.0.0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:8c11160913e3dd06c8ffdb5f233a4f254cb449f4dfc0f8f4549eda9e542c93d1"},
+    {file = "Pillow-10.0.0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:a74ba0c356aaa3bb8e3eb79606a87669e7ec6444be352870623025d75a14a2bf"},
+    {file = "Pillow-10.0.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:d5d0dae4cfd56969d23d94dc8e89fb6a217be461c69090768227beb8ed28c0a3"},
+    {file = "Pillow-10.0.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:22c10cc517668d44b211717fd9775799ccec4124b9a7f7b3635fc5386e584992"},
+    {file = "Pillow-10.0.0-cp310-cp310-win_amd64.whl", hash = "sha256:dffe31a7f47b603318c609f378ebcd57f1554a3a6a8effbc59c3c69f804296de"},
+    {file = "Pillow-10.0.0-cp311-cp311-macosx_10_10_x86_64.whl", hash = "sha256:9fb218c8a12e51d7ead2a7c9e101a04982237d4855716af2e9499306728fb485"},
+    {file = "Pillow-10.0.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:d35e3c8d9b1268cbf5d3670285feb3528f6680420eafe35cccc686b73c1e330f"},
+    {file = "Pillow-10.0.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3ed64f9ca2f0a95411e88a4efbd7a29e5ce2cea36072c53dd9d26d9c76f753b3"},
+    {file = "Pillow-10.0.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0b6eb5502f45a60a3f411c63187db83a3d3107887ad0d036c13ce836f8a36f1d"},
+    {file = "Pillow-10.0.0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:c1fbe7621c167ecaa38ad29643d77a9ce7311583761abf7836e1510c580bf3dd"},
+    {file = "Pillow-10.0.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:cd25d2a9d2b36fcb318882481367956d2cf91329f6892fe5d385c346c0649629"},
+    {file = "Pillow-10.0.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:3b08d4cc24f471b2c8ca24ec060abf4bebc6b144cb89cba638c720546b1cf538"},
+    {file = "Pillow-10.0.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:d737a602fbd82afd892ca746392401b634e278cb65d55c4b7a8f48e9ef8d008d"},
+    {file = "Pillow-10.0.0-cp311-cp311-win_amd64.whl", hash = "sha256:3a82c40d706d9aa9734289740ce26460a11aeec2d9c79b7af87bb35f0073c12f"},
+    {file = "Pillow-10.0.0-cp311-cp311-win_arm64.whl", hash = "sha256:bc2ec7c7b5d66b8ec9ce9f720dbb5fa4bace0f545acd34870eff4a369b44bf37"},
+    {file = "Pillow-10.0.0-cp312-cp312-macosx_10_10_x86_64.whl", hash = "sha256:d80cf684b541685fccdd84c485b31ce73fc5c9b5d7523bf1394ce134a60c6883"},
+    {file = "Pillow-10.0.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:76de421f9c326da8f43d690110f0e79fe3ad1e54be811545d7d91898b4c8493e"},
+    {file = "Pillow-10.0.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:81ff539a12457809666fef6624684c008e00ff6bf455b4b89fd00a140eecd640"},
+    {file = "Pillow-10.0.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ce543ed15570eedbb85df19b0a1a7314a9c8141a36ce089c0a894adbfccb4568"},
+    {file = "Pillow-10.0.0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:685ac03cc4ed5ebc15ad5c23bc555d68a87777586d970c2c3e216619a5476223"},
+    {file = "Pillow-10.0.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:d72e2ecc68a942e8cf9739619b7f408cc7b272b279b56b2c83c6123fcfa5cdff"},
+    {file = "Pillow-10.0.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:d50b6aec14bc737742ca96e85d6d0a5f9bfbded018264b3b70ff9d8c33485551"},
+    {file = "Pillow-10.0.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:00e65f5e822decd501e374b0650146063fbb30a7264b4d2744bdd7b913e0cab5"},
+    {file = "Pillow-10.0.0-cp312-cp312-win_amd64.whl", hash = "sha256:f31f9fdbfecb042d046f9d91270a0ba28368a723302786c0009ee9b9f1f60199"},
+    {file = "Pillow-10.0.0-cp312-cp312-win_arm64.whl", hash = "sha256:1ce91b6ec08d866b14413d3f0bbdea7e24dfdc8e59f562bb77bc3fe60b6144ca"},
+    {file = "Pillow-10.0.0-cp38-cp38-macosx_10_10_x86_64.whl", hash = "sha256:349930d6e9c685c089284b013478d6f76e3a534e36ddfa912cde493f235372f3"},
+    {file = "Pillow-10.0.0-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:3a684105f7c32488f7153905a4e3015a3b6c7182e106fe3c37fbb5ef3e6994c3"},
+    {file = "Pillow-10.0.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b4f69b3700201b80bb82c3a97d5e9254084f6dd5fb5b16fc1a7b974260f89f43"},
+    {file = "Pillow-10.0.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3f07ea8d2f827d7d2a49ecf1639ec02d75ffd1b88dcc5b3a61bbb37a8759ad8d"},
+    {file = "Pillow-10.0.0-cp38-cp38-manylinux_2_28_aarch64.whl", hash = "sha256:040586f7d37b34547153fa383f7f9aed68b738992380ac911447bb78f2abe530"},
+    {file = "Pillow-10.0.0-cp38-cp38-manylinux_2_28_x86_64.whl", hash = "sha256:f88a0b92277de8e3ca715a0d79d68dc82807457dae3ab8699c758f07c20b3c51"},
+    {file = "Pillow-10.0.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:c7cf14a27b0d6adfaebb3ae4153f1e516df54e47e42dcc073d7b3d76111a8d86"},
+    {file = "Pillow-10.0.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:3400aae60685b06bb96f99a21e1ada7bc7a413d5f49bce739828ecd9391bb8f7"},
+    {file = "Pillow-10.0.0-cp38-cp38-win_amd64.whl", hash = "sha256:dbc02381779d412145331789b40cc7b11fdf449e5d94f6bc0b080db0a56ea3f0"},
+    {file = "Pillow-10.0.0-cp39-cp39-macosx_10_10_x86_64.whl", hash = "sha256:9211e7ad69d7c9401cfc0e23d49b69ca65ddd898976d660a2fa5904e3d7a9baa"},
+    {file = "Pillow-10.0.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:faaf07ea35355b01a35cb442dd950d8f1bb5b040a7787791a535de13db15ed90"},
+    {file = "Pillow-10.0.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c9f72a021fbb792ce98306ffb0c348b3c9cb967dce0f12a49aa4c3d3fdefa967"},
+    {file = "Pillow-10.0.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9f7c16705f44e0504a3a2a14197c1f0b32a95731d251777dcb060aa83022cb2d"},
+    {file = "Pillow-10.0.0-cp39-cp39-manylinux_2_28_aarch64.whl", hash = "sha256:76edb0a1fa2b4745fb0c99fb9fb98f8b180a1bbceb8be49b087e0b21867e77d3"},
+    {file = "Pillow-10.0.0-cp39-cp39-manylinux_2_28_x86_64.whl", hash = "sha256:368ab3dfb5f49e312231b6f27b8820c823652b7cd29cfbd34090565a015e99ba"},
+    {file = "Pillow-10.0.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:608bfdee0d57cf297d32bcbb3c728dc1da0907519d1784962c5f0c68bb93e5a3"},
+    {file = "Pillow-10.0.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:5c6e3df6bdd396749bafd45314871b3d0af81ff935b2d188385e970052091017"},
+    {file = "Pillow-10.0.0-cp39-cp39-win_amd64.whl", hash = "sha256:7be600823e4c8631b74e4a0d38384c73f680e6105a7d3c6824fcf226c178c7e6"},
+    {file = "Pillow-10.0.0-pp310-pypy310_pp73-macosx_10_10_x86_64.whl", hash = "sha256:92be919bbc9f7d09f7ae343c38f5bb21c973d2576c1d45600fce4b74bafa7ac0"},
+    {file = "Pillow-10.0.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8f8182b523b2289f7c415f589118228d30ac8c355baa2f3194ced084dac2dbba"},
+    {file = "Pillow-10.0.0-pp310-pypy310_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:38250a349b6b390ee6047a62c086d3817ac69022c127f8a5dc058c31ccef17f3"},
+    {file = "Pillow-10.0.0-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:88af2003543cc40c80f6fca01411892ec52b11021b3dc22ec3bc9d5afd1c5334"},
+    {file = "Pillow-10.0.0-pp39-pypy39_pp73-macosx_10_10_x86_64.whl", hash = "sha256:c189af0545965fa8d3b9613cfdb0cd37f9d71349e0f7750e1fd704648d475ed2"},
+    {file = "Pillow-10.0.0-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ce7b031a6fc11365970e6a5686d7ba8c63e4c1cf1ea143811acbb524295eabed"},
+    {file = "Pillow-10.0.0-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:db24668940f82321e746773a4bc617bfac06ec831e5c88b643f91f122a785684"},
+    {file = "Pillow-10.0.0-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:efe8c0681042536e0d06c11f48cebe759707c9e9abf880ee213541c5b46c5bf3"},
+    {file = "Pillow-10.0.0.tar.gz", hash = "sha256:9c82b5b3e043c7af0d95792d0d20ccf68f61a1fec6b3530e718b688422727396"},
+]
+
+[package.extras]
+docs = ["furo", "olefile", "sphinx (>=2.4)", "sphinx-copybutton", "sphinx-inline-tabs", "sphinx-removed-in", "sphinxext-opengraph"]
+tests = ["check-manifest", "coverage", "defusedxml", "markdown2", "olefile", "packaging", "pyroma", "pytest", "pytest-cov", "pytest-timeout"]
+
 [[package]]
 name = "pluggy"
 version = "1.2.0"
@@ -1394,24 +1424,24 @@ testing = ["pytest", "pytest-benchmark"]
 
 [[package]]
 name = "protobuf"
-version = "4.23.4"
+version = "4.24.0"
 description = ""
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "protobuf-4.23.4-cp310-abi3-win32.whl", hash = "sha256:5fea3c64d41ea5ecf5697b83e41d09b9589e6f20b677ab3c48e5f242d9b7897b"},
-    {file = "protobuf-4.23.4-cp310-abi3-win_amd64.whl", hash = "sha256:7b19b6266d92ca6a2a87effa88ecc4af73ebc5cfde194dc737cf8ef23a9a3b12"},
-    {file = "protobuf-4.23.4-cp37-abi3-macosx_10_9_universal2.whl", hash = "sha256:8547bf44fe8cec3c69e3042f5c4fb3e36eb2a7a013bb0a44c018fc1e427aafbd"},
-    {file = "protobuf-4.23.4-cp37-abi3-manylinux2014_aarch64.whl", hash = "sha256:fee88269a090ada09ca63551bf2f573eb2424035bcf2cb1b121895b01a46594a"},
-    {file = "protobuf-4.23.4-cp37-abi3-manylinux2014_x86_64.whl", hash = "sha256:effeac51ab79332d44fba74660d40ae79985901ac21bca408f8dc335a81aa597"},
-    {file = "protobuf-4.23.4-cp37-cp37m-win32.whl", hash = "sha256:c3e0939433c40796ca4cfc0fac08af50b00eb66a40bbbc5dee711998fb0bbc1e"},
-    {file = "protobuf-4.23.4-cp37-cp37m-win_amd64.whl", hash = "sha256:9053df6df8e5a76c84339ee4a9f5a2661ceee4a0dab019e8663c50ba324208b0"},
-    {file = "protobuf-4.23.4-cp38-cp38-win32.whl", hash = "sha256:e1c915778d8ced71e26fcf43c0866d7499891bca14c4368448a82edc61fdbc70"},
-    {file = "protobuf-4.23.4-cp38-cp38-win_amd64.whl", hash = "sha256:351cc90f7d10839c480aeb9b870a211e322bf05f6ab3f55fcb2f51331f80a7d2"},
-    {file = "protobuf-4.23.4-cp39-cp39-win32.whl", hash = "sha256:6dd9b9940e3f17077e820b75851126615ee38643c2c5332aa7a359988820c720"},
-    {file = "protobuf-4.23.4-cp39-cp39-win_amd64.whl", hash = "sha256:0a5759f5696895de8cc913f084e27fd4125e8fb0914bb729a17816a33819f474"},
-    {file = "protobuf-4.23.4-py3-none-any.whl", hash = "sha256:e9d0be5bf34b275b9f87ba7407796556abeeba635455d036c7351f7c183ef8ff"},
-    {file = "protobuf-4.23.4.tar.gz", hash = "sha256:ccd9430c0719dce806b93f89c91de7977304729e55377f872a92465d548329a9"},
+    {file = "protobuf-4.24.0-cp310-abi3-win32.whl", hash = "sha256:81cb9c4621d2abfe181154354f63af1c41b00a4882fb230b4425cbaed65e8f52"},
+    {file = "protobuf-4.24.0-cp310-abi3-win_amd64.whl", hash = "sha256:6c817cf4a26334625a1904b38523d1b343ff8b637d75d2c8790189a4064e51c3"},
+    {file = "protobuf-4.24.0-cp37-abi3-macosx_10_9_universal2.whl", hash = "sha256:ae97b5de10f25b7a443b40427033e545a32b0e9dda17bcd8330d70033379b3e5"},
+    {file = "protobuf-4.24.0-cp37-abi3-manylinux2014_aarch64.whl", hash = "sha256:567fe6b0647494845d0849e3d5b260bfdd75692bf452cdc9cb660d12457c055d"},
+    {file = "protobuf-4.24.0-cp37-abi3-manylinux2014_x86_64.whl", hash = "sha256:a6b1ca92ccabfd9903c0c7dde8876221dc7d8d87ad5c42e095cc11b15d3569c7"},
+    {file = "protobuf-4.24.0-cp37-cp37m-win32.whl", hash = "sha256:a38400a692fd0c6944c3c58837d112f135eb1ed6cdad5ca6c5763336e74f1a04"},
+    {file = "protobuf-4.24.0-cp37-cp37m-win_amd64.whl", hash = "sha256:5ab19ee50037d4b663c02218a811a5e1e7bb30940c79aac385b96e7a4f9daa61"},
+    {file = "protobuf-4.24.0-cp38-cp38-win32.whl", hash = "sha256:e8834ef0b4c88666ebb7c7ec18045aa0f4325481d724daa624a4cf9f28134653"},
+    {file = "protobuf-4.24.0-cp38-cp38-win_amd64.whl", hash = "sha256:8bb52a2be32db82ddc623aefcedfe1e0eb51da60e18fcc908fb8885c81d72109"},
+    {file = "protobuf-4.24.0-cp39-cp39-win32.whl", hash = "sha256:ae7a1835721086013de193311df858bc12cd247abe4ef9710b715d930b95b33e"},
+    {file = "protobuf-4.24.0-cp39-cp39-win_amd64.whl", hash = "sha256:44825e963008f8ea0d26c51911c30d3e82e122997c3c4568fd0385dd7bacaedf"},
+    {file = "protobuf-4.24.0-py3-none-any.whl", hash = "sha256:82e6e9ebdd15b8200e8423676eab38b774624d6a1ad696a60d86a2ac93f18201"},
+    {file = "protobuf-4.24.0.tar.gz", hash = "sha256:5d0ceb9de6e08311832169e601d1fc71bd8e8c779f3ee38a97a78554945ecb85"},
 ]
 
 [[package]]
@@ -1575,99 +1605,99 @@ files = [
 
 [[package]]
 name = "regex"
-version = "2023.6.3"
+version = "2023.8.8"
 description = "Alternative regular expression module, to replace re."
 optional = false
 python-versions = ">=3.6"
 files = [
-    {file = "regex-2023.6.3-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:824bf3ac11001849aec3fa1d69abcb67aac3e150a933963fb12bda5151fe1bfd"},
-    {file = "regex-2023.6.3-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:05ed27acdf4465c95826962528f9e8d41dbf9b1aa8531a387dee6ed215a3e9ef"},
-    {file = "regex-2023.6.3-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:0b49c764f88a79160fa64f9a7b425620e87c9f46095ef9c9920542ab2495c8bc"},
-    {file = "regex-2023.6.3-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:8e3f1316c2293e5469f8f09dc2d76efb6c3982d3da91ba95061a7e69489a14ef"},
-    {file = "regex-2023.6.3-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:43e1dd9d12df9004246bacb79a0e5886b3b6071b32e41f83b0acbf293f820ee8"},
-    {file = "regex-2023.6.3-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:4959e8bcbfda5146477d21c3a8ad81b185cd252f3d0d6e4724a5ef11c012fb06"},
-    {file = "regex-2023.6.3-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:af4dd387354dc83a3bff67127a124c21116feb0d2ef536805c454721c5d7993d"},
-    {file = "regex-2023.6.3-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:2239d95d8e243658b8dbb36b12bd10c33ad6e6933a54d36ff053713f129aa536"},
-    {file = "regex-2023.6.3-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:890e5a11c97cf0d0c550eb661b937a1e45431ffa79803b942a057c4fb12a2da2"},
-    {file = "regex-2023.6.3-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:a8105e9af3b029f243ab11ad47c19b566482c150c754e4c717900a798806b222"},
-    {file = "regex-2023.6.3-cp310-cp310-musllinux_1_1_ppc64le.whl", hash = "sha256:25be746a8ec7bc7b082783216de8e9473803706723b3f6bef34b3d0ed03d57e2"},
-    {file = "regex-2023.6.3-cp310-cp310-musllinux_1_1_s390x.whl", hash = "sha256:3676f1dd082be28b1266c93f618ee07741b704ab7b68501a173ce7d8d0d0ca18"},
-    {file = "regex-2023.6.3-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:10cb847aeb1728412c666ab2e2000ba6f174f25b2bdc7292e7dd71b16db07568"},
-    {file = "regex-2023.6.3-cp310-cp310-win32.whl", hash = "sha256:dbbbfce33cd98f97f6bffb17801b0576e653f4fdb1d399b2ea89638bc8d08ae1"},
-    {file = "regex-2023.6.3-cp310-cp310-win_amd64.whl", hash = "sha256:c5f8037000eb21e4823aa485149f2299eb589f8d1fe4b448036d230c3f4e68e0"},
-    {file = "regex-2023.6.3-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:c123f662be8ec5ab4ea72ea300359023a5d1df095b7ead76fedcd8babbedf969"},
-    {file = "regex-2023.6.3-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:9edcbad1f8a407e450fbac88d89e04e0b99a08473f666a3f3de0fd292badb6aa"},
-    {file = "regex-2023.6.3-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:dcba6dae7de533c876255317c11f3abe4907ba7d9aa15d13e3d9710d4315ec0e"},
-    {file = "regex-2023.6.3-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:29cdd471ebf9e0f2fb3cac165efedc3c58db841d83a518b082077e612d3ee5df"},
-    {file = "regex-2023.6.3-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:12b74fbbf6cbbf9dbce20eb9b5879469e97aeeaa874145517563cca4029db65c"},
-    {file = "regex-2023.6.3-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0c29ca1bd61b16b67be247be87390ef1d1ef702800f91fbd1991f5c4421ebae8"},
-    {file = "regex-2023.6.3-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d77f09bc4b55d4bf7cc5eba785d87001d6757b7c9eec237fe2af57aba1a071d9"},
-    {file = "regex-2023.6.3-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:ea353ecb6ab5f7e7d2f4372b1e779796ebd7b37352d290096978fea83c4dba0c"},
-    {file = "regex-2023.6.3-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:10590510780b7541969287512d1b43f19f965c2ece6c9b1c00fc367b29d8dce7"},
-    {file = "regex-2023.6.3-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:e2fbd6236aae3b7f9d514312cdb58e6494ee1c76a9948adde6eba33eb1c4264f"},
-    {file = "regex-2023.6.3-cp311-cp311-musllinux_1_1_s390x.whl", hash = "sha256:6b2675068c8b56f6bfd5a2bda55b8accbb96c02fd563704732fd1c95e2083461"},
-    {file = "regex-2023.6.3-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:74419d2b50ecb98360cfaa2974da8689cb3b45b9deff0dcf489c0d333bcc1477"},
-    {file = "regex-2023.6.3-cp311-cp311-win32.whl", hash = "sha256:fb5ec16523dc573a4b277663a2b5a364e2099902d3944c9419a40ebd56a118f9"},
-    {file = "regex-2023.6.3-cp311-cp311-win_amd64.whl", hash = "sha256:09e4a1a6acc39294a36b7338819b10baceb227f7f7dbbea0506d419b5a1dd8af"},
-    {file = "regex-2023.6.3-cp36-cp36m-macosx_10_9_x86_64.whl", hash = "sha256:0654bca0cdf28a5956c83839162692725159f4cda8d63e0911a2c0dc76166525"},
-    {file = "regex-2023.6.3-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:463b6a3ceb5ca952e66550a4532cef94c9a0c80dc156c4cc343041951aec1697"},
-    {file = "regex-2023.6.3-cp36-cp36m-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:87b2a5bb5e78ee0ad1de71c664d6eb536dc3947a46a69182a90f4410f5e3f7dd"},
-    {file = "regex-2023.6.3-cp36-cp36m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:6343c6928282c1f6a9db41f5fd551662310e8774c0e5ebccb767002fcf663ca9"},
-    {file = "regex-2023.6.3-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b6192d5af2ccd2a38877bfef086d35e6659566a335b1492786ff254c168b1693"},
-    {file = "regex-2023.6.3-cp36-cp36m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:74390d18c75054947e4194019077e243c06fbb62e541d8817a0fa822ea310c14"},
-    {file = "regex-2023.6.3-cp36-cp36m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:742e19a90d9bb2f4a6cf2862b8b06dea5e09b96c9f2df1779e53432d7275331f"},
-    {file = "regex-2023.6.3-cp36-cp36m-musllinux_1_1_aarch64.whl", hash = "sha256:8abbc5d54ea0ee80e37fef009e3cec5dafd722ed3c829126253d3e22f3846f1e"},
-    {file = "regex-2023.6.3-cp36-cp36m-musllinux_1_1_i686.whl", hash = "sha256:c2b867c17a7a7ae44c43ebbeb1b5ff406b3e8d5b3e14662683e5e66e6cc868d3"},
-    {file = "regex-2023.6.3-cp36-cp36m-musllinux_1_1_ppc64le.whl", hash = "sha256:d831c2f8ff278179705ca59f7e8524069c1a989e716a1874d6d1aab6119d91d1"},
-    {file = "regex-2023.6.3-cp36-cp36m-musllinux_1_1_s390x.whl", hash = "sha256:ee2d1a9a253b1729bb2de27d41f696ae893507c7db224436abe83ee25356f5c1"},
-    {file = "regex-2023.6.3-cp36-cp36m-musllinux_1_1_x86_64.whl", hash = "sha256:61474f0b41fe1a80e8dfa70f70ea1e047387b7cd01c85ec88fa44f5d7561d787"},
-    {file = "regex-2023.6.3-cp36-cp36m-win32.whl", hash = "sha256:0b71e63226e393b534105fcbdd8740410dc6b0854c2bfa39bbda6b0d40e59a54"},
-    {file = "regex-2023.6.3-cp36-cp36m-win_amd64.whl", hash = "sha256:bbb02fd4462f37060122e5acacec78e49c0fbb303c30dd49c7f493cf21fc5b27"},
-    {file = "regex-2023.6.3-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:b862c2b9d5ae38a68b92e215b93f98d4c5e9454fa36aae4450f61dd33ff48487"},
-    {file = "regex-2023.6.3-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:976d7a304b59ede34ca2921305b57356694f9e6879db323fd90a80f865d355a3"},
-    {file = "regex-2023.6.3-cp37-cp37m-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:83320a09188e0e6c39088355d423aa9d056ad57a0b6c6381b300ec1a04ec3d16"},
-    {file = "regex-2023.6.3-cp37-cp37m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:9427a399501818a7564f8c90eced1e9e20709ece36be701f394ada99890ea4b3"},
-    {file = "regex-2023.6.3-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7178bbc1b2ec40eaca599d13c092079bf529679bf0371c602edaa555e10b41c3"},
-    {file = "regex-2023.6.3-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:837328d14cde912af625d5f303ec29f7e28cdab588674897baafaf505341f2fc"},
-    {file = "regex-2023.6.3-cp37-cp37m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:2d44dc13229905ae96dd2ae2dd7cebf824ee92bc52e8cf03dcead37d926da019"},
-    {file = "regex-2023.6.3-cp37-cp37m-musllinux_1_1_aarch64.whl", hash = "sha256:d54af539295392611e7efbe94e827311eb8b29668e2b3f4cadcfe6f46df9c777"},
-    {file = "regex-2023.6.3-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:7117d10690c38a622e54c432dfbbd3cbd92f09401d622902c32f6d377e2300ee"},
-    {file = "regex-2023.6.3-cp37-cp37m-musllinux_1_1_ppc64le.whl", hash = "sha256:bb60b503ec8a6e4e3e03a681072fa3a5adcbfa5479fa2d898ae2b4a8e24c4591"},
-    {file = "regex-2023.6.3-cp37-cp37m-musllinux_1_1_s390x.whl", hash = "sha256:65ba8603753cec91c71de423a943ba506363b0e5c3fdb913ef8f9caa14b2c7e0"},
-    {file = "regex-2023.6.3-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:271f0bdba3c70b58e6f500b205d10a36fb4b58bd06ac61381b68de66442efddb"},
-    {file = "regex-2023.6.3-cp37-cp37m-win32.whl", hash = "sha256:9beb322958aaca059f34975b0df135181f2e5d7a13b84d3e0e45434749cb20f7"},
-    {file = "regex-2023.6.3-cp37-cp37m-win_amd64.whl", hash = "sha256:fea75c3710d4f31389eed3c02f62d0b66a9da282521075061ce875eb5300cf23"},
-    {file = "regex-2023.6.3-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:8f56fcb7ff7bf7404becdfc60b1e81a6d0561807051fd2f1860b0d0348156a07"},
-    {file = "regex-2023.6.3-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:d2da3abc88711bce7557412310dfa50327d5769a31d1c894b58eb256459dc289"},
-    {file = "regex-2023.6.3-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a99b50300df5add73d307cf66abea093304a07eb017bce94f01e795090dea87c"},
-    {file = "regex-2023.6.3-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:5708089ed5b40a7b2dc561e0c8baa9535b77771b64a8330b684823cfd5116036"},
-    {file = "regex-2023.6.3-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:687ea9d78a4b1cf82f8479cab23678aff723108df3edeac098e5b2498879f4a7"},
-    {file = "regex-2023.6.3-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:4d3850beab9f527f06ccc94b446c864059c57651b3f911fddb8d9d3ec1d1b25d"},
-    {file = "regex-2023.6.3-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e8915cc96abeb8983cea1df3c939e3c6e1ac778340c17732eb63bb96247b91d2"},
-    {file = "regex-2023.6.3-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:841d6e0e5663d4c7b4c8099c9997be748677d46cbf43f9f471150e560791f7ff"},
-    {file = "regex-2023.6.3-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:9edce5281f965cf135e19840f4d93d55b3835122aa76ccacfd389e880ba4cf82"},
-    {file = "regex-2023.6.3-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:b956231ebdc45f5b7a2e1f90f66a12be9610ce775fe1b1d50414aac1e9206c06"},
-    {file = "regex-2023.6.3-cp38-cp38-musllinux_1_1_ppc64le.whl", hash = "sha256:36efeba71c6539d23c4643be88295ce8c82c88bbd7c65e8a24081d2ca123da3f"},
-    {file = "regex-2023.6.3-cp38-cp38-musllinux_1_1_s390x.whl", hash = "sha256:cf67ca618b4fd34aee78740bea954d7c69fdda419eb208c2c0c7060bb822d747"},
-    {file = "regex-2023.6.3-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:b4598b1897837067a57b08147a68ac026c1e73b31ef6e36deeeb1fa60b2933c9"},
-    {file = "regex-2023.6.3-cp38-cp38-win32.whl", hash = "sha256:f415f802fbcafed5dcc694c13b1292f07fe0befdb94aa8a52905bd115ff41e88"},
-    {file = "regex-2023.6.3-cp38-cp38-win_amd64.whl", hash = "sha256:d4f03bb71d482f979bda92e1427f3ec9b220e62a7dd337af0aa6b47bf4498f72"},
-    {file = "regex-2023.6.3-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:ccf91346b7bd20c790310c4147eee6ed495a54ddb6737162a36ce9dbef3e4751"},
-    {file = "regex-2023.6.3-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:b28f5024a3a041009eb4c333863d7894d191215b39576535c6734cd88b0fcb68"},
-    {file = "regex-2023.6.3-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e0bb18053dfcfed432cc3ac632b5e5e5c5b7e55fb3f8090e867bfd9b054dbcbf"},
-    {file = "regex-2023.6.3-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:9a5bfb3004f2144a084a16ce19ca56b8ac46e6fd0651f54269fc9e230edb5e4a"},
-    {file = "regex-2023.6.3-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:5c6b48d0fa50d8f4df3daf451be7f9689c2bde1a52b1225c5926e3f54b6a9ed1"},
-    {file = "regex-2023.6.3-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:051da80e6eeb6e239e394ae60704d2b566aa6a7aed6f2890a7967307267a5dc6"},
-    {file = "regex-2023.6.3-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:a4c3b7fa4cdaa69268748665a1a6ff70c014d39bb69c50fda64b396c9116cf77"},
-    {file = "regex-2023.6.3-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:457b6cce21bee41ac292d6753d5e94dcbc5c9e3e3a834da285b0bde7aa4a11e9"},
-    {file = "regex-2023.6.3-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:aad51907d74fc183033ad796dd4c2e080d1adcc4fd3c0fd4fd499f30c03011cd"},
-    {file = "regex-2023.6.3-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:0385e73da22363778ef2324950e08b689abdf0b108a7d8decb403ad7f5191938"},
-    {file = "regex-2023.6.3-cp39-cp39-musllinux_1_1_ppc64le.whl", hash = "sha256:c6a57b742133830eec44d9b2290daf5cbe0a2f1d6acee1b3c7b1c7b2f3606df7"},
-    {file = "regex-2023.6.3-cp39-cp39-musllinux_1_1_s390x.whl", hash = "sha256:3e5219bf9e75993d73ab3d25985c857c77e614525fac9ae02b1bebd92f7cecac"},
-    {file = "regex-2023.6.3-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:e5087a3c59eef624a4591ef9eaa6e9a8d8a94c779dade95d27c0bc24650261cd"},
-    {file = "regex-2023.6.3-cp39-cp39-win32.whl", hash = "sha256:20326216cc2afe69b6e98528160b225d72f85ab080cbdf0b11528cbbaba2248f"},
-    {file = "regex-2023.6.3-cp39-cp39-win_amd64.whl", hash = "sha256:bdff5eab10e59cf26bc479f565e25ed71a7d041d1ded04ccf9aee1d9f208487a"},
-    {file = "regex-2023.6.3.tar.gz", hash = "sha256:72d1a25bf36d2050ceb35b517afe13864865268dfb45910e2e17a84be6cbfeb0"},
+    {file = "regex-2023.8.8-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:88900f521c645f784260a8d346e12a1590f79e96403971241e64c3a265c8ecdb"},
+    {file = "regex-2023.8.8-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:3611576aff55918af2697410ff0293d6071b7e00f4b09e005d614686ac4cd57c"},
+    {file = "regex-2023.8.8-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b8a0ccc8f2698f120e9e5742f4b38dc944c38744d4bdfc427616f3a163dd9de5"},
+    {file = "regex-2023.8.8-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:c662a4cbdd6280ee56f841f14620787215a171c4e2d1744c9528bed8f5816c96"},
+    {file = "regex-2023.8.8-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:cf0633e4a1b667bfe0bb10b5e53fe0d5f34a6243ea2530eb342491f1adf4f739"},
+    {file = "regex-2023.8.8-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:551ad543fa19e94943c5b2cebc54c73353ffff08228ee5f3376bd27b3d5b9800"},
+    {file = "regex-2023.8.8-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:54de2619f5ea58474f2ac211ceea6b615af2d7e4306220d4f3fe690c91988a61"},
+    {file = "regex-2023.8.8-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:5ec4b3f0aebbbe2fc0134ee30a791af522a92ad9f164858805a77442d7d18570"},
+    {file = "regex-2023.8.8-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:3ae646c35cb9f820491760ac62c25b6d6b496757fda2d51be429e0e7b67ae0ab"},
+    {file = "regex-2023.8.8-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:ca339088839582d01654e6f83a637a4b8194d0960477b9769d2ff2cfa0fa36d2"},
+    {file = "regex-2023.8.8-cp310-cp310-musllinux_1_1_ppc64le.whl", hash = "sha256:d9b6627408021452dcd0d2cdf8da0534e19d93d070bfa8b6b4176f99711e7f90"},
+    {file = "regex-2023.8.8-cp310-cp310-musllinux_1_1_s390x.whl", hash = "sha256:bd3366aceedf274f765a3a4bc95d6cd97b130d1dda524d8f25225d14123c01db"},
+    {file = "regex-2023.8.8-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:7aed90a72fc3654fba9bc4b7f851571dcc368120432ad68b226bd593f3f6c0b7"},
+    {file = "regex-2023.8.8-cp310-cp310-win32.whl", hash = "sha256:80b80b889cb767cc47f31d2b2f3dec2db8126fbcd0cff31b3925b4dc6609dcdb"},
+    {file = "regex-2023.8.8-cp310-cp310-win_amd64.whl", hash = "sha256:b82edc98d107cbc7357da7a5a695901b47d6eb0420e587256ba3ad24b80b7d0b"},
+    {file = "regex-2023.8.8-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:1e7d84d64c84ad97bf06f3c8cb5e48941f135ace28f450d86af6b6512f1c9a71"},
+    {file = "regex-2023.8.8-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:ce0f9fbe7d295f9922c0424a3637b88c6c472b75eafeaff6f910494a1fa719ef"},
+    {file = "regex-2023.8.8-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:06c57e14ac723b04458df5956cfb7e2d9caa6e9d353c0b4c7d5d54fcb1325c46"},
+    {file = "regex-2023.8.8-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:e7a9aaa5a1267125eef22cef3b63484c3241aaec6f48949b366d26c7250e0357"},
+    {file = "regex-2023.8.8-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:9b7408511fca48a82a119d78a77c2f5eb1b22fe88b0d2450ed0756d194fe7a9a"},
+    {file = "regex-2023.8.8-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:14dc6f2d88192a67d708341f3085df6a4f5a0c7b03dec08d763ca2cd86e9f559"},
+    {file = "regex-2023.8.8-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:48c640b99213643d141550326f34f0502fedb1798adb3c9eb79650b1ecb2f177"},
+    {file = "regex-2023.8.8-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:0085da0f6c6393428bf0d9c08d8b1874d805bb55e17cb1dfa5ddb7cfb11140bf"},
+    {file = "regex-2023.8.8-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:964b16dcc10c79a4a2be9f1273fcc2684a9eedb3906439720598029a797b46e6"},
+    {file = "regex-2023.8.8-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:7ce606c14bb195b0e5108544b540e2c5faed6843367e4ab3deb5c6aa5e681208"},
+    {file = "regex-2023.8.8-cp311-cp311-musllinux_1_1_s390x.whl", hash = "sha256:40f029d73b10fac448c73d6eb33d57b34607f40116e9f6e9f0d32e9229b147d7"},
+    {file = "regex-2023.8.8-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:3b8e6ea6be6d64104d8e9afc34c151926f8182f84e7ac290a93925c0db004bfd"},
+    {file = "regex-2023.8.8-cp311-cp311-win32.whl", hash = "sha256:942f8b1f3b223638b02df7df79140646c03938d488fbfb771824f3d05fc083a8"},
+    {file = "regex-2023.8.8-cp311-cp311-win_amd64.whl", hash = "sha256:51d8ea2a3a1a8fe4f67de21b8b93757005213e8ac3917567872f2865185fa7fb"},
+    {file = "regex-2023.8.8-cp36-cp36m-macosx_10_9_x86_64.whl", hash = "sha256:e951d1a8e9963ea51efd7f150450803e3b95db5939f994ad3d5edac2b6f6e2b4"},
+    {file = "regex-2023.8.8-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:704f63b774218207b8ccc6c47fcef5340741e5d839d11d606f70af93ee78e4d4"},
+    {file = "regex-2023.8.8-cp36-cp36m-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:22283c769a7b01c8ac355d5be0715bf6929b6267619505e289f792b01304d898"},
+    {file = "regex-2023.8.8-cp36-cp36m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:91129ff1bb0619bc1f4ad19485718cc623a2dc433dff95baadbf89405c7f6b57"},
+    {file = "regex-2023.8.8-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:de35342190deb7b866ad6ba5cbcccb2d22c0487ee0cbb251efef0843d705f0d4"},
+    {file = "regex-2023.8.8-cp36-cp36m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b993b6f524d1e274a5062488a43e3f9f8764ee9745ccd8e8193df743dbe5ee61"},
+    {file = "regex-2023.8.8-cp36-cp36m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:3026cbcf11d79095a32d9a13bbc572a458727bd5b1ca332df4a79faecd45281c"},
+    {file = "regex-2023.8.8-cp36-cp36m-musllinux_1_1_aarch64.whl", hash = "sha256:293352710172239bf579c90a9864d0df57340b6fd21272345222fb6371bf82b3"},
+    {file = "regex-2023.8.8-cp36-cp36m-musllinux_1_1_i686.whl", hash = "sha256:d909b5a3fff619dc7e48b6b1bedc2f30ec43033ba7af32f936c10839e81b9217"},
+    {file = "regex-2023.8.8-cp36-cp36m-musllinux_1_1_ppc64le.whl", hash = "sha256:3d370ff652323c5307d9c8e4c62efd1956fb08051b0e9210212bc51168b4ff56"},
+    {file = "regex-2023.8.8-cp36-cp36m-musllinux_1_1_s390x.whl", hash = "sha256:b076da1ed19dc37788f6a934c60adf97bd02c7eea461b73730513921a85d4235"},
+    {file = "regex-2023.8.8-cp36-cp36m-musllinux_1_1_x86_64.whl", hash = "sha256:e9941a4ada58f6218694f382e43fdd256e97615db9da135e77359da257a7168b"},
+    {file = "regex-2023.8.8-cp36-cp36m-win32.whl", hash = "sha256:a8c65c17aed7e15a0c824cdc63a6b104dfc530f6fa8cb6ac51c437af52b481c7"},
+    {file = "regex-2023.8.8-cp36-cp36m-win_amd64.whl", hash = "sha256:aadf28046e77a72f30dcc1ab185639e8de7f4104b8cb5c6dfa5d8ed860e57236"},
+    {file = "regex-2023.8.8-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:423adfa872b4908843ac3e7a30f957f5d5282944b81ca0a3b8a7ccbbfaa06103"},
+    {file = "regex-2023.8.8-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4ae594c66f4a7e1ea67232a0846649a7c94c188d6c071ac0210c3e86a5f92109"},
+    {file = "regex-2023.8.8-cp37-cp37m-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:e51c80c168074faa793685656c38eb7a06cbad7774c8cbc3ea05552d615393d8"},
+    {file = "regex-2023.8.8-cp37-cp37m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:09b7f4c66aa9d1522b06e31a54f15581c37286237208df1345108fcf4e050c18"},
+    {file = "regex-2023.8.8-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2e73e5243af12d9cd6a9d6a45a43570dbe2e5b1cdfc862f5ae2b031e44dd95a8"},
+    {file = "regex-2023.8.8-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:941460db8fe3bd613db52f05259c9336f5a47ccae7d7def44cc277184030a116"},
+    {file = "regex-2023.8.8-cp37-cp37m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:f0ccf3e01afeb412a1a9993049cb160d0352dba635bbca7762b2dc722aa5742a"},
+    {file = "regex-2023.8.8-cp37-cp37m-musllinux_1_1_aarch64.whl", hash = "sha256:2e9216e0d2cdce7dbc9be48cb3eacb962740a09b011a116fd7af8c832ab116ca"},
+    {file = "regex-2023.8.8-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:5cd9cd7170459b9223c5e592ac036e0704bee765706445c353d96f2890e816c8"},
+    {file = "regex-2023.8.8-cp37-cp37m-musllinux_1_1_ppc64le.whl", hash = "sha256:4873ef92e03a4309b3ccd8281454801b291b689f6ad45ef8c3658b6fa761d7ac"},
+    {file = "regex-2023.8.8-cp37-cp37m-musllinux_1_1_s390x.whl", hash = "sha256:239c3c2a339d3b3ddd51c2daef10874410917cd2b998f043c13e2084cb191684"},
+    {file = "regex-2023.8.8-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:1005c60ed7037be0d9dea1f9c53cc42f836188227366370867222bda4c3c6bd7"},
+    {file = "regex-2023.8.8-cp37-cp37m-win32.whl", hash = "sha256:e6bd1e9b95bc5614a7a9c9c44fde9539cba1c823b43a9f7bc11266446dd568e3"},
+    {file = "regex-2023.8.8-cp37-cp37m-win_amd64.whl", hash = "sha256:9a96edd79661e93327cfeac4edec72a4046e14550a1d22aa0dd2e3ca52aec921"},
+    {file = "regex-2023.8.8-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:f2181c20ef18747d5f4a7ea513e09ea03bdd50884a11ce46066bb90fe4213675"},
+    {file = "regex-2023.8.8-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:a2ad5add903eb7cdde2b7c64aaca405f3957ab34f16594d2b78d53b8b1a6a7d6"},
+    {file = "regex-2023.8.8-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:9233ac249b354c54146e392e8a451e465dd2d967fc773690811d3a8c240ac601"},
+    {file = "regex-2023.8.8-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:920974009fb37b20d32afcdf0227a2e707eb83fe418713f7a8b7de038b870d0b"},
+    {file = "regex-2023.8.8-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:cd2b6c5dfe0929b6c23dde9624483380b170b6e34ed79054ad131b20203a1a63"},
+    {file = "regex-2023.8.8-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:96979d753b1dc3b2169003e1854dc67bfc86edf93c01e84757927f810b8c3c93"},
+    {file = "regex-2023.8.8-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2ae54a338191e1356253e7883d9d19f8679b6143703086245fb14d1f20196be9"},
+    {file = "regex-2023.8.8-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:2162ae2eb8b079622176a81b65d486ba50b888271302190870b8cc488587d280"},
+    {file = "regex-2023.8.8-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:c884d1a59e69e03b93cf0dfee8794c63d7de0ee8f7ffb76e5f75be8131b6400a"},
+    {file = "regex-2023.8.8-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:cf9273e96f3ee2ac89ffcb17627a78f78e7516b08f94dc435844ae72576a276e"},
+    {file = "regex-2023.8.8-cp38-cp38-musllinux_1_1_ppc64le.whl", hash = "sha256:83215147121e15d5f3a45d99abeed9cf1fe16869d5c233b08c56cdf75f43a504"},
+    {file = "regex-2023.8.8-cp38-cp38-musllinux_1_1_s390x.whl", hash = "sha256:3f7454aa427b8ab9101f3787eb178057c5250478e39b99540cfc2b889c7d0586"},
+    {file = "regex-2023.8.8-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:f0640913d2c1044d97e30d7c41728195fc37e54d190c5385eacb52115127b882"},
+    {file = "regex-2023.8.8-cp38-cp38-win32.whl", hash = "sha256:0c59122ceccb905a941fb23b087b8eafc5290bf983ebcb14d2301febcbe199c7"},
+    {file = "regex-2023.8.8-cp38-cp38-win_amd64.whl", hash = "sha256:c12f6f67495ea05c3d542d119d270007090bad5b843f642d418eb601ec0fa7be"},
+    {file = "regex-2023.8.8-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:82cd0a69cd28f6cc3789cc6adeb1027f79526b1ab50b1f6062bbc3a0ccb2dbc3"},
+    {file = "regex-2023.8.8-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:bb34d1605f96a245fc39790a117ac1bac8de84ab7691637b26ab2c5efb8f228c"},
+    {file = "regex-2023.8.8-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:987b9ac04d0b38ef4f89fbc035e84a7efad9cdd5f1e29024f9289182c8d99e09"},
+    {file = "regex-2023.8.8-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:9dd6082f4e2aec9b6a0927202c85bc1b09dcab113f97265127c1dc20e2e32495"},
+    {file = "regex-2023.8.8-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:7eb95fe8222932c10d4436e7a6f7c99991e3fdd9f36c949eff16a69246dee2dc"},
+    {file = "regex-2023.8.8-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7098c524ba9f20717a56a8d551d2ed491ea89cbf37e540759ed3b776a4f8d6eb"},
+    {file = "regex-2023.8.8-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:4b694430b3f00eb02c594ff5a16db30e054c1b9589a043fe9174584c6efa8033"},
+    {file = "regex-2023.8.8-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:b2aeab3895d778155054abea5238d0eb9a72e9242bd4b43f42fd911ef9a13470"},
+    {file = "regex-2023.8.8-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:988631b9d78b546e284478c2ec15c8a85960e262e247b35ca5eaf7ee22f6050a"},
+    {file = "regex-2023.8.8-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:67ecd894e56a0c6108ec5ab1d8fa8418ec0cff45844a855966b875d1039a2e34"},
+    {file = "regex-2023.8.8-cp39-cp39-musllinux_1_1_ppc64le.whl", hash = "sha256:14898830f0a0eb67cae2bbbc787c1a7d6e34ecc06fbd39d3af5fe29a4468e2c9"},
+    {file = "regex-2023.8.8-cp39-cp39-musllinux_1_1_s390x.whl", hash = "sha256:f2200e00b62568cfd920127782c61bc1c546062a879cdc741cfcc6976668dfcf"},
+    {file = "regex-2023.8.8-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:9691a549c19c22d26a4f3b948071e93517bdf86e41b81d8c6ac8a964bb71e5a6"},
+    {file = "regex-2023.8.8-cp39-cp39-win32.whl", hash = "sha256:6ab2ed84bf0137927846b37e882745a827458689eb969028af8032b1b3dac78e"},
+    {file = "regex-2023.8.8-cp39-cp39-win_amd64.whl", hash = "sha256:5543c055d8ec7801901e1193a51570643d6a6ab8751b1f7dd9af71af467538bb"},
+    {file = "regex-2023.8.8.tar.gz", hash = "sha256:fcbdc5f2b0f1cd0f6a56cdb46fe41d2cce1e644e3b68832f3eeebc5fb0f7712e"},
 ]
 
 [[package]]
@@ -1700,24 +1730,42 @@ python-versions = "*"
 files = [
     {file = "safetensors-0.3.2-cp310-cp310-macosx_11_0_x86_64.whl", hash = "sha256:b6a66989075c2891d743153e8ba9ca84ee7232c8539704488f454199b8b8f84d"},
     {file = "safetensors-0.3.2-cp310-cp310-macosx_12_0_arm64.whl", hash = "sha256:670d6bc3a3b377278ce2971fa7c36ebc0a35041c4ea23b9df750a39380800195"},
+    {file = "safetensors-0.3.2-cp310-cp310-macosx_13_0_arm64.whl", hash = "sha256:564f42838721925b5313ae864ba6caa6f4c80a9fbe63cf24310c3be98ab013cd"},
+    {file = "safetensors-0.3.2-cp310-cp310-macosx_13_0_x86_64.whl", hash = "sha256:7f80af7e4ab3188daaff12d43d078da3017a90d732d38d7af4eb08b6ca2198a5"},
+    {file = "safetensors-0.3.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:ec30d78f20f1235b252d59cbb9755beb35a1fde8c24c89b3c98e6a1804cfd432"},
+    {file = "safetensors-0.3.2-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:16063d94d8f600768d3c331b1e97964b1bf3772e19710105fe24ec5a6af63770"},
     {file = "safetensors-0.3.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:cbb44e140bf2aeda98d9dde669dbec15f7b77f96a9274469b91a6cf4bcc5ec3b"},
     {file = "safetensors-0.3.2-cp310-cp310-win32.whl", hash = "sha256:2961c1243fd0da46aa6a1c835305cc4595486f8ac64632a604d0eb5f2de76175"},
     {file = "safetensors-0.3.2-cp310-cp310-win_amd64.whl", hash = "sha256:c813920482c337d1424d306e1b05824a38e3ef94303748a0a287dea7a8c4f805"},
     {file = "safetensors-0.3.2-cp311-cp311-macosx_10_11_universal2.whl", hash = "sha256:707df34bd9b9047e97332136ad98e57028faeccdb9cfe1c3b52aba5964cc24bf"},
     {file = "safetensors-0.3.2-cp311-cp311-macosx_12_0_arm64.whl", hash = "sha256:becc5bb85b2947eae20ed23b407ebfd5277d9a560f90381fe2c42e6c043677ba"},
+    {file = "safetensors-0.3.2-cp311-cp311-macosx_13_0_arm64.whl", hash = "sha256:30a75707be5cc9686490bde14b9a371cede4af53244ea72b340cfbabfffdf58a"},
+    {file = "safetensors-0.3.2-cp311-cp311-macosx_13_0_universal2.whl", hash = "sha256:54ad6af663e15e2b99e2ea3280981b7514485df72ba6d014dc22dae7ba6a5e6c"},
+    {file = "safetensors-0.3.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:37764b3197656ef507a266c453e909a3477dabc795962b38e3ad28226f53153b"},
+    {file = "safetensors-0.3.2-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:4939067736783acd8391d83cd97d6c202f94181951ce697d519f9746381b6a39"},
     {file = "safetensors-0.3.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ada0fac127ff8fb04834da5c6d85a8077e6a1c9180a11251d96f8068db922a17"},
     {file = "safetensors-0.3.2-cp311-cp311-win32.whl", hash = "sha256:155b82dbe2b0ebff18cde3f76b42b6d9470296e92561ef1a282004d449fa2b4c"},
     {file = "safetensors-0.3.2-cp311-cp311-win_amd64.whl", hash = "sha256:a86428d196959619ce90197731be9391b5098b35100a7228ef4643957648f7f5"},
     {file = "safetensors-0.3.2-cp37-cp37m-macosx_11_0_x86_64.whl", hash = "sha256:c1f8ab41ed735c5b581f451fd15d9602ff51aa88044bfa933c5fa4b1d0c644d1"},
+    {file = "safetensors-0.3.2-cp37-cp37m-macosx_13_0_x86_64.whl", hash = "sha256:bc9cfb3c9ea2aec89685b4d656f9f2296f0f0d67ecf2bebf950870e3be89b3db"},
+    {file = "safetensors-0.3.2-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:ace5d471e3d78e0d93f952707d808b5ab5eac77ddb034ceb702e602e9acf2be9"},
+    {file = "safetensors-0.3.2-cp37-cp37m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:de3e20a388b444381bcda1a3193cce51825ddca277e4cf3ed1fe8d9b2d5722cd"},
     {file = "safetensors-0.3.2-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3d7d70d48585fe8df00725aa788f2e64fd24a4c9ae07cd6be34f6859d0f89a9c"},
     {file = "safetensors-0.3.2-cp37-cp37m-win32.whl", hash = "sha256:6ff59bc90cdc857f68b1023be9085fda6202bbe7f2fd67d06af8f976d6adcc10"},
     {file = "safetensors-0.3.2-cp37-cp37m-win_amd64.whl", hash = "sha256:8b05c93da15fa911763a89281906ca333ed800ab0ef1c7ce53317aa1a2322f19"},
     {file = "safetensors-0.3.2-cp38-cp38-macosx_11_0_x86_64.whl", hash = "sha256:8969cfd9e8d904e8d3c67c989e1bd9a95e3cc8980d4f95e4dcd43c299bb94253"},
+    {file = "safetensors-0.3.2-cp38-cp38-macosx_13_0_x86_64.whl", hash = "sha256:f54148ac027556eb02187e9bc1556c4d916c99ca3cb34ca36a7d304d675035c1"},
+    {file = "safetensors-0.3.2-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:caec25fedbcf73f66c9261984f07885680f71417fc173f52279276c7f8a5edd3"},
+    {file = "safetensors-0.3.2-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:50224a1d99927ccf3b75e27c3d412f7043280431ab100b4f08aad470c37cf99a"},
     {file = "safetensors-0.3.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:aa98f49e95f02eb750d32c4947e7d5aa43883149ebd0414920866446525b70f0"},
     {file = "safetensors-0.3.2-cp38-cp38-win32.whl", hash = "sha256:33409df5e28a83dc5cc5547a3ac17c0f1b13a1847b1eb3bc4b3be0df9915171e"},
     {file = "safetensors-0.3.2-cp38-cp38-win_amd64.whl", hash = "sha256:e04a7cbbb3856159ab99e3adb14521544f65fcb8548cce773a1435a0f8d78d27"},
     {file = "safetensors-0.3.2-cp39-cp39-macosx_11_0_x86_64.whl", hash = "sha256:7c864cf5dcbfb608c5378f83319c60cc9c97263343b57c02756b7613cd5ab4dd"},
     {file = "safetensors-0.3.2-cp39-cp39-macosx_12_0_arm64.whl", hash = "sha256:14e8c19d6dc51d4f70ee33c46aff04c8ba3f95812e74daf8036c24bc86e75cae"},
+    {file = "safetensors-0.3.2-cp39-cp39-macosx_13_0_arm64.whl", hash = "sha256:042a60f633c3c7009fdf6a7c182b165cb7283649d2a1e9c7a4a1c23454bd9a5b"},
+    {file = "safetensors-0.3.2-cp39-cp39-macosx_13_0_x86_64.whl", hash = "sha256:fafd95e5ef41e8f312e2a32b7031f7b9b2a621b255f867b221f94bb2e9f51ae8"},
+    {file = "safetensors-0.3.2-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8ed77cf358abce2307f03634694e0b2a29822e322a1623e0b1aa4b41e871bf8b"},
+    {file = "safetensors-0.3.2-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:5d344e8b2681a33aafc197c90b0def3229b3317d749531c72fa6259d0caa5c8c"},
     {file = "safetensors-0.3.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:87ff0024ef2e5722a79af24688ce4a430f70601d0cf712a744105ed4b8f67ba5"},
     {file = "safetensors-0.3.2-cp39-cp39-win32.whl", hash = "sha256:827af9478b78977248ba93e2fd97ea307fb63f463f80cef4824460f8c2542a52"},
     {file = "safetensors-0.3.2-cp39-cp39-win_amd64.whl", hash = "sha256:9b09f27c456efa301f98681ea14b12f81f2637889f6336223ccab71e42c34541"},
@@ -1945,19 +1993,31 @@ files = [
 
 [[package]]
 name = "torch"
-version = "2.0.1+cu118"
+version = "2.0.1"
 description = "Tensors and Dynamic neural networks in Python with strong GPU acceleration"
 optional = false
 python-versions = ">=3.8.0"
 files = [
-    {file = "torch-2.0.1+cu118-cp310-cp310-linux_x86_64.whl", hash = "sha256:a7a49d459bf4862f64f7bc1a68beccf8881c2fa9f3e0569608e16ba6f85ebf7b"},
-    {file = "torch-2.0.1+cu118-cp310-cp310-win_amd64.whl", hash = "sha256:f58d75619bc96e4322343c030b893613701caa2d6db8017155da226c14171335"},
-    {file = "torch-2.0.1+cu118-cp311-cp311-linux_x86_64.whl", hash = "sha256:143b6c658c17d43376e2dfbaa2c106d35639d615e5e8dec4429cf1e510dd8d61"},
-    {file = "torch-2.0.1+cu118-cp311-cp311-win_amd64.whl", hash = "sha256:b663a4ee744d574095dbd612644de345944247c0605692309fd9f6c7ccdea022"},
-    {file = "torch-2.0.1+cu118-cp38-cp38-linux_x86_64.whl", hash = "sha256:2ce38a6e4ea7c4b7f5baa51e65243a5f687f6e19ab7915ba5b2a431105f50bbe"},
-    {file = "torch-2.0.1+cu118-cp38-cp38-win_amd64.whl", hash = "sha256:e58d26a11bd57ac19761c018c3151c15bc71d068afc8ec409bfd9b4cfcc63a52"},
-    {file = "torch-2.0.1+cu118-cp39-cp39-linux_x86_64.whl", hash = "sha256:eb55f29db5744eda8a96f5594e637daed0d52278273005de759970e67cfa6a5a"},
-    {file = "torch-2.0.1+cu118-cp39-cp39-win_amd64.whl", hash = "sha256:fa225b6f941ee0e78978ac85ed7744d3c19fff462473821f8060c14faa60043e"},
+    {file = "torch-2.0.1-cp310-cp310-manylinux1_x86_64.whl", hash = "sha256:8ced00b3ba471856b993822508f77c98f48a458623596a4c43136158781e306a"},
+    {file = "torch-2.0.1-cp310-cp310-manylinux2014_aarch64.whl", hash = "sha256:359bfaad94d1cda02ab775dc1cc386d585712329bb47b8741607ef6ef4950747"},
+    {file = "torch-2.0.1-cp310-cp310-win_amd64.whl", hash = "sha256:7c84e44d9002182edd859f3400deaa7410f5ec948a519cc7ef512c2f9b34d2c4"},
+    {file = "torch-2.0.1-cp310-none-macosx_10_9_x86_64.whl", hash = "sha256:567f84d657edc5582d716900543e6e62353dbe275e61cdc36eda4929e46df9e7"},
+    {file = "torch-2.0.1-cp310-none-macosx_11_0_arm64.whl", hash = "sha256:787b5a78aa7917465e9b96399b883920c88a08f4eb63b5a5d2d1a16e27d2f89b"},
+    {file = "torch-2.0.1-cp311-cp311-manylinux1_x86_64.whl", hash = "sha256:e617b1d0abaf6ced02dbb9486803abfef0d581609b09641b34fa315c9c40766d"},
+    {file = "torch-2.0.1-cp311-cp311-manylinux2014_aarch64.whl", hash = "sha256:b6019b1de4978e96daa21d6a3ebb41e88a0b474898fe251fd96189587408873e"},
+    {file = "torch-2.0.1-cp311-cp311-win_amd64.whl", hash = "sha256:dbd68cbd1cd9da32fe5d294dd3411509b3d841baecb780b38b3b7b06c7754434"},
+    {file = "torch-2.0.1-cp311-none-macosx_10_9_x86_64.whl", hash = "sha256:ef654427d91600129864644e35deea761fb1fe131710180b952a6f2e2207075e"},
+    {file = "torch-2.0.1-cp311-none-macosx_11_0_arm64.whl", hash = "sha256:25aa43ca80dcdf32f13da04c503ec7afdf8e77e3a0183dd85cd3e53b2842e527"},
+    {file = "torch-2.0.1-cp38-cp38-manylinux1_x86_64.whl", hash = "sha256:5ef3ea3d25441d3957348f7e99c7824d33798258a2bf5f0f0277cbcadad2e20d"},
+    {file = "torch-2.0.1-cp38-cp38-manylinux2014_aarch64.whl", hash = "sha256:0882243755ff28895e8e6dc6bc26ebcf5aa0911ed81b2a12f241fc4b09075b13"},
+    {file = "torch-2.0.1-cp38-cp38-win_amd64.whl", hash = "sha256:f66aa6b9580a22b04d0af54fcd042f52406a8479e2b6a550e3d9f95963e168c8"},
+    {file = "torch-2.0.1-cp38-none-macosx_10_9_x86_64.whl", hash = "sha256:1adb60d369f2650cac8e9a95b1d5758e25d526a34808f7448d0bd599e4ae9072"},
+    {file = "torch-2.0.1-cp38-none-macosx_11_0_arm64.whl", hash = "sha256:1bcffc16b89e296826b33b98db5166f990e3b72654a2b90673e817b16c50e32b"},
+    {file = "torch-2.0.1-cp39-cp39-manylinux1_x86_64.whl", hash = "sha256:e10e1597f2175365285db1b24019eb6f04d53dcd626c735fc502f1e8b6be9875"},
+    {file = "torch-2.0.1-cp39-cp39-manylinux2014_aarch64.whl", hash = "sha256:423e0ae257b756bb45a4b49072046772d1ad0c592265c5080070e0767da4e490"},
+    {file = "torch-2.0.1-cp39-cp39-win_amd64.whl", hash = "sha256:8742bdc62946c93f75ff92da00e3803216c6cce9b132fbca69664ca38cfb3e18"},
+    {file = "torch-2.0.1-cp39-none-macosx_10_9_x86_64.whl", hash = "sha256:c62df99352bd6ee5a5a8d1832452110435d178b5164de450831a3a8cc14dc680"},
+    {file = "torch-2.0.1-cp39-none-macosx_11_0_arm64.whl", hash = "sha256:671a2565e3f63b8fe8e42ae3e36ad249fe5e567435ea27b94edaa672a7d0c416"},
 ]
 
 [package.dependencies]
@@ -1965,33 +2025,27 @@ filelock = "*"
 jinja2 = "*"
 networkx = "*"
 sympy = "*"
-triton = {version = "2.0.0", markers = "platform_system == \"Linux\" and platform_machine == \"x86_64\""}
 typing-extensions = "*"
 
 [package.extras]
 opt-einsum = ["opt-einsum (>=3.3)"]
 
-[package.source]
-type = "legacy"
-url = "https://download.pytorch.org/whl/cu118"
-reference = "pytorch-gpu-src"
-
 [[package]]
 name = "tqdm"
-version = "4.65.0"
+version = "4.66.1"
 description = "Fast, Extensible Progress Meter"
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "tqdm-4.65.0-py3-none-any.whl", hash = "sha256:c4f53a17fe37e132815abceec022631be8ffe1b9381c2e6e30aa70edc99e9671"},
-    {file = "tqdm-4.65.0.tar.gz", hash = "sha256:1871fb68a86b8fb3b59ca4cdd3dcccbc7e6d613eeed31f4c332531977b89beb5"},
+    {file = "tqdm-4.66.1-py3-none-any.whl", hash = "sha256:d302b3c5b53d47bce91fea46679d9c3c6508cf6332229aa1e7d8653723793386"},
+    {file = "tqdm-4.66.1.tar.gz", hash = "sha256:d88e651f9db8d8551a62556d3cff9e3034274ca5d66e93197cf2490e2dcb69c7"},
 ]
 
 [package.dependencies]
 colorama = {version = "*", markers = "platform_system == \"Windows\""}
 
 [package.extras]
-dev = ["py-make (>=0.1.0)", "twine", "wheel"]
+dev = ["pytest (>=6)", "pytest-cov", "pytest-timeout", "pytest-xdist"]
 notebook = ["ipywidgets (>=6)"]
 slack = ["slack-sdk"]
 telegram = ["requests"]
@@ -2065,43 +2119,6 @@ torchhub = ["filelock", "huggingface-hub (>=0.14.1,<1.0)", "importlib-metadata",
 video = ["av (==9.2.0)", "decord (==0.6.0)"]
 vision = ["Pillow (<10.0.0)"]
 
-[[package]]
-name = "triton"
-version = "2.0.0"
-description = "A language and compiler for custom Deep Learning operations"
-optional = false
-python-versions = "*"
-files = [
-    {file = "triton-2.0.0-1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:38806ee9663f4b0f7cd64790e96c579374089e58f49aac4a6608121aa55e2505"},
-    {file = "triton-2.0.0-1-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:226941c7b8595219ddef59a1fdb821e8c744289a132415ddd584facedeb475b1"},
-    {file = "triton-2.0.0-1-cp36-cp36m-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:4c9fc8c89874bc48eb7e7b2107a9b8d2c0bf139778637be5bfccb09191685cfd"},
-    {file = "triton-2.0.0-1-cp37-cp37m-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:d2684b6a60b9f174f447f36f933e9a45f31db96cb723723ecd2dcfd1c57b778b"},
-    {file = "triton-2.0.0-1-cp38-cp38-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:9d4978298b74fcf59a75fe71e535c092b023088933b2f1df933ec32615e4beef"},
-    {file = "triton-2.0.0-1-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:74f118c12b437fb2ca25e1a04759173b517582fcf4c7be11913316c764213656"},
-    {file = "triton-2.0.0-1-pp37-pypy37_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:9618815a8da1d9157514f08f855d9e9ff92e329cd81c0305003eb9ec25cc5add"},
-    {file = "triton-2.0.0-1-pp38-pypy38_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1aca3303629cd3136375b82cb9921727f804e47ebee27b2677fef23005c3851a"},
-    {file = "triton-2.0.0-1-pp39-pypy39_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:e3e13aa8b527c9b642e3a9defcc0fbd8ffbe1c80d8ac8c15a01692478dc64d8a"},
-    {file = "triton-2.0.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8f05a7e64e4ca0565535e3d5d3405d7e49f9d308505bb7773d21fb26a4c008c2"},
-    {file = "triton-2.0.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:bb4b99ca3c6844066e516658541d876c28a5f6e3a852286bbc97ad57134827fd"},
-    {file = "triton-2.0.0-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:47b4d70dc92fb40af553b4460492c31dc7d3a114a979ffb7a5cdedb7eb546c08"},
-    {file = "triton-2.0.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fedce6a381901b1547e0e7e1f2546e4f65dca6d91e2d8a7305a2d1f5551895be"},
-    {file = "triton-2.0.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:75834f27926eab6c7f00ce73aaf1ab5bfb9bec6eb57ab7c0bfc0a23fac803b4c"},
-    {file = "triton-2.0.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0117722f8c2b579cd429e0bee80f7731ae05f63fe8e9414acd9a679885fcbf42"},
-    {file = "triton-2.0.0-pp37-pypy37_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:bcd9be5d0c2e45d2b7e6ddc6da20112b6862d69741576f9c3dbaf941d745ecae"},
-    {file = "triton-2.0.0-pp38-pypy38_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:42a0d2c3fc2eab4ba71384f2e785fbfd47aa41ae05fa58bf12cb31dcbd0aeceb"},
-    {file = "triton-2.0.0-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:52c47b72c72693198163ece9d90a721299e4fb3b8e24fd13141e384ad952724f"},
-]
-
-[package.dependencies]
-cmake = "*"
-filelock = "*"
-lit = "*"
-torch = "*"
-
-[package.extras]
-tests = ["autopep8", "flake8", "isort", "numpy", "pytest", "scipy (>=1.7.1)"]
-tutorials = ["matplotlib", "pandas", "tabulate"]
-
 [[package]]
 name = "typer"
 version = "0.6.1"
@@ -2448,4 +2465,4 @@ quantize = ["accelerate", "datasets", "texttable"]
 [metadata]
 lock-version = "2.0"
 python-versions = ">=3.9,<3.13"
-content-hash = "6b6dbd9e11bc2f71cd88e2ab50adce249a722fea02e7cdcf5fe1b94c7b997a57"
+content-hash = "91a848038c08a44c67acfb4257781440ccc1e74a4b82f09513e75588fa33f72b"
diff --git a/server/pyproject.toml b/server/pyproject.toml
index 8eeae616..0e6d7882 100644
--- a/server/pyproject.toml
+++ b/server/pyproject.toml
@@ -31,8 +31,9 @@ einops = "^0.6.1"
 texttable = { version = "^1.6.7", optional = true }
 datasets = { version = "^2.14.0", optional = true }
 peft = "^0.4.0"
-torch = {version = "^2.0.1+cu118", source = "pytorch-gpu-src"}
+torch = { version = "^2.0.1" }
 scipy = "^1.11.1"
+pillow = "^10.0.0"
 
 [tool.poetry.extras]
 accelerate = ["accelerate"]
diff --git a/server/requirements.txt b/server/requirements.txt
index 5cc19619..012e91a2 100644
--- a/server/requirements.txt
+++ b/server/requirements.txt
@@ -1,11 +1,8 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-
 accelerate==0.20.3 ; python_version >= "3.9" and python_version < "3.13"
 backoff==2.2.1 ; python_version >= "3.9" and python_version < "3.13"
 certifi==2023.7.22 ; python_version >= "3.9" and python_version < "3.13"
 charset-normalizer==3.2.0 ; python_version >= "3.9" and python_version < "3.13"
 click==8.1.6 ; python_version >= "3.9" and python_version < "3.13"
-cmake==3.27.0 ; platform_system == "Linux" and platform_machine == "x86_64" and python_version >= "3.9" and python_version < "3.13"
 colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.13" and (sys_platform == "win32" or platform_system == "Windows")
 deprecated==1.2.14 ; python_version >= "3.9" and python_version < "3.13"
 einops==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
@@ -13,14 +10,13 @@ filelock==3.12.2 ; python_version >= "3.9" and python_version < "3.13"
 fsspec==2023.6.0 ; python_version >= "3.9" and python_version < "3.13"
 googleapis-common-protos==1.60.0 ; python_version >= "3.9" and python_version < "3.13"
 grpc-interceptor==0.15.2 ; python_version >= "3.9" and python_version < "3.13"
-grpcio-reflection==1.56.2 ; python_version >= "3.9" and python_version < "3.13"
-grpcio-status==1.56.2 ; python_version >= "3.9" and python_version < "3.13"
-grpcio==1.56.2 ; python_version >= "3.9" and python_version < "3.13"
+grpcio-reflection==1.57.0 ; python_version >= "3.9" and python_version < "3.13"
+grpcio-status==1.57.0 ; python_version >= "3.9" and python_version < "3.13"
+grpcio==1.57.0 ; python_version >= "3.9" and python_version < "3.13"
 hf-transfer==0.1.3 ; python_version >= "3.9" and python_version < "3.13"
 huggingface-hub==0.14.1 ; python_version >= "3.9" and python_version < "3.13"
 idna==3.4 ; python_version >= "3.9" and python_version < "3.13"
 jinja2==3.1.2 ; python_version >= "3.9" and python_version < "3.13"
-lit==16.0.6 ; platform_system == "Linux" and platform_machine == "x86_64" and python_version >= "3.9" and python_version < "3.13"
 loguru==0.6.0 ; python_version >= "3.9" and python_version < "3.13"
 markupsafe==2.1.3 ; python_version >= "3.9" and python_version < "3.13"
 mpmath==1.3.0 ; python_version >= "3.9" and python_version < "3.13"
@@ -37,10 +33,11 @@ opentelemetry-sdk==1.15.0 ; python_version >= "3.9" and python_version < "3.13"
 opentelemetry-semantic-conventions==0.36b0 ; python_version >= "3.9" and python_version < "3.13"
 packaging==23.1 ; python_version >= "3.9" and python_version < "3.13"
 peft==0.4.0 ; python_version >= "3.9" and python_version < "3.13"
-protobuf==4.23.4 ; python_version >= "3.9" and python_version < "3.13"
+pillow==10.0.0 ; python_version >= "3.9" and python_version < "3.13"
+protobuf==4.24.0 ; python_version >= "3.9" and python_version < "3.13"
 psutil==5.9.5 ; python_version >= "3.9" and python_version < "3.13"
 pyyaml==6.0.1 ; python_version >= "3.9" and python_version < "3.13"
-regex==2023.6.3 ; python_version >= "3.9" and python_version < "3.13"
+regex==2023.8.8 ; python_version >= "3.9" and python_version < "3.13"
 requests==2.31.0 ; python_version >= "3.9" and python_version < "3.13"
 safetensors==0.3.2 ; python_version >= "3.9" and python_version < "3.13"
 scipy==1.11.1 ; python_version >= "3.9" and python_version < "3.13"
@@ -48,10 +45,9 @@ sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.13"
 setuptools==68.0.0 ; python_version >= "3.9" and python_version < "3.13"
 sympy==1.12 ; python_version >= "3.9" and python_version < "3.13"
 tokenizers==0.13.3 ; python_version >= "3.9" and python_version < "3.13"
-torch==2.0.1+cu118 ; python_version >= "3.9" and python_version < "3.13"
-tqdm==4.65.0 ; python_version >= "3.9" and python_version < "3.13"
+torch==2.0.1 ; python_version >= "3.9" and python_version < "3.13"
+tqdm==4.66.1 ; python_version >= "3.9" and python_version < "3.13"
 transformers==4.31.0 ; python_version >= "3.9" and python_version < "3.13"
-triton==2.0.0 ; platform_system == "Linux" and platform_machine == "x86_64" and python_version >= "3.9" and python_version < "3.13"
 typer==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
 typing-extensions==4.7.1 ; python_version >= "3.9" and python_version < "3.13"
 urllib3==2.0.4 ; python_version >= "3.9" and python_version < "3.13"
diff --git a/server/text_generation_server/models/__init__.py b/server/text_generation_server/models/__init__.py
index 621652e8..932ab32e 100644
--- a/server/text_generation_server/models/__init__.py
+++ b/server/text_generation_server/models/__init__.py
@@ -54,6 +54,7 @@ try:
     from text_generation_server.models.flash_santacoder import (
         FlashSantacoderSharded,
     )
+    from text_generation_server.models.idefics import IDEFICSSharded
 
 except ImportError as e:
     logger.warning(f"Could not import Flash Attention enabled models: {e}")
@@ -64,6 +65,7 @@ if FLASH_ATTENTION:
     __all__.append(FlashRWSharded)
     __all__.append(FlashSantacoderSharded)
     __all__.append(FlashLlama)
+    __all__.append(IDEFICSSharded)
 
 
 def get_model(
@@ -248,6 +250,17 @@ def get_model(
             dtype=dtype,
             trust_remote_code=trust_remote_code,
         )
+    elif model_type == "idefics":
+        if FLASH_ATTENTION:
+           return IDEFICSSharded(
+               model_id,
+               revision,
+               quantize=quantize,
+               dtype=dtype,
+               trust_remote_code=trust_remote_code,
+           )
+        else:
+            raise NotImplementedError(FLASH_ATT_ERROR_MESSAGE.format("Idefics"))
 
     if sharded:
         raise ValueError("sharded is not supported for AutoModel")
diff --git a/server/text_generation_server/models/custom_modeling/idefics_config.py b/server/text_generation_server/models/custom_modeling/idefics_config.py
new file mode 100644
index 00000000..34925087
--- /dev/null
+++ b/server/text_generation_server/models/custom_modeling/idefics_config.py
@@ -0,0 +1,323 @@
+# coding=utf-8
+# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
+#
+# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
+# and OPT implementations in this library. It has been modified from its
+# original forms to accommodate minor architectural differences compared
+# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Idefics model configuration"""
+import copy
+
+from transformers import PretrainedConfig
+
+IDEFICS_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "HuggingFaceM4/idefics-9b": "https://huggingface.co/HuggingFaceM4/idefics-9b/blob/main/config.json",
+    "HuggingFaceM4/idefics-80b": "https://huggingface.co/HuggingFaceM4/idefics-80b/blob/main/config.json",
+}
+
+
+class IdeficsVisionConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`IdeficsModel`]. It is used to instantiate an
+    Idefics model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of the Idefics-9B.
+    e.g. [HuggingFaceM4/idefics-9b](https://huggingface.co/HuggingFaceM4/idefics-9b)
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+    Args:
+        hidden_size (`int`, *optional*, defaults to 768):
+            Dimensionality of the encoder layers and the pooler layer. (elsewhere referred to as `hidden_size`)
+        image_size (`int`, *optional*, defaults to 224):
+            The size (resolution) of each image.
+        intermediate_size (`int`, *optional*, defaults to 5120):
+            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        patch_size (`int`, *optional*, defaults to 14):
+            The size (resolution) of each patch.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 16):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        image_num_channels (`int`, *optional*, defaults to `3`):
+            Number of image channels.
+        hidden_act (`str` or `function`, *optional*, defaults to `"quick_gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"selu"` and `"gelu_new"` ``"quick_gelu"` are supported.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        initializer_factor (`float`, *optional*, defaults to 1.0):
+            A factor for initializing all weight matrices (should be kept to 1.0, used internally for initialization
+            testing).
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+    """
+    model_type = "idefics"
+    attribute_map = {
+        "hidden_size": "embed_dim",
+    }
+
+    def __init__(
+        self,
+        embed_dim=768,
+        image_size=224,
+        intermediate_size=5120,
+        patch_size=14,
+        num_hidden_layers=32,
+        num_attention_heads=16,
+        num_channels=3,
+        hidden_act="quick_gelu",
+        layer_norm_eps=1e-5,
+        attention_dropout=0.0,
+        initializer_range=0.02,
+        initializer_factor=1.0,
+        **kwargs,
+    ):
+        self.embed_dim = embed_dim
+        self.image_size = image_size
+        self.intermediate_size = intermediate_size
+        self.patch_size = patch_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_channels = num_channels
+        self.layer_norm_eps = layer_norm_eps
+        self.attention_dropout = attention_dropout
+        self.initializer_range = initializer_range
+        self.initializer_factor = initializer_factor
+        self.hidden_act = hidden_act
+
+        super().__init__(**kwargs)
+
+
+class IdeficsPerceiverConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`IdeficsModel`]. It is used to instantiate an
+    Idefics model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of the Idefics-9B.
+    e.g. [HuggingFaceM4/idefics-9b](https://huggingface.co/HuggingFaceM4/idefics-9b)
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+    Args:
+        use_resampler (`bool`, *optional*, defaults to `False`):
+            Whether or not to use the resampler
+        resampler_n_latents (`int`, *optional*, defaults to ):
+            Number of latent embeddings to resample ("compress") the input sequence to (usually < 128).
+        resampler_depth (`int`, *optional*, defaults to 6):
+            Depth of the Perceiver Resampler (Transformer w/ cross attention). Should be shallow (< 3).
+        resampler_n_heads (`int`, *optional*, defaults to 16):
+            Number of heads in each Transformer block (for multi-headed self-attention).
+        resampler_head_dim (`int`, *optional*, defaults to 96):
+            Dimensionality of each head projection in the Transformer block.
+        qk_layer_norms_perceiver (`bool`, *optional*, defaults to `False`):
+            Whether or not to use qk layer norms in perceiver
+    """
+    model_type = "idefics"
+
+    def __init__(
+        self,
+        use_resampler=False,
+        resampler_n_latents=64,
+        resampler_depth=6,
+        resampler_n_heads=16,
+        resampler_head_dim=96,
+        qk_layer_norms_perceiver=False,
+        **kwargs,
+    ):
+        self.use_resampler = use_resampler
+        self.resampler_n_latents = resampler_n_latents
+        self.resampler_depth = resampler_depth
+        self.resampler_n_heads = resampler_n_heads
+        self.resampler_head_dim = resampler_head_dim
+        self.qk_layer_norms_perceiver = qk_layer_norms_perceiver
+
+        super().__init__(**kwargs)
+
+
+class IdeficsConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`IdeficsModel`]. It is used to instantiate an
+    Idefics model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of the Idefics-9B.
+    e.g. [HuggingFaceM4/idefics-9b](https://huggingface.co/HuggingFaceM4/idefics-9b)
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+    Args:
+        additional_vocab_size (`int`, *optional`, defaults to 0):
+            Additional vocabulary size of the model, typically for the special "<img>" token. Additional vocab tokens
+            are always trainable whereas regular vocab tokens can be frozen or not.
+        vocab_size (`int`, *optional*, defaults to 32000):
+            Vocabulary size of the Idefics model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`~IdeficsModel`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 11008):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        dropout (`float`, *optional*, defaults to 0.0):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        alpha_initializer (`str`, *optional*, defaults to `"zeros"`):
+            Initialization type for the alphas.
+        alphas_initializer_range (`float`, *optional*, defaults to 0.0):
+            The standard deviation of the truncated_normal_initializer for initializing the alphas in the Gated Cross
+            Attention.
+        alpha_type (`str`, *optional*, defaults to `"float"`):
+            Whether the gating alphas should be vectors or single floats.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-6):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*, defaults to 0)
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 1)
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 2)
+            End of stream token id.
+        tie_word_embeddings(`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        cross_layer_interval (`int`, *optional*, default to 1)
+            Interval for cross attention (from text to image) layers.
+        qk_layer_norms (`bool`, *optional*, defaults to `False`): Whether to add layer norm after q and k
+        freeze_text_layers (`bool`, *optional*, defaults to `True`): Whether to freeze text layers
+        freeze_text_module_exceptions (`bool`, *optional*, defaults to `[]`):
+            Exceptions to freezing text layers when `freeze_text_layers` is `True`
+        freeze_lm_head (`bool`, *optional*, defaults to `False`): Whether to freeze lm head
+        freeze_vision_layers (`bool`, *optional*, defaults to `True`):  Whether to freeze vision layers
+        freeze_vision_module_exceptions (`bool`, *optional*, defaults to `[]`):
+            Exceptions to freezing vision layers when `freeze_vision_layers` is `True`
+        use_resampler (`bool`, *optional*, defaults to `False`): Whether to use the Resampler
+        vision_config (`IdeficsVisionConfig`,  *optional*): Custom vision config or dict
+        perceiver_config (`IdeficsPerceiverConfig`,  *optional*): Custom perceiver config or dict
+    Example:
+    ```python
+    >>> from transformers import IdeficsModel, IdeficsConfig
+    >>> # Initializing a Idefics idefics-9b style configuration
+    >>> configuration = IdeficsConfig()
+    >>> # Initializing a model from the idefics-9b style configuration
+    >>> model = IdeficsModel(configuration)
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "idefics"
+    is_composition = True
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        additional_vocab_size=0,
+        hidden_size=4096,
+        intermediate_size=11008,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        dropout=0.0,
+        hidden_act="silu",
+        initializer_range=0.02,
+        alpha_initializer="zeros",
+        alphas_initializer_range=0.0,
+        alpha_type="float",
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=0,
+        bos_token_id=1,
+        eos_token_id=2,
+        tie_word_embeddings=False,
+        cross_layer_interval=1,
+        qk_layer_norms=False,
+        freeze_text_layers=True,
+        freeze_text_module_exceptions=[],
+        freeze_lm_head=False,
+        freeze_vision_layers=True,
+        freeze_vision_module_exceptions=[],
+        use_resampler=False,
+        vision_config=None,
+        perceiver_config=None,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.additional_vocab_size = additional_vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.dropout = dropout
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.alpha_initializer = alpha_initializer
+        self.alphas_initializer_range = alphas_initializer_range
+        self.alpha_type = alpha_type
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+
+        self.cross_layer_interval = cross_layer_interval
+        self.qk_layer_norms = qk_layer_norms
+        self.freeze_vision_layers = freeze_vision_layers
+
+        self.freeze_text_layers = freeze_text_layers
+        self.freeze_text_module_exceptions = freeze_text_module_exceptions
+        self.freeze_vision_module_exceptions = freeze_vision_module_exceptions
+        self.freeze_lm_head = freeze_lm_head
+
+        self.use_resampler = use_resampler
+
+        if perceiver_config is None:
+            self.perceiver_config = IdeficsPerceiverConfig()
+        elif isinstance(perceiver_config, dict):
+            self.perceiver_config = IdeficsPerceiverConfig(**perceiver_config)
+        elif isinstance(perceiver_config, IdeficsPerceiverConfig):
+            self.perceiver_config = perceiver_config
+
+        if vision_config is None:
+            self.vision_config = IdeficsVisionConfig()
+        elif isinstance(vision_config, dict):
+            self.vision_config = IdeficsVisionConfig(**vision_config)
+        elif isinstance(vision_config, IdeficsVisionConfig):
+            self.vision_config = vision_config
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+        # IMPORTANT: Do not do any __init__ args-based checks in the constructor, since
+        # PretrainedConfig.from_dict first instantiates the class with the config dict and only then
+        # updates the config object with `kwargs` from from_pretrained, so during the instantiation
+        # of this object many attributes have default values and haven't yet been overridden.
+        # Do any required checks inside `from_pretrained` once the superclass' `from_pretrained` was run.
+
+    def to_dict(self):
+        """
+        Serializes this instance to a Python dictionary. Override the default [`~PretrainedConfig.to_dict`].
+        Returns:
+            `Dict[str, any]`: Dictionary of all the attributes that make up this configuration instance,
+        """
+        output = copy.deepcopy(self.__dict__)
+
+        output["vision_config"] = self.vision_config.to_dict()
+        output["perceiver_config"] = self.perceiver_config.to_dict()
+        output["model_type"] = self.__class__.model_type
+
+        return output
diff --git a/server/text_generation_server/models/custom_modeling/idefics_image_processing.py b/server/text_generation_server/models/custom_modeling/idefics_image_processing.py
new file mode 100644
index 00000000..aec9a3dc
--- /dev/null
+++ b/server/text_generation_server/models/custom_modeling/idefics_image_processing.py
@@ -0,0 +1,264 @@
+# coding=utf-8
+# Copyright 2022 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Image processor class for Idefics."""
+
+from typing import Callable, Dict, List, Optional, Union, Iterable
+import numpy as np
+
+from PIL import Image
+
+from transformers.image_processing_utils import BaseImageProcessor, BatchFeature
+from transformers.image_transforms import resize, to_channel_dimension_format, rescale, normalize
+from transformers.image_utils import (
+    ChannelDimension,
+    ImageInput,
+    PILImageResampling,
+    make_list_of_images,
+    to_numpy_array,
+    valid_images,
+)
+from io import BytesIO
+import requests
+from transformers import TensorType, is_torch_available
+
+
+IDEFICS_STANDARD_MEAN = [0.48145466, 0.4578275, 0.40821073]
+IDEFICS_STANDARD_STD = [0.26862954, 0.26130258, 0.27577711]
+
+
+def convert_to_rgb(image):
+    # `image.convert("RGB")` would only work for .jpg images, as it creates a wrong background
+    # for transparent images. The call to `alpha_composite` handles this case
+    if image.mode == "RGB":
+        return image
+
+    image_rgba = image.convert("RGBA")
+    background = Image.new("RGBA", image_rgba.size, (255, 255, 255))
+    alpha_composite = Image.alpha_composite(background, image_rgba)
+    alpha_composite = alpha_composite.convert("RGB")
+    return alpha_composite
+
+
+class IdeficsImageProcessor(BaseImageProcessor):
+    r"""
+    Constructs a Idefics image processor.
+    Args:
+        image_size (`int`, *optional*, defaults to `224`):
+            Resize to image size
+        image_num_channels (`int`, *optional*, defaults to `3`):
+            Number of image channels.
+        image_mean (`float` or `List[float]`, *optional*, defaults to `IDEFICS_STANDARD_MEAN`):
+            Mean to use if normalizing the image. This is a float or list of floats the length of the number of
+            channels in the image. Can be overridden by the `image_mean` parameter in the `preprocess` method. Can be
+            overridden by the `image_mean` parameter in the `preprocess` method.
+        image_std (`float` or `List[float]`, *optional*, defaults to `IDEFICS_STANDARD_STD`):
+            Standard deviation to use if normalizing the image. This is a float or list of floats the length of the
+            number of channels in the image. Can be overridden by the `image_std` parameter in the `preprocess` method.
+            Can be overridden by the `image_std` parameter in the `preprocess` method.
+    """
+
+    model_input_names = ["pixel_values"]
+
+    def __init__(
+        self,
+        image_size: int = 224,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        image_num_channels: Optional[int] = 3,
+        **kwargs,
+    ) -> None:
+        super().__init__(**kwargs)
+
+        self.image_size = image_size
+        self.image_num_channels = image_num_channels
+        self.image_mean = image_mean
+        self.image_std = image_std
+
+    def preprocess(
+        self,
+        images: ImageInput,
+        image_num_channels: Optional[int] = 3,
+        image_size: Optional[Dict[str, int]] = None,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        transform: Callable = None,
+        **kwargs,
+    ) -> TensorType.PYTORCH:
+        """
+        Preprocess a batch of images.
+        Args:
+            images (`ImageInput`):
+                A list of images to preprocess.
+            image_size (`int`, *optional*, defaults to `self.image_size`):
+                Resize to image size
+            image_num_channels (`int`, *optional*, defaults to `self.image_num_channels`):
+                Number of image channels.
+            image_mean (`float` or `List[float]`, *optional*, defaults to `IDEFICS_STANDARD_MEAN`):
+                Mean to use if normalizing the image. This is a float or list of floats the length of the number of
+                channels in the image. Can be overridden by the `image_mean` parameter in the `preprocess` method. Can
+                be overridden by the `image_mean` parameter in the `preprocess` method.
+            image_std (`float` or `List[float]`, *optional*, defaults to `IDEFICS_STANDARD_STD`):
+                Standard deviation to use if normalizing the image. This is a float or list of floats the length of the
+                number of channels in the image. Can be overridden by the `image_std` parameter in the `preprocess`
+                method. Can be overridden by the `image_std` parameter in the `preprocess` method.
+            transform (`Callable`, *optional*, defaults to `None`):
+                A custom transform function that accepts a single image can be passed for training. For example,
+                `torchvision.Compose` can be used to compose multiple transforms. If `None` - an inference mode is
+                assumed - and then a preset of inference-specific transforms will be applied to the images
+        Returns:
+            a PyTorch tensor of the processed images
+        """
+        image_size = image_size if image_size is not None else self.image_size
+        image_num_channels = image_num_channels if image_num_channels is not None else self.image_num_channels
+        image_mean = image_mean if image_mean is not None else self.image_mean
+        image_std = image_std if image_std is not None else self.image_std
+        size = (image_size, image_size)
+
+        if len(images) == 0:
+            return []
+
+        images = make_list_of_images(images)
+
+        if not valid_images(images):
+            raise ValueError(
+                "Invalid image type. Must be of type PIL.Image.Image, numpy.ndarray, "
+                "torch.Tensor, tf.Tensor or jax.ndarray."
+            )
+
+        # For training a user needs to pass their own set of transforms as a Callable.
+        # For reference this is what was used in the original IDEFICS training:
+        # transform = transforms.Compose([
+        #     convert_to_rgb,
+        #     transforms.RandomResizedCrop((size, size), scale=(0.9, 1.0), interpolation=transforms.InterpolationMode.BICUBIC),
+        #     transforms.ToTensor(),
+        #     transforms.Normalize(mean=image_mean, std=image_std),
+        # ])
+        if transform is not None:
+            if not is_torch_available():
+                raise ImportError("To pass in `transform` torch must be installed")
+            import torch
+
+            images = [transform(x) for x in images]
+            return torch.stack(images)
+
+        # for inference we do the exact transforms that were used to train IDEFICS
+        images = [convert_to_rgb(x) for x in images]
+        # further transforms expect numpy arrays
+        images = [to_numpy_array(x) for x in images]
+        images = [resize(x, size, resample=PILImageResampling.BICUBIC) for x in images]
+        images = [self.rescale(image=image, scale=1 / 255) for image in images]
+        images = [self.normalize(x, mean=image_mean, std=image_std) for x in images]
+        images = [to_channel_dimension_format(x, ChannelDimension.FIRST) for x in images]
+        # TODO: this converts to torch tensors - switch to convert_to_tensors once it becomes available
+        images = BatchFeature(data={"pixel_values": images}, tensor_type=TensorType.PYTORCH)["pixel_values"]
+
+        return images
+
+    def fetch_images(self, image_url_or_urls: Union[str, List[str]]):
+        """
+        Convert a single or a list of urls into the corresponding `PIL.Image` objects.
+        If a single url is passed, the return value will be a single object. If a list is passed a list of objects is
+        returned.
+        """
+        headers = {
+            "User-Agent": (
+                "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/114.0.0.0"
+                " Safari/537.36"
+            )
+        }
+        if isinstance(image_url_or_urls, list):
+            return [self.fetch_images(x) for x in image_url_or_urls]
+        elif isinstance(image_url_or_urls, str):
+            response = requests.get(image_url_or_urls, stream=True, headers=headers)
+            response.raise_for_status()
+            return Image.open(BytesIO(response.content))
+        else:
+            raise ValueError(f"only a single or a list of entries is supported but got type={type(image_url_or_urls)}")
+
+    def rescale(
+        self,
+        image: np.ndarray,
+        scale: float,
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs,
+    ) -> np.ndarray:
+        """
+        Rescale an image by a scale factor. image = image * scale.
+
+        Args:
+            image (`np.ndarray`):
+                Image to rescale.
+            scale (`float`):
+                The scaling factor to rescale pixel values by.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The channel dimension format for the output image. If unset, the channel dimension format of the input
+                image is used. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format for the input image. If unset, the channel dimension format is inferred
+                from the input image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+
+        Returns:
+            `np.ndarray`: The rescaled image.
+        """
+        # return rescale(image, scale=scale, data_format=data_format, input_data_format=input_data_format, **kwargs)
+        # requires 4.32
+        return rescale(image, scale=scale, data_format=data_format, **kwargs)
+
+    def normalize(
+        self,
+        image: np.ndarray,
+        mean: Union[float, Iterable[float]],
+        std: Union[float, Iterable[float]],
+        data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        **kwargs,
+    ) -> np.ndarray:
+        """
+        Normalize an image. image = (image - image_mean) / image_std.
+
+        Args:
+            image (`np.ndarray`):
+                Image to normalize.
+            mean (`float` or `Iterable[float]`):
+                Image mean to use for normalization.
+            std (`float` or `Iterable[float]`):
+                Image standard deviation to use for normalization.
+            data_format (`str` or `ChannelDimension`, *optional*):
+                The channel dimension format for the output image. If unset, the channel dimension format of the input
+                image is used. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format for the input image. If unset, the channel dimension format is inferred
+                from the input image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+
+        Returns:
+            `np.ndarray`: The normalized image.
+        """
+        # TODO 4.32
+        return normalize(
+            image, mean=mean, std=std, data_format=data_format, **kwargs
+        )
+
+
+import transformers
+transformers.IdeficsImageProcessor = IdeficsImageProcessor
diff --git a/server/text_generation_server/models/custom_modeling/idefics_modeling.py b/server/text_generation_server/models/custom_modeling/idefics_modeling.py
new file mode 100644
index 00000000..8b43ae4d
--- /dev/null
+++ b/server/text_generation_server/models/custom_modeling/idefics_modeling.py
@@ -0,0 +1,1327 @@
+# coding=utf-8
+# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
+#
+# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
+# and OPT implementations in this library. It has been modified from its
+# original forms to accommodate minor architectural differences compared
+# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" PyTorch Idefics model."""
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+
+from transformers import PreTrainedModel
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast, dataclass
+from transformers.modeling_utils import PretrainedConfig
+from transformers.utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+from text_generation_server.models.custom_modeling.idefics_config import IdeficsConfig
+from text_generation_server.models.custom_modeling.idefics_vision import IdeficsVisionTransformer
+from text_generation_server.models.custom_modeling.idefics_perceiver import IdeficsPerceiverResampler
+from text_generation_server.utils.layers import (
+    TensorParallelColumnLinear,
+    TensorParallelEmbedding,
+    TensorParallelRowLinear,
+    TensorParallelHead,
+    PositionRotaryEmbedding,
+    FastLinear,
+)
+import dropout_layer_norm
+
+@dataclass
+class BaseModelOutputWithPastImage(BaseModelOutputWithPast):
+    image_hidden_states: Optional[torch.FloatTensor] = None
+
+@dataclass
+class CausalLMOutputWithPastImage(CausalLMOutputWithPast):
+    image_hidden_states: Optional[torch.FloatTensor] = None
+
+
+# logger = logging.get_logger(__name__)
+
+# _CONFIG_FOR_DOC = "IdeficsConfig"
+
+# IDEFICS_PRETRAINED_MODEL_ARCHIVE_LIST = [
+#     "HuggingFaceM4/idefics-9b",
+#     "HuggingFaceM4/idefics-80b",
+#     # See all Idefics models at https://huggingface.co/models?filter=idefics
+# ]
+
+
+def expand_inputs_for_generation(
+    input_ids,
+    expand_size=1,
+    is_encoder_decoder=False,
+    attention_mask=None,
+    encoder_outputs=None,
+    **model_kwargs,
+):
+    expanded_return_idx = (
+        torch.arange(input_ids.shape[0]).view(-1, 1).repeat(1, expand_size).view(-1).to(input_ids.device)
+    )
+    input_ids = input_ids.index_select(0, expanded_return_idx)
+
+    if "token_type_ids" in model_kwargs:
+        token_type_ids = model_kwargs["token_type_ids"]
+        model_kwargs["token_type_ids"] = token_type_ids.index_select(0, expanded_return_idx)
+
+    if attention_mask is not None:
+        model_kwargs["attention_mask"] = attention_mask.index_select(0, expanded_return_idx)
+        model_kwargs["image_attention_mask"] = model_kwargs["image_attention_mask"].index_select(
+            0, expanded_return_idx
+        )
+        model_kwargs["pixel_values"] = model_kwargs["pixel_values"].index_select(0, expanded_return_idx)
+
+    if is_encoder_decoder:
+        if encoder_outputs is None:
+            raise ValueError("If `is_encoder_decoder` is True, make sure that `encoder_outputs` is defined.")
+        encoder_outputs["last_hidden_state"] = encoder_outputs.last_hidden_state.index_select(
+            0, expanded_return_idx.to(encoder_outputs.last_hidden_state.device)
+        )
+        model_kwargs["encoder_outputs"] = encoder_outputs
+    return input_ids, model_kwargs
+
+
+def update_model_kwargs_for_generation(outputs, model_kwargs, is_encoder_decoder=False):
+    # must have this key set to at least None
+    model_kwargs["past_key_values"] = model_kwargs.get("past_key_values", None)
+
+    # update past
+    if "past_key_values" in outputs:
+        model_kwargs["past"] = outputs.past_key_values
+    elif "mems" in outputs:
+        model_kwargs["past"] = outputs.mems
+    elif "past_buckets_states" in outputs:
+        model_kwargs["past"] = outputs.past_buckets_states
+    else:
+        model_kwargs["past"] = None
+
+    # update token_type_ids with last value
+    if "token_type_ids" in model_kwargs:
+        token_type_ids = model_kwargs["token_type_ids"]
+        model_kwargs["token_type_ids"] = torch.cat([token_type_ids, token_type_ids[:, -1].unsqueeze(-1)], dim=-1)
+
+    # update attention masks
+    if not is_encoder_decoder:
+        if "attention_mask" in model_kwargs:
+            attention_mask = model_kwargs["attention_mask"]
+            model_kwargs["attention_mask"] = torch.cat(
+                [attention_mask, attention_mask.new_ones((attention_mask.shape[0], 1))], dim=-1
+            )
+        if "image_attention_mask" in model_kwargs:
+            image_attention_mask = model_kwargs["image_attention_mask"]
+            last_mask = image_attention_mask[:, -1, :].unsqueeze(1)
+            model_kwargs["image_attention_mask"] = last_mask
+
+    return model_kwargs
+
+
+def prepare_inputs_for_generation(input_ids, past=None, **kwargs):
+    token_type_ids = kwargs.get("token_type_ids", None)
+    # only last token for inputs_ids if past is defined in kwargs
+    if past:
+        input_ids = input_ids[:, -1].unsqueeze(-1)
+        if token_type_ids is not None:
+            token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
+
+    attention_mask = kwargs.get("attention_mask", None)
+    position_ids = kwargs.get("position_ids", None)
+
+    if attention_mask is not None and position_ids is None:
+        # create position_ids on the fly for batch generation
+        position_ids = attention_mask.long().cumsum(-1) - 1
+        position_ids.masked_fill_(attention_mask == 0, 1)
+        if past:
+            position_ids = position_ids[:, -1].unsqueeze(-1)
+
+    pixel_values = kwargs.get("pixel_values", None)
+    image_attention_mask = kwargs.get("image_attention_mask", None)
+    # if pixel_values is None or image_attention_mask is None:
+    #     raise ValueError("pixel values and image attention mask cannot be None")
+
+    return {
+        "input_ids": input_ids,
+        "past_key_values": past,
+        "use_cache": kwargs.get("use_cache"),
+        "position_ids": position_ids,
+        "attention_mask": attention_mask,
+        "token_type_ids": token_type_ids,
+        "pixel_values": pixel_values,
+        "image_attention_mask": image_attention_mask,
+    }
+
+
+def freeze_model(model, module_exceptions=[]):
+    mapping = {
+        "LayerNorm": nn.LayerNorm,
+        "Linear": nn.Linear,
+        "Embedding": nn.Embedding,
+    }
+    module_exceptions_mapped = [mapping[m] for m in module_exceptions]
+    for module in model.modules():
+        if module_exceptions and any([isinstance(module, t) for t in module_exceptions_mapped]):
+            module.requires_grad_(True)  # Explicitely setting it to true to avoid any mistakes
+        else:
+            module.requires_grad_(False)
+    return model
+
+
+class IdeficsDecoupledPartialTPEmbedding(nn.Module):
+    def __init__(
+        self,
+        config,
+        weights,
+    ):
+        super().__init__()
+        self.num_embeddings = config.vocab_size
+        self.weight = TensorParallelEmbedding(prefix="model.embed_tokens", weights=weights)
+        self.additional_weight = nn.Parameter(weights.get_tensor(f"model.embed_tokens.additional_embedding.weight"))
+
+    def forward(self, input_ids):
+        # Clone so that we don't modify the original input_ids later on
+        input_ids = input_ids.clone()
+        additional_vocab_indices = torch.where(input_ids >= self.num_embeddings)
+        input_ids_additional_vocab = input_ids[additional_vocab_indices]
+        additional_embeddings = torch.nn.functional.embedding(input_ids_additional_vocab - self.num_embeddings, self.additional_weight)
+
+        # for successful lookup replace input_ids with 0, the results of these will be discarded anyway
+        input_ids[additional_vocab_indices] = 0
+        full_vector = self.weight(input_ids)
+
+        # overwrite the records with high indices
+        full_vector[additional_vocab_indices] = additional_embeddings
+
+        return full_vector
+
+
+class IdeficsDecoupledTensorParallelLinear(nn.Module):
+    # Derived from https://pytorch.org/docs/stable/_modules/torch/nn/modules/linear.html#Linear
+    """
+    Implements a decoupling of parameters to allow freezing (or not) a subset of the parameters. In practise, the
+    regular `weight` can be trained or frozen (i.e. `partially_freeze=True`), and if `out_additional_features` > 0,
+    then it will create `out_additional_features * in_features` additional parameters that are always trained. If
+    `out_additional_features=0`, then the module defaults back to the regular behavior of `nn.Linear`.
+    """
+
+    def __init__(
+        self,
+        config,
+        weights,
+    ) -> None:
+        super().__init__()
+        self.fc = TensorParallelHead.load(
+            config=config, prefix="lm_head", weights=weights
+        )
+        self.additional_fc = FastLinear.load(
+            config=config, prefix="lm_head.additional_fc", weights=weights, bias=False,
+        )
+
+    def forward(self, input: torch.Tensor) -> torch.Tensor:
+        output = self.fc(input)
+        additional_features = self.additional_fc(input)
+        output = torch.cat((output, additional_features), -1)
+
+        return output
+
+    def extra_repr(self) -> str:
+        """Overwriting `nn.Linear.extra_repr` to include new parameters."""
+        return "in_features={}, out_features={}, out_additional_features={}, bias={}, partially_freeze={}".format(
+            self.in_features,
+            self.out_features,
+            self.out_additional_features,
+            self.bias is not None,
+            self.partially_freeze,
+        )
+
+
+# Copied from transformers.models.bart.modeling_bart._make_causal_mask
+def _make_causal_mask(
+    input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
+):
+    """
+    Make causal mask used for bi-directional self-attention.
+    """
+    bsz, tgt_len = input_ids_shape
+    mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
+    mask_cond = torch.arange(mask.size(-1), device=device)
+    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
+    mask = mask.to(dtype)
+
+    if past_key_values_length > 0:
+        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
+    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
+
+
+def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+
+class IdeficsRMSNorm(nn.Module):
+    def __init__(self, prefix, weights, eps=1e-6):
+        """
+        LlamaRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+
+        weight = weights.get_tensor(f"{prefix}.weight")
+        self.weight = nn.Parameter(weight)
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states, residual=None):
+        if hidden_states.shape[-1] > 8192:
+            if residual is not None:
+                hidden_states += residual
+            residual = hidden_states
+
+            hidden_states = hidden_states.to(torch.float32)
+            variance = hidden_states.pow(2).mean(-1, keepdim=True)
+            hidden_states = hidden_states * torch.rsqrt(
+                variance + self.variance_epsilon
+            )
+
+            # convert into half-precision if necessary
+            if self.weight.dtype in [torch.float16, torch.bfloat16]:
+                hidden_states = hidden_states.to(self.weight.dtype)
+
+            return self.weight * hidden_states
+        else:
+            # faster post attention rms norm
+            unwrap = False
+            if len(hidden_states.shape) > 2:
+                unwrap = True
+                shape = hidden_states.shape
+                hidden_states = hidden_states.reshape(-1, shape[-1])
+
+            normed_hidden_states, res, *rest = dropout_layer_norm.dropout_add_ln_fwd(
+                hidden_states,
+                residual,
+                self.weight,
+                None,
+                None,
+                None,
+                None,
+                None,
+                0.0,
+                self.variance_epsilon,
+                1.0,
+                0,
+                None,
+                False,
+                True,  # Activate RMSNorm
+            )
+            if res is None:
+                res = hidden_states
+
+            if unwrap:
+                normed_hidden_states = normed_hidden_states.view(*shape)
+
+
+            return normed_hidden_states
+
+
+# this was adapted from LlamaMLP
+class IdeficsMLP(nn.Module):
+    def __init__(
+        self,
+        config,
+        prefix,
+        weights,
+    ):
+        super().__init__()
+        self.gate_up_proj = TensorParallelColumnLinear.load_multi(
+            config,
+            prefixes=[f"{prefix}.gate_proj", f"{prefix}.up_proj"],
+            weights=weights,
+            dim=0,
+            bias=False,
+        )
+        self.down_proj = TensorParallelRowLinear.load(
+            config, prefix=f"{prefix}.down_proj", weights=weights, bias=False,
+        )
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, hidden_states):
+        gate_up_states = self.gate_up_proj(hidden_states)
+        shape = gate_up_states.shape
+        gate_up_states = gate_up_states.view(*shape[:-1], 2, shape[-1] // 2)
+        return self.down_proj(self.act_fn(gate_up_states[:, :, 0]) * gate_up_states[:, :, 1])
+
+
+# this was adapted from LlamaAttention
+class IdeficsAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(
+        self,
+        config,
+        prefix,
+        weights,
+        qk_layer_norms: bool = False,
+        is_cross_attention: bool = False,
+    ):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.dropout = config.dropout
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+
+        self.is_cross_attention = is_cross_attention
+
+        # if not hasattr(nn.functional, "scaled_dot_product_attention"):
+        #     raise ValueError("this model requires pytorch 2.0 or higher")
+
+        process_group = weights.process_group
+        if self.num_heads % weights.process_group.size() != 0:
+            raise ValueError(
+                f"`num_heads` must be divisible by `num_shards` (got `num_heads`: {self.num_heads} "
+                f"and `num_shards`: {weights.process_group.size()}"
+            )
+        self.num_heads //= weights.process_group.size()
+
+        if self.is_cross_attention:
+            # kv_input_dim = (
+            #     self.hidden_size if not hasattr(config.vision_config, "embed_dim") else config.vision_config.embed_dim
+            # )
+            self.q_proj = TensorParallelColumnLinear.load(
+                config, prefix=f"{prefix}.q_proj", weights=weights, bias=False
+            )
+            self.k_proj = TensorParallelColumnLinear.load(
+                config, prefix=f"{prefix}.k_proj", weights=weights, bias=False
+            )
+            self.v_proj = TensorParallelColumnLinear.load(
+                config, prefix=f"{prefix}.v_proj", weights=weights, bias=False
+            )
+        else:
+            self.qkv = TensorParallelColumnLinear.load_multi(
+                config,
+                prefixes=[f"{prefix}.q_proj", f"{prefix}.k_proj", f"{prefix}.v_proj"],
+                dim=0,
+                weights=weights,
+                bias=False,
+            )
+        self.o_proj = TensorParallelRowLinear.load(
+            config, prefix=f"{prefix}.o_proj", weights=weights, bias=False
+        )
+        self.rotary_emb = PositionRotaryEmbedding.static(
+            config=config, dim=self.head_dim, base=10000.0, device=weights.device
+        )
+        self.qk_layer_norms = qk_layer_norms
+        if self.qk_layer_norms:
+            self.q_layer_norm = IdeficsRMSNorm(
+            prefix=f"{prefix}.q_layer_norm", weights=weights, eps=config.rms_norm_eps
+        )
+            self.k_layer_norm = IdeficsRMSNorm(
+            prefix=f"{prefix}.q_layer_norm", weights=weights, eps=config.rms_norm_eps
+        )
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        key_value_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        # if key_value_states are provided this layer is used as a cross-attention layer
+        is_cross_attention = self.is_cross_attention or key_value_states is not None
+
+        bsz, q_len, _ = hidden_states.size()
+
+        if is_cross_attention:
+            query_states = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim)# .transpose(1, 2)
+            query_states = query_states.transpose(1, 2)
+            _, kv_len, _ = key_value_states.size()  # Note that, in this case, `kv_len` == `kv_seq_len`
+            key_states = self.k_proj(key_value_states).view(bsz, kv_len, self.num_heads, self.head_dim).transpose(1, 2)
+            value_states = (
+                self.v_proj(key_value_states).view(bsz, kv_len, self.num_heads, self.head_dim).transpose(1, 2)
+            )
+        else:
+            qkv = self.qkv(hidden_states)
+            query_states, key_states, value_states = qkv.split(self.num_heads * self.head_dim, dim=2)
+
+            query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim)# .transpose(1, 2)
+            key_states = key_states.view(bsz, q_len, self.num_heads, self.head_dim)# . transpose(1, 2)
+            value_states = value_states.view(bsz, q_len, self.num_heads, self.head_dim)# .transpose(1, 2)
+            kv_seq_len = q_len
+            if past_key_value is not None:
+                kv_seq_len += past_key_value[0].shape[-2]
+            max_s = max(kv_seq_len, q_len)
+            cos, sin = self.rotary_emb.get_cos_sin(
+                position_ids.view(-1), max_s, hidden_states.dtype
+            )
+
+            shape = query_states.shape
+            query_states = self.rotary_emb(query_states.view(-1, *shape[2:]), cos, sin).view(shape)
+
+            shape = key_states.shape
+            key_states = self.rotary_emb(key_states.reshape(-1, *shape[2:]), cos, sin).view(shape)
+
+            query_states = query_states.transpose(1, 2)
+            key_states = key_states.transpose(1, 2)
+            value_states = value_states.transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+        # [bsz, nh, t, hd]
+
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        if self.qk_layer_norms:
+            query_states = self.q_layer_norm(query_states)
+            key_states = self.k_layer_norm(key_states)
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+
+        attn_output = nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=attention_mask,
+            dropout_p=self.dropout,
+        )
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2)
+        attn_output = attn_output.reshape(bsz, q_len, -1)
+
+        attn_output = self.o_proj(attn_output)
+
+        attn_weights = None
+        if output_attentions:
+            logger.warning_once(
+                "attn_weights are not extracted in scaled_dot_product_attention. The model returns None instead"
+            )
+
+        return attn_output, attn_weights, past_key_value
+
+
+# this was adapted from LlamaDecoderLayer
+class IdeficsDecoderLayer(nn.Module):
+    def __init__(self, layer_id: int, config: IdeficsConfig, weights):
+        super().__init__()
+        self.process_group = weights.process_group
+        self.hidden_size = config.hidden_size
+        prefix = f"model.layers.{layer_id}"
+        self.self_attn = IdeficsAttention(
+            config=config,
+            prefix=f"{prefix}.self_attn",
+            weights=weights,
+            qk_layer_norms=False,
+            is_cross_attention=False,
+        )
+        self.mlp = IdeficsMLP(
+            config=config,
+            prefix=f"{prefix}.mlp",
+            weights=weights,
+        )
+        self.input_layernorm = IdeficsRMSNorm(prefix=f"{prefix}.input_layernorm", weights=weights, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = IdeficsRMSNorm(prefix=f"{prefix}.post_attention_layernorm", weights=weights, eps=config.rms_norm_eps)
+        self.dropout = config.dropout
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        # hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        # hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+class IdeficsGatedCrossAttentionLayer(nn.Module):
+    def __init__(self, layer_id, config: IdeficsConfig, weights):
+        super().__init__()
+        self.process_group = weights.process_group
+        self.hidden_size = config.hidden_size
+        prefix = f"model.gated_cross_attn_layers.{layer_id}"
+        self.cross_attn = IdeficsAttention(
+            config=config,
+            prefix=f"{prefix}.cross_attn",
+            weights=weights,
+            qk_layer_norms=True,
+            is_cross_attention=True,
+        )
+        self.mlp = IdeficsMLP(
+            config=config,
+            prefix=f"{prefix}.mlp",
+            weights=weights,
+        )
+        self.input_layernorm = IdeficsRMSNorm(prefix=f"{prefix}.input_layernorm", weights=weights, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = IdeficsRMSNorm(prefix=f"{prefix}.post_attention_layernorm", weights=weights, eps=config.rms_norm_eps)
+        self.config = config.dropout
+
+        self.act_cross_attn = nn.Tanh()
+        self.act_dense = nn.Tanh()
+
+        self.alpha_cross_attn = nn.Parameter(weights.get_tensor(f"{prefix}.alpha_cross_attn"))
+        self.alpha_dense = nn.Parameter(weights.get_tensor(f"{prefix}.alpha_dense"))
+
+        if not (hasattr(self, "alpha_cross_attn") and hasattr(self, "alpha_dense")):
+            raise ValueError("Alpha parameters not initialized correctly!")
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        image_hidden_states: Optional[torch.Tensor] = None,
+        image_attention_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        no_images: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            no_images (`bool`, *optional*, defaults to `False`): If `True` the vision part is ignored
+        """
+        if image_hidden_states is None:
+            raise ValueError(
+                "`image_hidden_states` is required for Idefics cross attention module which are visual features to be"
+                " conditioned on."
+            )
+
+        if past_key_value is not None:
+            raise NotImplementedError("Past key value states are not implemented for Idefics cross attention module.")
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.cross_attn(
+            hidden_states=hidden_states,
+            key_value_states=image_hidden_states,
+            attention_mask=image_attention_mask,
+            output_attentions=output_attentions,
+        )
+        # hidden_states = nn.functional.dropout(hidden_states, p=self.config, training=self.training)
+        # when there are no images the model is used in pure language mode
+        gate = 0 if no_images else 1
+        hidden_states = residual + gate * self.act_cross_attn(self.alpha_cross_attn) * hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        # hidden_states = nn.functional.dropout(hidden_states, p=self.config, training=self.training)
+        hidden_states = residual + self.act_dense(self.alpha_dense) * hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+LLAMA_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`IdeficsConfig`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+# @add_start_docstrings(
+#     "The bare LLaMA Model outputting raw hidden-states without any specific head on top.",
+#     LLAMA_START_DOCSTRING,
+# )
+class IdeficsPreTrainedModel(PreTrainedModel):
+    config_class = IdeficsConfig
+    # base_model_prefix = "model"
+    # supports_gradient_checkpointing = True
+    # _no_split_modules = ["IdeficsDecoderLayer", "IdeficsGatedCrossAttentionLayer"]
+
+    # def _init_weights(self, module):
+    #     # important: this ported version of Idefics isn't meant for training from scratch - only
+    #     # inference and fine-tuning - so the proper init weights code has been removed - the m4 code
+    #     # base should be used for training from scratch and it contains the correct code.
+    #     std = self.config.initializer_range
+    #     if isinstance(module, nn.Linear):
+    #         module.weight.data.normal_(mean=0.0, std=std)
+    #         if module.bias is not None:
+    #             module.bias.data.zero_()
+    #     elif isinstance(module, nn.Embedding):
+    #         module.weight.data.normal_(mean=0.0, std=std)
+    #         if module.padding_idx is not None:
+    #             module.weight.data[module.padding_idx].zero_()
+
+    # def _set_gradient_checkpointing(self, module, value=False):
+    #     if isinstance(module, IdeficsModel):
+    #         module.gradient_checkpointing = value
+
+
+# LLAMA_INPUTS_DOCSTRING = r"""
+#     Args:
+#         input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+#             Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+#             it.
+
+#             Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+#             [`PreTrainedTokenizer.__call__`] for details.
+
+#             [What are input IDs?](../glossary#input-ids)
+#         attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+#             Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+#             - 1 for tokens that are **not masked**,
+#             - 0 for tokens that are **masked**.
+
+#             [What are attention masks?](../glossary#attention-mask)
+
+#             Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+#             [`PreTrainedTokenizer.__call__`] for details.
+
+#             If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+#             `past_key_values`).
+
+#             If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+#             and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+#             information on the default strategy.
+
+#             - 1 indicates the head is **not masked**,
+#             - 0 indicates the head is **masked**.
+#         position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+#             Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+#             config.n_positions - 1]`. [What are position IDs?](../glossary#position-ids)
+#         past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+#             Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+#             `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+#             `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+#             Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+#             blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+#             If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+#             don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+#             `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+#         inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+#             Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+#             is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+#             model's internal embedding lookup matrix.
+#         use_cache (`bool`, *optional*):
+#             If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+#             `past_key_values`).
+#         output_attentions (`bool`, *optional*):
+#             Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+#             tensors for more detail.
+#         output_hidden_states (`bool`, *optional*):
+#             Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+#             more detail.
+#         return_dict (`bool`, *optional*):
+#             Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+# """
+
+
+# @add_start_docstrings(
+#     "The bare LLaMA Model outputting raw hidden-states without any specific head on top.",
+#     LLAMA_START_DOCSTRING,
+# )
+class IdeficsModel(IdeficsPreTrainedModel):
+    # """
+    # Transformer decoder consisting of `config.num_hidden_layers` layers. Each layer is a [`IdeficsDecoderLayer`]
+
+    # Args:
+    #     config: IdeficsConfig
+    # """
+
+    def __init__(self, config: IdeficsConfig, weights):
+        super().__init__(config)
+        self.config = config
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = IdeficsDecoupledPartialTPEmbedding(
+            config=config,
+            weights=weights,
+        )
+
+        self.image_size = config.vision_config.image_size
+        self.vision_config = config.vision_config
+        self.vision_model = IdeficsVisionTransformer(
+            prefix="model.vision_model",
+            config=config.vision_config,
+            weights=weights,
+        )
+
+        # Perceiver Resampler
+        if config.use_resampler:
+            perceiver_config = config.perceiver_config
+            self.perceiver_resampler = IdeficsPerceiverResampler(
+                prefix=f"model.perceiver_resampler",
+                config=config,
+                embed_dim=config.vision_config.embed_dim,
+                depth=perceiver_config.resampler_depth,
+                n_heads=perceiver_config.resampler_n_heads,
+                head_dim=perceiver_config.resampler_head_dim,
+                n_latents=perceiver_config.resampler_n_latents,
+                weights=weights,
+            )
+
+        self.layers = nn.ModuleList(
+            [
+                IdeficsDecoderLayer(layer_id, config, weights)
+                for layer_id in range(config.num_hidden_layers)
+            ]
+        )
+
+        self.cross_layer_interval = config.cross_layer_interval
+        num_cross_layers = config.num_hidden_layers // self.cross_layer_interval
+        self.gated_cross_attn_layers = nn.ModuleList(
+            [
+                IdeficsGatedCrossAttentionLayer(layer_id, config, weights)
+                for layer_id in range(num_cross_layers)]
+        )
+        # self.gradient_checkpointing = False
+
+        self.norm = IdeficsRMSNorm(prefix=f"model.norm", weights=weights, eps=config.rms_norm_eps)
+
+        # self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        # self.post_init()
+
+        # self.freeze_relevant_params(config)
+
+    # def freeze_relevant_params(self, config=None):
+    #     if config is None:
+    #         config = self.config
+
+    #     if config.freeze_text_layers:
+    #         self.freeze_text_layers(config.freeze_text_module_exceptions)
+
+    #     if config.freeze_vision_layers:
+    #         freeze_model(self.vision_model, module_exceptions=config.freeze_vision_module_exceptions)
+
+    # def freeze_text_layers(self, module_exceptions=[]):
+    #     for module in [self.layers, self.norm]:
+    #         freeze_model(module, module_exceptions=module_exceptions)
+
+    # def freeze_vision_layers(self, module_exceptions=[]):
+    #     freeze_model(self.vision_model, module_exceptions=module_exceptions)
+
+    # def get_input_embeddings(self):
+    #     return self.embed_tokens
+
+    # def set_input_embeddings(self, value):
+    #     self.embed_tokens = value
+
+    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
+    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
+        # create causal mask
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        combined_attention_mask = None
+        if input_shape[-1] > 1:
+            combined_attention_mask = _make_causal_mask(
+                input_shape,
+                inputs_embeds.dtype,
+                device=inputs_embeds.device,
+                past_key_values_length=past_key_values_length,
+            )
+
+        if attention_mask is not None:
+            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
+                inputs_embeds.device
+            )
+            combined_attention_mask = (
+                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
+            )
+
+        return combined_attention_mask
+
+    # @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        pixel_values: Optional[torch.FloatTensor] = None,
+        image_hidden_states: Optional[torch.FloatTensor] = None,
+        image_embeddings: Optional[torch.FloatTensor] = None,
+        image_attention_mask: Optional[torch.Tensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPastImage]:
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
+
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+
+        if past_key_values is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+        elif position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        no_images = False
+
+        if image_hidden_states is None:
+            if pixel_values is None and image_embeddings is None:
+                raise ValueError("Either pixel_values and image_embeddings have to be not-None.")
+
+            elif pixel_values is not None and image_embeddings is not None:
+                raise ValueError("You cannot specify both pixel_values and image_embeddings at the same time")
+
+            elif pixel_values is not None:
+                no_images = len(torch.nonzero(pixel_values)) == 0
+                pixel_values = pixel_values.to(dtype=self.dtype, device=device)  # fp16 compatibility
+                batch_size, num_images = pixel_values.shape[:2]
+                pixel_values = pixel_values.contiguous().view(batch_size * num_images, *pixel_values.shape[2:])
+
+                # Get sequence from the vision encoder
+                image_hidden_states = self.vision_model(pixel_values=pixel_values).last_hidden_state
+
+            elif image_embeddings is not None:
+                batch_size, num_images, image_seq_len, image_hidden_size = image_embeddings.size()
+                image_hidden_states = image_embeddings.to(dtype=self.dtype, device=input_ids.device)
+                image_hidden_states = image_hidden_states.view(batch_size * num_images, image_seq_len, image_hidden_size)
+
+            if self.config.use_resampler:
+                image_hidden_states = self.perceiver_resampler(image_hidden_states)
+            image_seq_len, image_hidden_size = image_hidden_states.size(1), image_hidden_states.size(2)
+            image_hidden_states = image_hidden_states.view(batch_size, num_images * image_seq_len, image_hidden_size)
+        else:
+            no_images = False
+            num_images = pixel_values.shape[1]
+            image_seq_len = image_hidden_states.shape[1] // num_images
+
+        # # Hack to use the model in full language modeling mode
+        # image_attention_mask = torch.zeros(batch_size, seq_length, 1, dtype=torch.long, device=image_hidden_states.device)
+        # Make image_attention_mask compatible with hidden states
+        text_seq_len = image_attention_mask.size(1)
+        image_attention_mask = image_attention_mask.unsqueeze(-1)
+        image_attention_mask = image_attention_mask.repeat(1, 1, 1, image_seq_len)
+        image_attention_mask = image_attention_mask.view(batch_size, text_seq_len, num_images * image_seq_len)
+        image_batch_size, image_sequence_length, _ = image_hidden_states.size()
+        image_hidden_shape = (image_batch_size, image_sequence_length)
+        if image_attention_mask is None:
+            image_attention_mask = torch.ones(image_hidden_shape, device=device)
+        image_attention_mask = self.invert_attention_mask(image_attention_mask)
+
+        # if list(image_attention_mask.shape) != [4, 1, 1024, 64]:
+        #     raise ValueError(f"Image hidden_states {image_hidden_states.shape} - mask {image_attention_mask.shape} {num_images} {image_seq_len} {text_seq_len}")
+
+
+        # if image_hidden_states is not None:
+        # else:
+        #     image_attention_mask = None
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        # embed positions
+        if attention_mask is None:
+            attention_mask = torch.ones(
+                (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
+            )
+        attention_mask = self._prepare_decoder_attention_mask(
+            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
+        )
+
+        hidden_states = inputs_embeds
+
+        # if self.gradient_checkpointing and self.training:
+        #     if use_cache:
+        #         logger.warning_once(
+        #             "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+        #         )
+        #         use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+            def vblock(
+                main_block,
+                hidden_states,
+                attention_mask,
+                position_ids,
+                past_key_value,
+                image_hidden_states,
+                image_attention_mask,
+                output_attentions,
+                use_cache,
+                no_images,
+                layer_idx,
+                cross_layer_interval,
+                gated_cross_attn_layers,
+            ):
+                # TODO(ls): Add cross attention values to respective lists
+                if layer_idx % cross_layer_interval == 0:
+                    xblock = gated_cross_attn_layers[layer_idx // cross_layer_interval]
+                    outputs = xblock(
+                        hidden_states,
+                        attention_mask=attention_mask,
+                        image_hidden_states=image_hidden_states,
+                        image_attention_mask=image_attention_mask,
+                        output_attentions=output_attentions,
+                        use_cache=use_cache,
+                        past_key_value=None,  # not implemented
+                        no_images=no_images,
+                    )
+                    hidden_states = outputs[0]
+
+                layer_outputs = main_block(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+                return layer_outputs
+
+            # if self.gradient_checkpointing and self.training:
+            #     past_key_value = None
+            #     if use_cache:
+            #         logger.warning_once(
+            #             "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+            #         )
+            #         use_cache = False
+
+            #     layer_outputs = torch.utils.checkpoint.checkpoint(
+            #         vblock,
+            #         decoder_layer,
+            #         hidden_states,
+            #         attention_mask,
+            #         position_ids,
+            #         past_key_value,
+            #         image_hidden_states,
+            #         image_attention_mask,
+            #         output_attentions,
+            #         use_cache,
+            #         no_images,
+            #         idx,
+            #         self.cross_layer_interval,
+            #         self.gated_cross_attn_layers,
+            #     )
+            # else:
+            layer_outputs = vblock(
+                decoder_layer,
+                hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                image_hidden_states=image_hidden_states,
+                image_attention_mask=image_attention_mask,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+                no_images=no_images,
+                layer_idx=idx,
+                cross_layer_interval=self.cross_layer_interval,
+                gated_cross_attn_layers=self.gated_cross_attn_layers,
+            )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPastImage(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            image_hidden_states=image_hidden_states,
+        )
+
+
+class IdeficsForVisionText2Text(IdeficsPreTrainedModel):
+    def __init__(
+        self,
+        config,
+        weights,
+    ):
+        super().__init__(config)
+        self.model = IdeficsModel(
+            config=config,
+            weights=weights,
+        )
+
+        self.lm_head = IdeficsDecoupledTensorParallelLinear(
+            config=config,
+            weights=weights,
+        )
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        pixel_values: Optional[torch.FloatTensor] = None,
+        image_embeddings: Optional[torch.FloatTensor] = None,
+        image_hidden_states:  Optional[torch.FloatTensor] = None,
+        image_attention_mask: Optional[torch.Tensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, CausalLMOutputWithPastImage]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, LlamaForCausalLM
+
+        >>> model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
+        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
+
+        >>> prompt = "Hey, are you consciours? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            pixel_values=pixel_values,
+            image_embeddings=image_embeddings,
+            image_hidden_states=image_hidden_states,
+            image_attention_mask=image_attention_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+
+        loss = None
+
+        return CausalLMOutputWithPastImage(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            image_hidden_states=outputs.image_hidden_states
+        )
+
+    def prepare_inputs_for_generation(self, input_ids, past=None, **kwargs):
+        inputs = prepare_inputs_for_generation(input_ids, past=past, **kwargs)
+        unwanted_kwargs = ["token_type_ids"]
+        for kwarg in unwanted_kwargs:
+            inputs.pop(kwarg, None)
+        return inputs
+
+    @staticmethod
+    def _expand_inputs_for_generation(
+        *args,
+        **model_kwargs,
+    ):
+        return expand_inputs_for_generation(*args, **model_kwargs)
+
+    @staticmethod
+    def _update_model_kwargs_for_generation(outputs, model_kwargs, is_encoder_decoder=False):
+        return update_model_kwargs_for_generation(outputs, model_kwargs, is_encoder_decoder=is_encoder_decoder)
+
+    @staticmethod
+    def _reorder_cache(past, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past
diff --git a/server/text_generation_server/models/custom_modeling/idefics_perceiver.py b/server/text_generation_server/models/custom_modeling/idefics_perceiver.py
new file mode 100644
index 00000000..def78390
--- /dev/null
+++ b/server/text_generation_server/models/custom_modeling/idefics_perceiver.py
@@ -0,0 +1,246 @@
+# This code was adapted from https://github.com/lucidrains/flamingo-pytorch licensed under the MIT License.
+#
+# MIT License
+#
+# Copyright (c) 2020  The Google AI Language Team Authors, The HuggingFace Inc. team and github/lonePatient
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
+
+"""
+
+Generic interface to various configurations of the Perceiver Resampler, that simply takes in a series of (potentially
+time-indexed) contextual embeddings, and "resamples" (compresses) them down to a pre-specified number of latents! Note
+that the Perceiver in general resamples based solely off the *long-range* context; there's a nice opportunity here to
+prime the Perceiver Resampler with say a single layer's worth of language embeddings (the target domain), and use that
+to softly "retrieve & compress" what we need --> this would be a novel contribution we should explore.
+
+References:
+    - DeepMind's Flamingo: https://www.deepmind.com/blog/tackling-multiple-tasks-with-a-single-visual-language-model
+    - Code borrowed w/ love from: https://github.com/lucidrains/flamingo-pytorch
+
+"""
+from typing import Optional, Tuple
+
+import torch
+import torch.nn as nn
+
+from text_generation_server.utils.layers import (
+    TensorParallelColumnLinear,
+    TensorParallelRowLinear,
+)
+
+EPS=1e-5
+
+class IdeficsPerceiverResampler(nn.Module):
+    def __init__(
+        self,
+        prefix,
+        config,
+        embed_dim: int,
+        depth: int,
+        n_heads: int,
+        head_dim: int,
+        n_latents: int,
+        weights,
+    ) -> None:
+        """
+        Instantiates a Perceiver Resampler that operates over a sequence of embeddings (say from a ResNet or ViT or
+        MAE) of a given dimension, performs `depth` blocks of cross-attention with a fixed `n_latents` inputs, then
+        returns a Tensor of shape [bsz, n_latents, embed_dim]. :param embed_dim: Dimensionality of embeddings being fed
+        to the Perceiver Resampler (also dimensionality of latent embeddings *returned* by the Perceiver Resampler.
+        Could be e.g., VIT embed_dim, ResNet pool dim, and so on.
+
+        Args:
+            config (`IdeficsConfig`): config object
+            embed_dim (`int`): The size of each embedding vector
+            depth (`int`): Depth of the Perceiver Resampler (Transformer w/ cross attention). Should be shallow (< 3).
+            n_heads (`int`): Number of heads in each Transformer block (for multi-headed self-attention).
+            head_dim (`int`): Dimensionality of each head projection in the Transformer block.
+            n_latents (`int`):
+                Number of latent embeddings to resample ("compress") the input sequence to (usually < 128).
+
+        """
+        super().__init__()
+        self.embed_dim, self.n_heads, self.head_dim, self.n_latents = embed_dim, n_heads, head_dim, n_latents
+        self.qk_layer_norms = config.perceiver_config.qk_layer_norms_perceiver
+
+        # Create Latents for Perceiver
+        self.latents = nn.Parameter(weights.get_tensor(f"{prefix}.latents"))
+
+        self.intermediate_dim = (
+            self.embed_dim * 4
+            if not hasattr(config.vision_config, "embed_dim")
+            else config.vision_config.embed_dim * 4
+        )
+        # Create Transformer Blocks
+        self.blocks = nn.ModuleList(
+            [
+                nn.ModuleList(
+                    [
+                        IdeficsPerceiverAttention(
+                            prefix=f"{prefix}.blocks.{layer_id}.0",
+                            config=config,
+                            embed_dim=self.embed_dim,
+                            n_heads=self.n_heads,
+                            head_dim=self.head_dim,
+                            qk_layer_norms=self.qk_layer_norms,
+                            weights=weights,
+                        ),
+                        IdeficsMLP(
+                            prefix=f"{prefix}.blocks.{layer_id}.1",
+                            intermediate_size=self.intermediate_dim,
+                            config=config,
+                            weights=weights
+                        ),
+                    ]
+                )
+                for layer_id in range(depth)
+            ]
+        )
+        self.layer_norm = nn.LayerNorm.load(prefix=f"{prefix}.layer_norm", weights=weights, eps=EPS)
+
+    def forward(self, context: torch.Tensor) -> torch.Tensor:
+        """Resample arbitrary length context & *compress* down to self.n_latents latent embeddings"""
+        # einsum.repeat(self.latents, "seq embed -> bsz seq embed", bsz=context.shape[0])
+        latents = self.latents.repeat(context.shape[0], 1, 1)
+
+        # Feed through Perceiver Attention blocks...
+        for attn, ff in self.blocks:
+            latents = attn(context, latents) + latents
+            latents = ff(latents) + latents
+
+        return self.layer_norm(latents)
+
+
+class IdeficsPerceiverAttention(nn.Module):
+    def __init__(self,
+            prefix,
+            config,
+            embed_dim: int,
+            n_heads: int,
+            head_dim: int,
+            qk_layer_norms: bool,
+            weights
+        ) -> None:
+        """Perceiver Cross-Attention Module --> let long-form inputs be `context`, resampled embeddings be `latents`"""
+        super().__init__()
+        self.embed_dim, self.n_heads, self.head_dim = embed_dim, n_heads, head_dim
+        self.qk_layer_norms = qk_layer_norms
+        # Normalization & Scaling
+        self.context_layer_norm = nn.LayerNorm.load(prefix=f"{prefix}.context_layer_norm", weights=weights, eps=EPS)
+        self.latents_layer_norm = nn.LayerNorm.load(prefix=f"{prefix}.latents_layer_norm", weights=weights, eps=EPS)
+        if self.qk_layer_norms:
+            self.q_layer_norm = nn.LayerNorm.load(prefix=f"{prefix}.q_layer_norm", weights=weights, eps=EPS)
+            self.k_layer_norm = nn.LayerNorm.load(prefix=f"{prefix}.k_layer_norm", weights=weights, eps=EPS)
+
+        self.qk_scale = self.head_dim**-0.5
+
+        process_group = weights.process_group
+        if n_heads % weights.process_group.size() != 0:
+            raise ValueError(
+                f"`num_heads` must be divisible by `num_shards` (got `num_heads`: {n_heads} "
+                f"and `num_shards`: {weights.process_group.size()}"
+            )
+        self.n_heads //= weights.process_group.size()
+
+        # Q, K, V Projection (no bias -- detail from Perceiver/Flamingo Papers).
+        self.q_proj = TensorParallelColumnLinear.load(
+            config=config, prefix=f"{prefix}.q_proj", weights=weights, bias=False
+        )
+        self.k_proj =  TensorParallelColumnLinear.load(
+            config=config, prefix=f"{prefix}.k_proj", weights=weights, bias=False
+        )
+        self.v_proj =  TensorParallelColumnLinear.load(
+            config=config, prefix=f"{prefix}.v_proj", weights=weights, bias=False
+        )
+
+        self.output_proj = TensorParallelRowLinear.load(
+            config=config, prefix=f"{prefix}.output_proj", weights=weights, bias=False
+        )
+
+    def forward(self, context: torch.Tensor, latents: torch.Tensor) -> torch.Tensor:
+        """
+        Runs Perceiver Self-Attention, with special (context, latents) appended along the `seq` dimension!
+
+        Args:
+            context (`torch.Tensor`):
+                Tensor of shape `[bsz, seq, embed_dim]` representing long-form context to resample.
+            latents (`torch.Tensor`):
+                Tensor of shape `[bsz, n_latents, embed_dim]` representing fixed length latents to compress to.
+
+        Returns:
+            `torch.Tensor`: Tensor of shape `[bsz, n_latents, embed_dim]` representing attention over latents w/ cross
+            from context.
+        """
+        context = self.context_layer_norm(context)
+        latents = self.latents_layer_norm(latents)
+        batch_size, seq_length, embed_dim = context.shape[:3]
+
+        # Query, Key, Value Projections --> Note that in Flamingo, latents are *concatenated* with context prior to attn!
+        #   Note: This results in queries w/ `seq = n_latents`, and keys, values with `seq = len(context) + n_latents`
+        q = self.q_proj(latents)
+        k = self.k_proj(torch.cat([context, latents], dim=-2))
+        v = self.v_proj(torch.cat([context, latents], dim=-2))
+
+        # Multiheaded Self-Attention w/ stable softmax (subtract per-row max -- `amax` -- before softmax call)
+        #   =>> `attn` should be a 2D matrix of shape [n_latents x (context + n_latents)]
+        # einsum.rearrange(x, "bsz seq (heads embed) -> bsz heads seq embed", heads=self.n_heads)
+        q, k, v = [x.reshape(batch_size, x.shape[1], self.n_heads, self.head_dim).transpose(1, 2) for x in (q, k, v)]
+
+        if self.qk_layer_norms:
+            q = self.q_layer_norm(q)
+            k = self.k_layer_norm(k)
+
+        scores = torch.einsum("... i d, ... j d -> ... i j", q * self.qk_scale, k)
+        stabilized_scores = scores - (scores.amax(dim=-1, keepdim=True).detach())
+        attn = stabilized_scores.softmax(dim=-1)
+
+        # Attend & project back to output...
+        resampled = torch.einsum("... i j, ... j d -> ... i d", attn, v)
+        # einsum.rearrange(resampled, "bsz heads seq embed -> bsz seq (heads embed)", heads=self.n_heads)
+        return self.output_proj(resampled.transpose(1, 2).flatten(-2))
+
+
+class IdeficsMLP(nn.Module):
+    def __init__(self,
+            prefix,
+            intermediate_size,
+            config,
+            weights,
+        ):
+        """Simple MLP block with intermediate_size and embedding size"""
+        super().__init__()
+        self.embed_dim = config.vision_config.embed_dim
+        self.ln = nn.LayerNorm.load(prefix=f"{prefix}.ln", weights=weights, eps=EPS)
+        self.fc = TensorParallelColumnLinear.load(
+            config=config, prefix=f"{prefix}.fc", weights=weights, bias=False,
+        )
+        self.act = nn.ReLU()
+        self.c_proj = TensorParallelRowLinear.load(
+            config=config, prefix=f"{prefix}.c_proj", weights=weights, bias=False,
+        )
+
+    def forward(self, hidden_states: Optional[Tuple[torch.FloatTensor]]) -> torch.FloatTensor:
+        hidden_states = self.ln(hidden_states)
+        hidden_states = self.fc(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.c_proj(hidden_states)
+
+        return hidden_states
diff --git a/server/text_generation_server/models/custom_modeling/idefics_processing.py b/server/text_generation_server/models/custom_modeling/idefics_processing.py
new file mode 100644
index 00000000..e24fc7bd
--- /dev/null
+++ b/server/text_generation_server/models/custom_modeling/idefics_processing.py
@@ -0,0 +1,413 @@
+# coding=utf-8
+# Copyright 2022 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Processor class for IDEFICS.
+"""
+
+from typing import Callable, List, Optional, Union
+from urllib.parse import urlparse
+
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.processing_utils import ProcessorMixin
+from transformers.tokenization_utils_base import BatchEncoding, PaddingStrategy, TextInput, TruncationStrategy
+from transformers.utils import TensorType, is_torch_available
+from text_generation_server.models.custom_modeling.idefics_image_processing import IdeficsImageProcessor
+
+
+if is_torch_available():
+    import torch
+
+
+IMAGE_TOKEN = "<image>"
+
+
+# copied from m4.training.packing
+def incremental_to_binary_attention_mask(incremental_mask, num_classes=-1):
+    # This function converts: [-1, 0, 1] => [[0, 0], [1, 0], [0, 1]]
+
+    # If any of images index are more than num_classes, set them to -1.
+    # Words after the max number of images allowed have been seen don't attend on anything
+    if num_classes != -1:
+        incremental_mask[incremental_mask >= num_classes] = -1
+
+    negatives = incremental_mask == -1
+    incremental_mask[negatives] = 0
+    attn_mask = torch.nn.functional.one_hot(incremental_mask, num_classes=num_classes)
+    attn_mask[negatives, :] = 0
+    return attn_mask
+
+
+# copied from m4.training.packing
+def image_attention_mask_for_packed_input_ids(input_ids, tokenizer):
+    image_attention_mask = torch.full_like(input_ids, fill_value=-1)
+    next_image_attention_mask = torch.full_like(input_ids, fill_value=-1)
+    image_token_id = tokenizer.convert_tokens_to_ids(IMAGE_TOKEN)
+    eod_token_id = tokenizer.eos_token_id
+    for batch_idx in range(input_ids.size(0)):
+        count = -1
+        seen_eod = False
+        for idx, token_id in enumerate(input_ids[batch_idx]):
+            if token_id == image_token_id:
+                count += 1
+                image_attention_mask[batch_idx][idx] = count
+                seen_eod = False
+            else:
+                image_attention_mask[batch_idx][idx] = count
+
+            if seen_eod:
+                image_attention_mask[batch_idx][idx] = -1
+
+            if token_id == eod_token_id:
+                seen_eod = True
+
+    for batch_idx in range(input_ids.size(0)):
+        count = -1
+        seen_eod = False
+        for idx in range(input_ids[batch_idx].size(0) - 1, -1, -1):
+            token_id = input_ids[batch_idx][idx]
+            if token_id == image_token_id:
+                count += 1
+                next_image_attention_mask[batch_idx][idx] = count
+                seen_eod = False
+            else:
+                next_image_attention_mask[batch_idx][idx] = count
+
+            if token_id == eod_token_id:
+                seen_eod = True
+
+            if seen_eod:
+                next_image_attention_mask[batch_idx][idx] = -1
+
+        non_negative_indices = next_image_attention_mask[batch_idx] != -1
+        next_image_attention_mask[batch_idx][non_negative_indices] -= count
+        next_image_attention_mask[batch_idx][non_negative_indices] *= -1
+
+    return image_attention_mask, next_image_attention_mask
+
+
+def is_url(string):
+    """Checks if the passed string contains a valid url and nothing else. e.g. if space is included it's immediately
+    invalidated the url"""
+    if " " in string:
+        return False
+    result = urlparse(string)
+    return all([result.scheme, result.netloc])
+
+
+class IdeficsProcessor(ProcessorMixin):
+    r"""
+    Constructs a IDEFICS processor which wraps a LLama tokenizer and IDEFICS image processor into a single processor.
+
+    [`IdeficsProcessor`] offers all the functionalities of [`IdeficsImageProcessor`] and [`LlamaTokenizerFast`]. See
+    the docstring of [`~IdeficsProcessor.__call__`] and [`~IdeficsProcessor.decode`] for more information.
+
+    Args:
+        image_processor (`IdeficsImageProcessor`):
+            An instance of [`IdeficsImageProcessor`]. The image processor is a required input.
+        tokenizer (`LlamaTokenizerFast`):
+            An instance of [`LlamaTokenizerFast`]. The tokenizer is a required input.
+        image_size (`int`, *optional*, defaults to 224): Image size (assuming a square image)
+    """
+    attributes = ["image_processor", "tokenizer"]
+    image_processor_class = "IdeficsImageProcessor"
+    tokenizer_class = "LlamaTokenizerFast"
+
+    def __init__(self, image_processor, tokenizer=None, image_size=224, add_end_of_utterance_token=None, **kwargs):
+        if image_processor is None:
+            raise ValueError("You need to specify an `image_processor`.")
+        if tokenizer is None:
+            raise ValueError("You need to specify a `tokenizer`.")
+
+        super().__init__(image_processor, tokenizer)
+        self.current_processor = self.image_processor
+        self.image_token_id = tokenizer.convert_tokens_to_ids(IMAGE_TOKEN)
+
+        self.default_image_dims = (
+            self.image_processor.image_num_channels,
+            self.image_processor.image_size,
+            self.image_processor.image_size,
+        )
+
+        self.tokenizer_was_trained_with_end_of_utterance_token = (
+            True
+            if "<end_of_utterance>" in self.tokenizer.special_tokens_map.get("additional_special_tokens", [])
+            else False
+        )
+
+    def __call__(
+        self,
+        prompts: Union[List[TextInput], List[List[TextInput]]],
+        padding: Union[bool, str, PaddingStrategy] = False,
+        truncation: Union[bool, str, TruncationStrategy] = None,
+        max_length: Optional[int] = None,
+        transform: Callable = None,
+        add_eos_token=False,
+        add_end_of_utterance_token=None,
+        debug=False,
+        return_tensors: Optional[Union[str, TensorType]] = TensorType.PYTORCH,
+    ) -> BatchEncoding:
+        """This method takes batched or non-batched prompts made of text and images and converts them into prompts that
+        the model was trained on and prepares the image pixel values for the model to process.
+
+        Args:
+            prompts (`Union[List[TextInput], [List[List[TextInput]]]]`):
+                either a single prompt or a batched list of prompts - see the detailed description immediately after
+                the end of the arguments doc section.
+            padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `False`):
+                Select a strategy to pad the returned sequences (according to the model's padding side and padding
+                index) among:
+                - `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
+                  sequence if provided).
+                - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum
+                  acceptable input length for the model if that argument is not provided.
+                - `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different
+                  lengths).
+            max_length (`int`, *optional*):
+                Maximum length of the returned list and optionally padding length (see above).
+            truncation (`bool`, *optional*):
+                Activates truncation to cut input sequences longer than `max_length` to `max_length`.
+            transform (`Callable`, *optional*):
+                A custom transform function that accepts a single image can be passed for training. For example,
+                `torchvision.Compose` can be used to compose multiple functions. If `None` a preset inference-specific
+                set of transforms will be applied to the images
+            add_eos_token (`bool`, *optional*, defaults to `False`):
+                Adds `eos_token` at the end of the final prompt if True`
+            add_end_of_utterance_token (`bool`, *optional*)
+                Whether to automatically add `<end_of_utterance>` after each prompt's text input (unless followed by an
+                image). If `None` the tokenizer will be checked instead and if this token is found in
+                `additional_special_tokens` then the value will be `True`.
+            debug (`bool`, *optional*, defaults to `False`):
+                `True` value will help debug prompt generation by dumping useful information
+            return_tensors (`str` or `TensorType`, *optional*, defaults to `TensorType.PYTORCH`):
+                The type of tensors to return. Can be one of:
+                    - `TensorType.PYTORCH` or `'pt'`: Return a batch of type `torch.Tensor`.
+
+        Returns:
+            a dict with entries: `input_ids`, `attention_mask`, `pixel_values`, `image_attention_mask` which can be
+            directly passed to `model.generate`
+
+        Detailed explanation:
+
+        Each entry in `prompts` is either a text to be passed as is or an image that will be processed.
+
+        An image can be either an image object (`PIL.Image`) or a url from which the image can be retrieved.
+
+        When the processor encounters an image it'll inject `<fake_token_around_image><image><fake_token_around_image>`
+        entry into the prompt.
+
+        Example:
+
+        ```python
+        checkpoint = "HuggingFaceM4/idefics-9b"
+        processor = AutoProcessor.from_pretrained(checkpoint)
+        url = "https://hips.hearstapps.com/hmg-prod/images/cute-photos-of-cats-in-grass-1593184777.jpg"
+        img = processor.image_processor.fetch_images([url])[0]
+
+        prompts = [
+            "User:",
+            img,
+            "Describe this image.\nAssistant: An image of two kittens in grass.\n",
+            "User:",
+            "https://hips.hearstapps.com/hmg-prod/images/dog-puns-1581708208.jpg",
+            "Describe this image.\nAssistant:",
+        ]
+
+        inputs = processor(prompts, return_tensors="pt")
+        generated_ids = model.generate(**inputs, max_length=100)
+        generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+        ```
+
+        In this example the `prompts` will be converted into:
+
+        ```
+        <s>User:<fake_token_around_image><image><fake_token_around_image>Describe this image.
+        Assistant: An image of two kittens in grass.
+        User:<fake_token_around_image><image><fake_token_around_image>Describe this image.
+        Assistant:'
+        ```
+
+        and the two images will be massaged using [`IdeficsImageProcessor.__call__`] method and placed inside the
+        `pixel_values` dict entry of the return value.
+
+        This example also examplifies that images can be passed as objects or as text urls. It can be seen that the
+        first image is passed as object and the second one as a url.
+
+        To do training do:
+
+        ```python
+        image_transform = transforms.Compose(
+            [
+                transforms.RandomResizedCrop(
+                    (w, h), scale=(0.9, 1.0), interpolation=transforms.InterpolationMode.BICUBIC
+                ),
+                transforms.ToTensor(),
+                transforms.Normalize(mean=self.image_mean, std=self.image_std),
+            ]
+        )
+        inputs = processor(prompts, transform=image_transform, return_tensors="pt")
+        ```
+
+        In order to help debug prompt generation enable `debug=True` which will show you what's happening.
+
+        """
+
+        # if the value isn't overriden by the user, check if the tokenizer was trained with this token and then use it
+        if add_end_of_utterance_token is None:
+            add_end_of_utterance_token = self.tokenizer_was_trained_with_end_of_utterance_token
+
+        # turn non-batched prompts into batched
+        if not any(isinstance(i, list) for i in prompts):
+            prompts = [prompts]
+
+        fake_token = "<fake_token_around_image>"
+        image_token = "<image>"
+        end_of_utterance_token = "<end_of_utterance>"
+
+        def image_tokens(last_was_image):
+            if last_was_image:
+                return image_token + fake_token
+            else:
+                return fake_token + image_token + fake_token
+
+        all_texts = []
+        all_images = []
+        for sample in prompts:
+            # the model was trained on samples starting with <s>
+            full_text = f"{self.tokenizer.bos_token}"
+
+            # an image can either be an image object in the item or the url, everything else is a verbatim prompt text
+            image_objects = []
+            last_was_image = False
+            last_was_text = False
+            for i, item in enumerate(sample):
+                if i > 0:
+                    last_was_text = True if not last_was_image else False
+
+                if isinstance(item, str):
+                    item = item.strip(" ")
+                    if is_url(item):
+                        image = self.image_processor.fetch_images(item)
+                        full_text += image_tokens(last_was_image)
+                        image_objects.append(image)
+                        last_was_image = True
+                    else:
+                        # we add end_of_utterance_token between each subsequent text prompts (but not at the last one!)
+                        if add_end_of_utterance_token and last_was_text:
+                            full_text += end_of_utterance_token
+                        full_text += item
+                        last_was_image = False
+                else:
+                    # must be an image obj
+                    full_text += image_tokens(last_was_image)
+                    image_objects.append(item)
+                    last_was_image = True
+
+            if add_eos_token:
+                full_text += self.tokenizer.eos_token
+
+            if debug is True:
+                print(f"{full_text=}")
+
+            image_objects = self.image_processor(image_objects, transform=transform)
+
+            text_encoding = self.tokenizer(
+                text=full_text,
+                add_special_tokens=False,
+                padding=padding,
+                truncation=truncation,
+                max_length=max_length,
+            )
+
+            all_texts.append(text_encoding["input_ids"])
+            all_images.append(image_objects)
+
+        max_seq_len = max(len(x) for x in all_texts)
+
+        # max_num_images has to be at least 1 even when there are no images
+        max_num_images = max(len(x) for x in all_images)
+        max_num_images = max(1, max_num_images)
+
+        at_least_one_image = sum(len(x) for x in all_images) > 0
+        output_input_ids = []
+        output_images = []
+        output_attention_masks = []
+        for text, images in zip(all_texts, all_images):
+            padded_input_ids = [self.tokenizer.pad_token_id] * max_seq_len
+            unpadded_seq_len = len(text)
+            start = max_seq_len - unpadded_seq_len
+            padded_input_ids[start:] = text[:max_seq_len]
+
+            attention_mask = torch.zeros((max_seq_len,), dtype=torch.long)
+            attention_mask[start:] = 1
+
+            image_count = padded_input_ids.count(self.image_token_id)
+            local_max_num_images = min(image_count, max_num_images)
+
+            current_images = images[:local_max_num_images]
+
+            if len(current_images) > 0:
+                padded_image_tensor = torch.zeros(max_num_images, *current_images.size()[1:])
+                padded_image_tensor[: current_images.size(0)] = current_images
+            else:
+                padded_image_tensor = torch.zeros(max_num_images, *self.default_image_dims)
+
+            output_images.append(padded_image_tensor)
+            output_input_ids.append(torch.tensor(padded_input_ids))
+
+            output_attention_masks.append(attention_mask)
+
+        output_input_ids = torch.stack(output_input_ids)
+        output_images = torch.stack(output_images)
+        output_attention_masks = torch.stack(output_attention_masks)
+
+        if at_least_one_image:
+            image_attention_mask, _ = image_attention_mask_for_packed_input_ids(output_input_ids, self.tokenizer)
+            image_attention_mask = incremental_to_binary_attention_mask(
+                image_attention_mask, num_classes=max_num_images
+            )
+        else:
+            # in full language mode we set the image mask to all-0s
+            image_attention_mask = torch.zeros(
+                output_input_ids.shape[0], output_input_ids.shape[1], 1, dtype=torch.bool
+            )
+
+        return BatchFeature(
+            data={
+                "input_ids": output_input_ids,
+                "attention_mask": output_attention_masks,
+                "pixel_values": output_images,
+                "image_attention_mask": image_attention_mask,
+            }
+        )
+
+    def batch_decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
+        refer to the docstring of this method for more information.
+        """
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    def decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
+        the docstring of this method for more information.
+        """
+        return self.tokenizer.decode(*args, **kwargs)
+
+    @property
+    def model_input_names(self):
+        tokenizer_input_names = self.tokenizer.model_input_names
+        image_processor_input_names = self.image_processor.model_input_names
+        return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
diff --git a/server/text_generation_server/models/custom_modeling/idefics_vision.py b/server/text_generation_server/models/custom_modeling/idefics_vision.py
new file mode 100644
index 00000000..d933d7c1
--- /dev/null
+++ b/server/text_generation_server/models/custom_modeling/idefics_vision.py
@@ -0,0 +1,476 @@
+# coding=utf-8
+# Copyright 2021 The OpenAI Team Authors and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" PyTorch IdeficsVision model: a copy of CLIPVisionModel using a simpler config object"""
+
+
+from dataclasses import dataclass
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
+from transformers.utils import (
+    ModelOutput,
+    logging,
+)
+from text_generation_server.utils.layers import (
+    TensorParallelColumnLinear,
+    TensorParallelRowLinear,
+    TensorParallelEmbedding,
+)
+
+logger = logging.get_logger(__name__)
+
+
+@dataclass
+class IdeficsVisionModelOutput(ModelOutput):
+    """
+    Base class for vision model's outputs that also contains image embeddings of the pooling of the last hidden states.
+
+    Args:
+        image_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim)` *optional* returned when model is initialized with `with_projection=True`):
+            The image embeddings obtained by applying the projection layer to the pooler_output.
+        last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model.
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    image_embeds: Optional[torch.FloatTensor] = None
+    last_hidden_state: torch.FloatTensor = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+
+
+# Copied from transformers.models.clip.modeling_clip.CLIPVisionEmbeddings with CLIP->Idefics
+class IdeficsVisionEmbeddings(nn.Module):
+    def __init__(self, prefix, config, weights):
+        super().__init__()
+        self.config = config
+        self.embed_dim = config.hidden_size
+        self.image_size = config.image_size
+        self.patch_size = config.patch_size
+
+        self.class_embedding = nn.Parameter(weights.get_tensor(f"{prefix}.class_embedding"))
+
+        self.patch_embedding = nn.Conv2d.load_no_bias(
+            prefix=f"{prefix}.patch_embedding",
+            weights=weights,
+            in_channels=config.num_channels,
+            out_channels=self.embed_dim,
+            kernel_size=self.patch_size,
+            stride=self.patch_size,
+        )
+
+        self.num_patches = (self.image_size // self.patch_size) ** 2
+        self.num_positions = self.num_patches + 1
+        # self.position_embedding = nn.Embedding(self.num_positions, self.embed_dim)
+        self.position_embedding = TensorParallelEmbedding(
+            prefix="model.vision_model.embeddings.position_embedding", weights=weights
+        )
+        # self.register_buffer("position_ids", torch.arange(self.num_positions).expand((1, -1)), persistent=False)
+        self.position_ids = weights.get_tensor(f"{prefix}.position_ids")
+
+    def forward(self, pixel_values: torch.FloatTensor) -> torch.Tensor:
+        batch_size = pixel_values.shape[0]
+        target_dtype = self.patch_embedding.weight.dtype
+        patch_embeds = self.patch_embedding(pixel_values.to(dtype=target_dtype))  # shape = [*, width, grid, grid]
+        patch_embeds = patch_embeds.flatten(2).transpose(1, 2)
+
+        class_embeds = self.class_embedding.expand(batch_size, 1, -1)
+        embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
+        embeddings = embeddings + self.position_embedding(self.position_ids)
+        return embeddings
+
+
+# Copied from transformers.models.clip.modeling_clip.CLIPAttention with CLIP->IdeficsVision
+class IdeficsVisionAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, prefix, config, weights):
+        super().__init__()
+        self.config = config
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_heads
+        if self.head_dim * self.num_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
+                f" {self.num_heads})."
+            )
+        self.scale = self.head_dim**-0.5
+        self.dropout = config.attention_dropout
+
+        process_group = weights.process_group
+        if self.num_heads % weights.process_group.size() != 0:
+            raise ValueError(
+                f"`num_heads` must be divisible by `num_shards` (got `num_heads`: {self.num_heads} "
+                f"and `num_shards`: {weights.process_group.size()}"
+            )
+        self.num_heads = self.num_heads // weights.process_group.size()
+        self.embed_dim = self.embed_dim // weights.process_group.size()
+
+
+        self.k_proj = TensorParallelColumnLinear.load(
+            config, prefix=f"{prefix}.k_proj", weights=weights, bias=True
+        )
+        self.v_proj = TensorParallelColumnLinear.load(
+            config, prefix=f"{prefix}.v_proj", weights=weights, bias=True
+        )
+        self.q_proj = TensorParallelColumnLinear.load(
+            config, prefix=f"{prefix}.q_proj", weights=weights, bias=True
+        )
+        self.out_proj = TensorParallelRowLinear.load(
+            config, prefix=f"{prefix}.out_proj", weights=weights, bias=True
+        )
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        causal_attention_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        """Input shape: Batch x Time x Channel"""
+
+        bsz, tgt_len, _ = hidden_states.size()
+
+        # get query proj
+        query_states = self.q_proj(hidden_states) * self.scale
+        key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
+        value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
+
+        proj_shape = (bsz * self.num_heads, -1, self.head_dim)
+        query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
+        key_states = key_states.view(*proj_shape)
+        value_states = value_states.view(*proj_shape)
+
+        src_len = key_states.size(1)
+        attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))
+
+        if attn_weights.size() != (bsz * self.num_heads, tgt_len, src_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz * self.num_heads, tgt_len, src_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        # apply the causal_attention_mask first
+        if causal_attention_mask is not None:
+            if causal_attention_mask.size() != (bsz, 1, tgt_len, src_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is"
+                    f" {causal_attention_mask.size()}"
+                )
+            attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + causal_attention_mask
+            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, tgt_len, src_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + attention_mask
+            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
+
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1)
+
+        if output_attentions:
+            # this operation is a bit akward, but it's required to
+            # make sure that attn_weights keeps its gradient.
+            # In order to do so, attn_weights have to reshaped
+            # twice and have to be reused in the following
+            attn_weights_reshaped = attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
+            attn_weights = attn_weights_reshaped.view(bsz * self.num_heads, tgt_len, src_len)
+        else:
+            attn_weights_reshaped = None
+
+        attn_probs = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)
+
+        attn_output = torch.bmm(attn_probs, value_states)
+
+        if attn_output.size() != (bsz * self.num_heads, tgt_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, tgt_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.view(bsz, self.num_heads, tgt_len, self.head_dim)
+        attn_output = attn_output.transpose(1, 2)
+        attn_output = attn_output.reshape(bsz, tgt_len, self.embed_dim)
+
+        attn_output = self.out_proj(attn_output)
+
+        return attn_output, attn_weights_reshaped
+
+
+# Copied from transformers.models.clip.modeling_clip.CLIPMLP with CLIP->IdeficsVision
+class IdeficsVisionMLP(nn.Module):
+    def __init__(self, prefix, config, weights):
+        super().__init__()
+        self.config = config
+        self.activation_fn = ACT2FN[config.hidden_act]
+        self.fc1 = TensorParallelColumnLinear.load(
+            config, prefix=f"{prefix}.fc1", weights=weights, bias=True
+        )
+        self.fc2 = TensorParallelRowLinear.load(
+            config, prefix=f"{prefix}.fc2", weights=weights, bias=True
+        )
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.fc1(hidden_states)
+        hidden_states = self.activation_fn(hidden_states)
+        hidden_states = self.fc2(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.clip.modeling_clip.CLIPEncoderLayer with CLIP->IdeficsVision
+class IdeficsVisionEncoderLayer(nn.Module):
+    def __init__(self, prefix, config, weights):
+        super().__init__()
+        self.embed_dim = config.hidden_size
+        self.self_attn = IdeficsVisionAttention(prefix=f"{prefix}.self_attn", config=config, weights=weights)
+        self.layer_norm1 = nn.LayerNorm.load(
+            prefix=f"{prefix}.layer_norm1", weights=weights, eps=config.layer_norm_eps
+        )
+        self.mlp = IdeficsVisionMLP(prefix=f"{prefix}.mlp", config=config, weights=weights)
+        self.layer_norm2 = nn.LayerNorm.load(
+            prefix=f"{prefix}.layer_norm2", weights=weights, eps=config.layer_norm_eps
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor,
+        causal_attention_mask: torch.Tensor,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+                `(config.encoder_attention_heads,)`.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+        """
+        residual = hidden_states
+
+        hidden_states = self.layer_norm1(hidden_states)
+        hidden_states, attn_weights = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            causal_attention_mask=causal_attention_mask,
+            output_attentions=output_attentions,
+        )
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.layer_norm2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs
+
+
+# Copied from transformers.models.clip.modeling_clip.CLIPEncoder with CLIP->IdeficsVision
+class IdeficsVisionEncoder(nn.Module):
+    """
+    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
+    [`IdeficsVisionEncoderLayer`].
+
+    Args:
+        config: IdeficsVisionConfig
+    """
+
+    def __init__(self, prefix, config, weights):
+        super().__init__()
+        self.config = config
+        self.layers = nn.ModuleList(
+            [
+                IdeficsVisionEncoderLayer(prefix=f"{prefix}.encoder.layers.{layer_id}", config=config, weights=weights)
+                for layer_id in range(config.num_hidden_layers)
+            ]
+        )
+        # self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        inputs_embeds,
+        attention_mask: Optional[torch.Tensor] = None,
+        causal_attention_mask: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutput]:
+        r"""
+        Args:
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
+                Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
+                This is useful if you want more control over how to convert `input_ids` indices into associated vectors
+                than the model's internal embedding lookup matrix.
+            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+                - 1 for tokens that are **not masked**,
+                - 0 for tokens that are **masked**.
+
+                [What are attention masks?](../glossary#attention-mask)
+            causal_attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Causal mask for the text model. Mask values selected in `[0, 1]`:
+
+                - 1 for tokens that are **not masked**,
+                - 0 for tokens that are **masked**.
+
+                [What are attention masks?](../glossary#attention-mask)
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            output_hidden_states (`bool`, *optional*):
+                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
+                for more detail.
+            return_dict (`bool`, *optional*):
+                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        encoder_states = () if output_hidden_states else None
+        all_attentions = () if output_attentions else None
+
+        hidden_states = inputs_embeds
+        for idx, encoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                encoder_states = encoder_states + (hidden_states,)
+            # if self.gradient_checkpointing and self.training:
+
+            #     def create_custom_forward(module):
+            #         def custom_forward(*inputs):
+            #             return module(*inputs, output_attentions)
+
+            #         return custom_forward
+
+            #     layer_outputs = torch.utils.checkpoint.checkpoint(
+            #         create_custom_forward(encoder_layer),
+            #         hidden_states,
+            #         attention_mask,
+            #         causal_attention_mask,
+            #     )
+            # else:
+            layer_outputs = encoder_layer(
+                hidden_states,
+                attention_mask,
+                causal_attention_mask,
+                output_attentions=output_attentions,
+            )
+
+            hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_attentions = all_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            encoder_states = encoder_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, encoder_states, all_attentions] if v is not None)
+        return BaseModelOutput(
+            last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
+        )
+
+
+# Adapted from transformers.models.clip.modeling_clip.CLIPVisionTransformer
+class IdeficsVisionTransformer(nn.Module):
+    def __init__(self, prefix, config, weights):
+        super().__init__()
+        self.config = config
+        embed_dim = config.hidden_size
+
+        self.embeddings = IdeficsVisionEmbeddings(prefix=f"{prefix}.embeddings", config=config, weights=weights)
+        self.pre_layrnorm = nn.LayerNorm.load(
+            prefix=f"{prefix}.pre_layrnorm", weights=weights, eps=config.layer_norm_eps
+        )
+        self.encoder = IdeficsVisionEncoder(prefix=prefix, config=config, weights=weights)
+        self.post_layernorm = nn.LayerNorm.load(
+            prefix=f"{prefix}.post_layernorm", weights=weights, eps=config.layer_norm_eps
+        )
+
+    # copied from transformers.models.clip.modeling_clip.CLIPVisionTransformer.forward
+    def forward(
+        self,
+        pixel_values: Optional[torch.FloatTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPooling]:
+        r"""
+        Returns:
+
+        """
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if pixel_values is None:
+            raise ValueError("You have to specify pixel_values")
+
+        hidden_states = self.embeddings(pixel_values)
+        hidden_states = self.pre_layrnorm(hidden_states)
+
+        encoder_outputs = self.encoder(
+            inputs_embeds=hidden_states,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        last_hidden_state = encoder_outputs[0]
+        pooled_output = last_hidden_state[:, 0, :]
+        pooled_output = self.post_layernorm(pooled_output)
+
+        if not return_dict:
+            return (last_hidden_state, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=last_hidden_state,
+            pooler_output=pooled_output,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+        )
diff --git a/server/text_generation_server/models/idefics.py b/server/text_generation_server/models/idefics.py
new file mode 100644
index 00000000..c54b539b
--- /dev/null
+++ b/server/text_generation_server/models/idefics.py
@@ -0,0 +1,91 @@
+import torch
+import torch.distributed
+
+from typing import List, Optional, Tuple
+
+from transformers import (
+    AutoTokenizer,
+    AutoConfig,
+    AutoProcessor,
+)
+
+from text_generation_server.models.custom_modeling.idefics_config import IdeficsConfig
+from text_generation_server.models.custom_modeling.idefics_processing import (
+    IdeficsProcessor,
+)
+from transformers import LlamaTokenizerFast
+from text_generation_server.models.custom_modeling.idefics_modeling import (
+    IdeficsForVisionText2Text,
+)
+from text_generation_server.models.idefics_causal_lm import IdeficsCausalLM
+from text_generation_server.utils import (
+    initialize_torch_distributed,
+    weight_files,
+    Weights,
+)
+
+
+class IDEFICSSharded(IdeficsCausalLM):
+    def __init__(
+        self,
+        model_id: str,
+        revision: Optional[str] = None,
+        quantize: Optional[str] = None,
+        dtype: Optional[torch.dtype] = None,
+        trust_remote_code: bool = False,
+    ):
+        self.process_group, rank, world_size = initialize_torch_distributed()
+        if torch.cuda.is_available():
+            device = torch.device(f"cuda:{rank}")
+            # 9b seems to work correctly enough in float16, but 80b seems
+            # to be really saturating for f16.
+            dtype = torch.bfloat16 if dtype is None else dtype
+        else:
+            device = torch.device("cpu")
+            dtype = torch.float32
+        self.device, self.dtype = device, dtype
+
+        config = IdeficsConfig.from_pretrained(
+            model_id,
+            revision=revision,
+            trust_remote_code=trust_remote_code,
+        )
+        config.quantize = quantize
+        config.vision_config.quantize = quantize
+
+        tokenizer = LlamaTokenizerFast.from_pretrained(
+            model_id,
+            revision=revision,
+            padding_side="left",
+            truncation_side="left",
+            trust_remote_code=trust_remote_code,
+        )
+        self.processor = IdeficsProcessor.from_pretrained(
+            model_id,
+            revision=revision,
+            padding_side="left",
+            truncation_side="left",
+            trust_remote_code=trust_remote_code,
+        )
+
+        torch.distributed.barrier(group=self.process_group)
+        filenames = weight_files(model_id, revision=revision, extension=".safetensors")
+        weights = Weights(
+            filenames,
+            device=device,
+            dtype=dtype,
+            process_group=self.process_group,
+        )
+
+        model = IdeficsForVisionText2Text(config, weights)
+
+        torch.distributed.barrier(group=self.process_group)
+        super(IdeficsCausalLM, self).__init__(
+            model=model,
+            tokenizer=tokenizer,
+            requires_padding=True,
+            dtype=dtype,
+            device=device,
+            rank=rank,
+            world_size=world_size,
+        )
diff --git a/server/text_generation_server/models/idefics_causal_lm.py b/server/text_generation_server/models/idefics_causal_lm.py
new file mode 100644
index 00000000..faad63ba
--- /dev/null
+++ b/server/text_generation_server/models/idefics_causal_lm.py
@@ -0,0 +1,806 @@
+import torch
+import inspect
+import re
+from io import BytesIO
+import base64
+from PIL import Image
+import re
+
+from dataclasses import dataclass
+from opentelemetry import trace
+from transformers import AutoProcessor, AutoTokenizer, AutoModelForCausalLM, PreTrainedTokenizerBase, ProcessorMixin
+from typing import Optional, Tuple, List, Type, Dict
+
+from text_generation_server.models import Model
+from text_generation_server.models.types import (
+    Batch,
+    PrefillTokens,
+    Generation,
+    GeneratedText,
+)
+from text_generation_server.pb import generate_pb2
+from text_generation_server.utils import NextTokenChooser, StoppingCriteria, Sampling
+
+import re
+
+IMAGES = re.compile(r'!\[[^\]]*\]\((.*?)\s*(\"(?:.*[^\"])\")?\s*\)')
+
+def split(string):
+    parts = []
+    cursor = 0
+    for pattern in IMAGES.finditer(string):
+        start = pattern.start()
+        if start != cursor:
+            parts.append(string[cursor:start])
+
+        parts.append(pattern.group(1))
+        cursor = pattern.end()
+
+    if cursor != len(string):
+        parts.append(string[cursor:])
+
+    return parts
+
+tracer = trace.get_tracer(__name__)
+
+
+@dataclass
+class IdeficsCausalLMBatch(Batch):
+    batch_id: int
+    requests: List[generate_pb2.Request]
+    requests_idx_mapping: Dict[int, int]
+
+    # Decoder values
+    input_ids: torch.Tensor
+    attention_mask: torch.Tensor
+    position_ids: torch.Tensor
+    pixel_values: Optional[torch.Tensor]
+    image_hidden_states: Optional[torch.Tensor]
+    image_attention_mask: Optional[torch.Tensor]
+    past_key_values: Optional[List[Tuple]]
+
+    # All tokens
+    all_input_ids: List[torch.Tensor]
+
+    # Lengths of all generations present in the batch
+    input_lengths: List[int]
+    prefix_offsets: List[int]
+    read_offsets: List[int]
+
+    # Generation helpers
+    next_token_choosers: List[NextTokenChooser]
+    stopping_criterias: List[StoppingCriteria]
+
+    # Metadata used for padding
+    max_input_length: int
+    padding_right_offset: int
+
+    # Maximum number of tokens this batch will grow to
+    max_tokens: int
+
+    # Past metadata
+    keys_head_dim_last: bool = True
+
+    def to_pb(self) -> generate_pb2.CachedBatch:
+        return generate_pb2.CachedBatch(
+            id=self.batch_id,
+            request_ids=[r.id for r in self.requests],
+            size=len(self),
+            max_tokens=self.max_tokens,
+        )
+
+    @classmethod
+    def from_pb(
+        cls,
+        pb: generate_pb2.Batch,
+        tokenizer: PreTrainedTokenizerBase,
+        processor: ProcessorMixin, # Hack
+        dtype: torch.dtype,
+        device: torch.device,
+    ) -> "IdeficsCausalLMBatch":
+        inputs = []
+        next_token_choosers = []
+        stopping_criterias = []
+        prefix_offsets = []
+        read_offsets = []
+        requests_idx_mapping = {}
+
+        # Parse batch
+        max_truncation = 0
+        padding_right_offset = 0
+        max_decode_tokens = 0
+        for i, r in enumerate(pb.requests):
+            requests_idx_mapping[r.id] = i
+            inputs.append(r.inputs)
+            next_token_choosers.append(NextTokenChooser.from_pb(r.parameters, device))
+            stopping_criteria = StoppingCriteria.from_pb(
+                r.stopping_parameters, tokenizer
+            )
+            stopping_criterias.append(stopping_criteria)
+            max_truncation = max(max_truncation, r.truncate)
+            max_decode_tokens += stopping_criteria.max_new_tokens
+            padding_right_offset = max(
+                padding_right_offset, stopping_criteria.max_new_tokens
+            )
+
+        prompts = []
+        for inp in inputs:
+            # Each input is encoded into a list, where each element of this input list is either a string or a URL
+            prompts.append(split(inp))
+
+        # The processor replaces the call to tokenizer, and
+        # a/ takes care of fetching images from the URL
+        # b/ generate the correct input_ids, attention_mask, pixel_values, image_attention_mask to feed to the model
+        tokenized_inputs = processor(
+            prompts,
+            return_tensors="pt",
+            padding=True,
+            truncation=True,
+            max_length=max_truncation,
+            add_end_of_utterance_token=False, # Already taken care of inside the prompts, so bypassing the processor's handling of this token
+        ).to(device)
+        for _ in pb.requests:
+            input_len = tokenized_inputs["input_ids"].shape[1]
+            prefix_offsets.append(input_len - 5) # To decode without potential fallbacks errors
+            read_offsets.append(input_len) # To decode without potential fallbacks errors
+
+        input_lengths = tokenized_inputs["attention_mask"].sum(1)
+        max_input_length = input_lengths.max()
+
+        input_ids = tokenized_inputs["input_ids"]
+        pixel_values = tokenized_inputs["pixel_values"]
+        image_hidden_states = None
+        # Allocate maximum attention_mask
+        attention_mask = input_ids.new_zeros(
+            (pb.size, max_input_length + padding_right_offset)
+        )
+        # Copy tokenizer attention_mask into fully allocated attention_mask
+        attention_mask[:, :max_input_length] = tokenized_inputs["attention_mask"]
+        # Do the same for image_attention_mask
+        image_attention_mask = input_ids.new_zeros(
+            (pb.size, max_input_length + padding_right_offset, tokenized_inputs["pixel_values"].size(1))
+        )
+        image_attention_mask[:, :max_input_length, :] = tokenized_inputs["image_attention_mask"]
+
+
+        position_ids = tokenized_inputs["attention_mask"].long().cumsum(-1) - 1
+        position_ids.masked_fill_(tokenized_inputs["attention_mask"] == 0, 1)
+        all_input_ids = tokenized_inputs["input_ids"].T.split(1, dim=1) # It's input_ids but splitted into a tuple of tensors where each tensor is (seq_len, 1) size. It is then transformed into a list
+
+        max_tokens = len(inputs) * (max_input_length + max_decode_tokens)
+
+        return cls(
+            batch_id=pb.id,
+            requests=pb.requests,
+            requests_idx_mapping=requests_idx_mapping,
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            pixel_values=pixel_values,
+            image_hidden_states=image_hidden_states,
+            image_attention_mask=image_attention_mask,
+            past_key_values=None,
+            all_input_ids=list(all_input_ids),
+            input_lengths=input_lengths.tolist(),
+            prefix_offsets=prefix_offsets,
+            read_offsets=read_offsets,
+            next_token_choosers=next_token_choosers,
+            stopping_criterias=stopping_criterias,
+            max_input_length=max_input_length.item(),
+            padding_right_offset=padding_right_offset,
+            max_tokens=max_tokens,
+        )
+
+    @tracer.start_as_current_span("filter")
+    def filter(self, request_ids: List[int]) -> Optional["IdeficsCausalLMBatch"]:
+        # It deletes requests from the batch. For instance when client lost connection
+        if len(request_ids) == 0:
+            raise ValueError("Batch must have at least one request")
+        if len(request_ids) == len(self):
+            return self
+
+        keep_indices = []
+
+        # New values after filtering
+        requests_idx_mapping = {}
+        requests = []
+        input_lengths = []
+        prefix_offsets = []
+        read_offsets = []
+        all_input_ids = []
+        max_input_length = 0
+
+        next_token_choosers = []
+        stopping_criterias = []
+
+        total_remaining_decode_tokens = 0
+        new_padding_right_offset = 0
+
+        for i, request_id in enumerate(request_ids):
+            idx = self.requests_idx_mapping[request_id]
+            requests_idx_mapping[request_id] = i
+            keep_indices.append(idx)
+
+            requests.append(self.requests[idx])
+            prefix_offsets.append(self.prefix_offsets[idx])
+            read_offsets.append(self.read_offsets[idx])
+            all_input_ids.append(self.all_input_ids[idx])
+
+            request_input_length = self.input_lengths[idx]
+            input_lengths.append(request_input_length)
+            max_input_length = max(max_input_length, request_input_length)
+
+            next_token_choosers.append(self.next_token_choosers[idx])
+            stopping_criteria = self.stopping_criterias[idx]
+            stopping_criterias.append(stopping_criteria)
+            remaining_decode_tokens = (
+                stopping_criteria.max_new_tokens - stopping_criteria.current_tokens
+            )
+            total_remaining_decode_tokens += remaining_decode_tokens
+            new_padding_right_offset = max(
+                new_padding_right_offset, remaining_decode_tokens
+            )
+
+        # Apply indices to input_ids, attention mask, past key values and other items that need to be cached
+        input_ids = self.input_ids[keep_indices]
+        position_ids = self.position_ids[keep_indices]
+        self.attention_mask = self.attention_mask[
+            keep_indices,
+            -(self.padding_right_offset + max_input_length) : (
+                self.attention_mask.shape[1] - self.padding_right_offset
+            )
+            + new_padding_right_offset,
+        ]
+        # Do the same for pixel_values and image_attention_mask
+        pixel_values = self.pixel_values[keep_indices]
+        self.image_attention_mask = self.image_attention_mask[
+            keep_indices,
+            -(self.padding_right_offset + max_input_length) : (
+                self.image_attention_mask.shape[1] - self.padding_right_offset
+            )
+            + new_padding_right_offset,
+            :
+        ]
+        if self.image_hidden_states is None:
+            image_hidden_states = None
+        else:
+            image_hidden_states = self.image_hidden_states[keep_indices]
+
+        # Ensure that past_key_values tensors can be updated in-place
+        if type(self.past_key_values[0]) == tuple:
+            self.past_key_values = [list(layer) for layer in self.past_key_values]
+
+        # Update tensors in-place to allow incremental garbage collection
+        past_kv_length = max_input_length - 1
+        for layer in self.past_key_values:
+            past_keys, past_values = layer
+            if len(past_keys.shape) == 3:
+                # Force past to be of dim [self_size, num_heads, ...] for easy indexing
+                past_keys = past_keys.view(len(self), -1, *past_keys.shape[-2:])
+                past_values = past_values.view(len(self), -1, *past_values.shape[-2:])
+            if self.keys_head_dim_last:
+                layer[0] = past_keys[keep_indices, :, -past_kv_length:, :]
+            else:
+                layer[0] = past_keys[keep_indices, :, :, -past_kv_length:]
+            del past_keys
+            layer[1] = past_values[keep_indices, :, -past_kv_length:, :]
+            del past_values
+
+        max_tokens = len(request_ids) * max_input_length + total_remaining_decode_tokens
+
+        self.requests = requests
+        self.requests_idx_mapping = requests_idx_mapping
+        self.input_ids = input_ids
+        self.pixel_values = pixel_values
+        self.image_hidden_states = image_hidden_states
+        self.position_ids = position_ids
+        self.all_input_ids = all_input_ids
+        self.input_lengths = input_lengths
+        self.prefix_offsets = prefix_offsets
+        self.read_offsets = read_offsets
+        self.next_token_choosers = next_token_choosers
+        self.stopping_criterias = stopping_criterias
+        self.max_input_length = max_input_length
+        self.padding_right_offset = new_padding_right_offset
+        self.max_tokens = max_tokens
+
+        return self
+
+    @classmethod
+    @tracer.start_as_current_span("concatenate")
+    def concatenate(cls, batches: List["IdeficsCausalLMBatch"]) -> "IdeficsCausalLMBatch":
+        # It adds new requests to the batch
+        # Used for padding
+        total_batch_size = 0
+        max_input_length = 0
+        max_num_images = 0
+        padding_right_offset = 0
+        for batch in batches:
+            total_batch_size += len(batch)
+            max_input_length = max(max_input_length, batch.max_input_length)
+            max_num_images = max(max_num_images, batch.pixel_values.size(1))
+            padding_right_offset = max(padding_right_offset, batch.padding_right_offset)
+
+        # Batch attributes
+        requests = []
+        requests_idx_mapping = {}
+        input_lengths = []
+        prefix_offsets = []
+        read_offsets = []
+        all_input_ids = []
+        next_token_choosers = []
+        stopping_criterias = []
+        max_tokens = 0
+
+        # Batch tensors
+        input_ids = None
+        attention_mask = None
+        position_ids = None
+        pixel_values = None
+        image_hidden_states = None
+        image_attention_mask = None
+        past_key_values = []
+
+        # Used for slicing correctly inside the tensors
+        # Equivalent to a cumsum on batch sizes
+        start_index = 0
+        for i, batch in enumerate(batches):
+            requests.extend(batch.requests)
+            input_lengths.extend(batch.input_lengths)
+            prefix_offsets.extend(batch.prefix_offsets)
+            read_offsets.extend(batch.read_offsets)
+            all_input_ids.extend(batch.all_input_ids)
+            next_token_choosers.extend(batch.next_token_choosers)
+            stopping_criterias.extend(batch.stopping_criterias)
+
+            if i == 0:
+                requests_idx_mapping = batch.requests_idx_mapping
+            else:
+                # We need to offset the mapping for each batch by the cumulative batch size
+                for k, v in batch.requests_idx_mapping.items():
+                    requests_idx_mapping[k] = v + start_index
+
+            # Slicing end index for this batch
+            end_index = start_index + len(batch)
+
+            # We only concatenate batches that did at least one step
+            if batch.past_key_values is None:
+                raise ValueError("only concatenate prefilled batches")
+
+            # Create empty tensor
+            # input_ids is always of shape [batch_size, 1]
+            # We do not need to pad it
+            if input_ids is None:
+                input_ids = batch.input_ids.new_empty((total_batch_size, 1))
+            # Copy to correct indices
+            input_ids[start_index:end_index] = batch.input_ids
+
+            # Create padded tensor
+            if attention_mask is None:
+                attention_mask = batch.attention_mask.new_zeros(
+                    (total_batch_size, max_input_length + padding_right_offset),
+                )
+
+            curr_batch_max_num_images = batch.pixel_values.size(1)
+            if pixel_values is None:
+                pixel_values = batch.pixel_values.new_zeros((total_batch_size, max_num_images, 3, 224, 224))
+            pixel_values[start_index:end_index, :curr_batch_max_num_images] = batch.pixel_values
+
+            if image_attention_mask is None:
+                image_attention_mask = batch.image_attention_mask.new_zeros(
+                    (total_batch_size, max_input_length + padding_right_offset, max_num_images)
+                )
+
+            # We need to slice the attention mask to remove padding from previous steps
+            # and to remove unused allocated space
+            left_offset = max_input_length - batch.max_input_length
+            batch_left_offset = (
+                batch.attention_mask.shape[1]
+                - batch.max_input_length
+                - batch.padding_right_offset
+            )
+            attention_mask[
+                start_index:end_index,
+                left_offset:-padding_right_offset,
+            ] = batch.attention_mask[
+                :,
+                batch_left_offset : -batch.padding_right_offset,
+            ]
+            image_attention_mask[
+                start_index:end_index,
+                left_offset:-padding_right_offset,
+                :curr_batch_max_num_images
+            ] = batch.image_attention_mask[
+                :,
+                batch_left_offset : - batch.padding_right_offset,
+                :
+            ]
+
+            # Create empty tensor
+            # position_ids is always of shape [batch_size, 1]
+            if position_ids is None:
+                position_ids = batch.position_ids.new_empty((total_batch_size, 1))
+            position_ids[start_index:end_index] = batch.position_ids
+
+            # Shenanigans to get dimensions because BLOOM outputs a past with a different shape
+            # BLOOM Keys:   [batch_size * num_heads, head_dim, seq_length]
+            # BLOOM Values: [batch_size * num_heads, seq_length, head_dim]
+            # And ensure that we can update tensors in-place
+            if type(batch.past_key_values[0]) == tuple:
+                batch.past_key_values = [
+                    [t.view(len(batch), -1, *t.shape[-2:]) for t in layer]
+                    for layer in batch.past_key_values
+                ]
+            elif len(batch.past_key_values[0][0].shape) == 3:
+                for layer in batch.past_key_values:
+                    for k, t in enumerate(layer):
+                        layer[k] = t.view(len(batch), -1, *t.shape[-2:])
+
+            # Add eventual padding tokens that were added while concatenating
+            max_tokens += batch.max_tokens + (
+                max_input_length - batch.max_input_length
+            ) * len(batch)
+
+            start_index = end_index
+
+        first_past_kvs = batches[0].past_key_values
+        _, num_heads, padded_sequence_length, head_dim = first_past_kvs[0][1].shape
+
+        padded_past_values_shape = (
+            total_batch_size,
+            num_heads,
+            max_input_length - 1,
+            head_dim,
+        )
+
+        if batches[0].keys_head_dim_last:
+            padded_past_keys_shape = padded_past_values_shape
+        else:
+            # seq_length is last for BLOOM
+            padded_past_keys_shape = (
+                total_batch_size,
+                num_heads,
+                head_dim,
+                max_input_length - 1,
+            )
+
+        # Iterate over attention layers
+        # Concatenate past key values layer by layer to allow incremental garbage collection
+        for j in range(len(first_past_kvs)):
+            padded_past_keys = first_past_kvs[j][0].new_zeros(padded_past_keys_shape)
+            start_index = 0
+            for batch in batches:
+                past_keys = batch.past_key_values[j][0]
+                # Clear reference to the original tensor
+                batch.past_key_values[j][0] = None
+
+                # Slicing end index for this batch
+                end_index = start_index + len(batch)
+                # We slice the keys to remove the padding from previous batches
+                past_seq_len = batch.max_input_length - 1
+                if batch.keys_head_dim_last:
+                    padded_past_keys[
+                        start_index:end_index, :, -past_seq_len:, :
+                    ] = past_keys[:, :, -past_seq_len:, :]
+                else:
+                    # BLOOM case
+                    padded_past_keys[
+                        start_index:end_index, :, :, -past_seq_len:
+                    ] = past_keys[:, :, :, -past_seq_len:]
+                del past_keys
+
+                start_index = end_index
+
+            padded_past_values = first_past_kvs[j][1].new_zeros(
+                padded_past_values_shape
+            )
+            start_index = 0
+            for batch in batches:
+                past_values = batch.past_key_values[j][1]
+                # Clear reference to the original tensor
+                batch.past_key_values[j][1] = None
+
+                # Slicing end index for this batch
+                end_index = start_index + len(batch)
+                # We slice the past values to remove the padding from previous batches
+                past_seq_len = batch.max_input_length - 1
+                padded_past_values[
+                    start_index:end_index, :, -past_seq_len:, :
+                ] = past_values[:, :, -past_seq_len:, :]
+                del past_values
+
+                # Update values
+                start_index = end_index
+
+            past_key_values.append([padded_past_keys, padded_past_values])
+
+        return cls(
+            batch_id=batches[0].batch_id,
+            requests=requests,
+            requests_idx_mapping=requests_idx_mapping,
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            pixel_values=pixel_values,
+            image_hidden_states=image_hidden_states,
+            image_attention_mask=image_attention_mask,
+            past_key_values=past_key_values,
+            all_input_ids=all_input_ids,
+            input_lengths=input_lengths,
+            prefix_offsets=prefix_offsets,
+            read_offsets=read_offsets,
+            next_token_choosers=next_token_choosers,
+            stopping_criterias=stopping_criterias,
+            max_input_length=max_input_length,
+            padding_right_offset=padding_right_offset,
+            keys_head_dim_last=batches[0].keys_head_dim_last,
+            max_tokens=max_tokens,
+        )
+
+    def __len__(self):
+        return len(self.requests)
+
+
+class IdeficsCausalLM(Model):
+    def __init__(
+        self,
+        model_id: str,
+        revision: Optional[str] = None,
+        quantize: Optional[str] = None,
+        dtype: Optional[torch.dtype] = None,
+        trust_remote_code: bool = False,
+    ):
+        from text_generation_server.models.custom_modeling.idefics_modeling import IdeficsForVisionText2Text
+
+        if torch.cuda.is_available():
+            device = torch.device("cuda")
+            dtype = torch.float16 if dtype is None else dtype
+        else:
+            if quantize:
+                raise ValueError("quantization is not available on CPU")
+
+            device = torch.device("cpu")
+            dtype = torch.float32
+
+        tokenizer = AutoTokenizer.from_pretrained(
+            model_id,
+            revision=revision,
+            padding_side="left",
+            truncation_side="left",
+            trust_remote_code=trust_remote_code,
+        )
+        self.processor = AutoProcessor.from_pretrained(
+            model_id,
+            revision=revision,
+            padding_side="left",
+            truncation_side="left",
+            trust_remote_code=trust_remote_code,
+        )
+        model = IdeficsForVisionText2Text.from_pretrained(
+            model_id,
+            revision=revision,
+            torch_dtype=dtype,
+            device_map="auto"
+            if torch.cuda.is_available() and torch.cuda.device_count() > 1
+            else None,
+            load_in_8bit=quantize == "bitsandbytes",
+            trust_remote_code=trust_remote_code,
+        )
+        if torch.cuda.is_available() and torch.cuda.device_count() == 1:
+            model = model.cuda()
+
+        if tokenizer.pad_token_id is None:
+            if model.config.pad_token_id is not None:
+                tokenizer.pad_token_id = model.config.pad_token_id
+            elif model.config.eos_token_id is not None:
+                tokenizer.pad_token_id = model.config.eos_token_id
+            elif tokenizer.eos_token_id is not None:
+                tokenizer.pad_token_id = tokenizer.eos_token_id
+            else:
+                tokenizer.add_special_tokens({"pad_token": "<unk>"})
+
+        super(IdeficsCausalLM, self).__init__(
+            model=model,
+            tokenizer=tokenizer,
+            requires_padding=True,
+            dtype=dtype,
+            device=device,
+        )
+
+    @property
+    def batch_type(self) -> Type[IdeficsCausalLMBatch]:
+        return IdeficsCausalLMBatch
+
+    def decode(self, generated_ids: List[int]) -> str:
+        return self.tokenizer.decode(
+            generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
+        )
+
+    def forward(
+        self,
+        input_ids,
+        attention_mask,
+        position_ids,
+        pixel_values,
+        image_hidden_states,
+        image_attention_mask,
+        past_key_values: Optional = None,
+    ) -> Tuple[torch.Tensor, List[Tuple[torch.Tensor, torch.Tensor]]]:
+        # Model Forward
+        kwargs = {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+            "pixel_values": pixel_values,
+            "image_hidden_states": image_hidden_states,
+            "image_attention_mask": image_attention_mask,
+            "past_key_values": past_key_values,
+            "use_cache": True,
+            "return_dict": True,
+        }
+        if self.has_position_ids:
+            kwargs["position_ids"] = position_ids
+
+        outputs = self.model.forward(**kwargs)
+        return outputs.logits, outputs.past_key_values, outputs.image_hidden_states
+
+    @tracer.start_as_current_span("generate_token")
+    def generate_token(
+        self, batch: IdeficsCausalLMBatch
+    ) -> Tuple[List[Generation], Optional[IdeficsCausalLMBatch]]:
+        # slice the attention mask to the correct shape
+        attention_mask = batch.attention_mask[:, : -batch.padding_right_offset]
+        if batch.input_ids.size(1) == 1:
+            # THIS is a hack: when calling idefics.generate, the first time, we need the whole image_attention_mask (size bs x max_seq_len x max_num_images),
+            # but the subsequent times, we only need the last attention mask along the `max_seq_len` dimension
+            # this is due to the nature IDEFICS: it's an encoder decoder, and so when decoding, only the currently generated
+            # token need to attend to the encoder hidden states (i.e. the vision encoder)
+            # Also see seq2seq_lm.Seq2SeqLM.generate_token which has roughly the same logic
+            image_attention_mask = batch.image_attention_mask[:, -(batch.padding_right_offset+1)].unsqueeze(1)
+        else:
+            image_attention_mask = batch.image_attention_mask[:, : -batch.padding_right_offset]
+
+        logits, past, image_hidden_states = self.forward(
+            input_ids=batch.input_ids,
+            attention_mask=attention_mask,
+            position_ids=batch.position_ids,
+            pixel_values=batch.pixel_values,
+            image_hidden_states=batch.image_hidden_states,
+            image_attention_mask=image_attention_mask,
+            past_key_values=batch.past_key_values,
+        )
+        # Hardcoded remove image tokens
+        logits[:, 32000:32001] = torch.finfo(logits.dtype).min
+
+        # Results
+        generations: List[Generation] = []
+        stopped = True
+
+        # Zipped iterator
+        iterator = zip(
+            batch.requests,
+            batch.input_lengths,
+            batch.prefix_offsets,
+            batch.read_offsets,
+            logits,
+            batch.next_token_choosers,
+            batch.stopping_criterias,
+            batch.all_input_ids,
+        )
+
+        # For each member of the batch
+        for i, (
+            request,
+            input_length,
+            prefix_offset,
+            read_offset,
+            logits,
+            next_token_chooser,
+            stopping_criteria,
+            all_input_ids,
+        ) in enumerate(iterator):
+            # Select next token
+            next_token_id, logprobs = next_token_chooser(
+                all_input_ids.view(1, -1), logits[-1:, :]
+            )
+
+            # Append next token to all tokens
+            all_input_ids = torch.cat([all_input_ids, next_token_id])
+            new_input_length = input_length + 1
+
+            # Generated token
+            next_token_logprob = logprobs[-1, next_token_id]
+            next_token_id_squeezed = next_token_id.squeeze()
+            next_token_text, prefix_offset, read_offset = self.decode_token(
+                all_input_ids[:, 0], prefix_offset, read_offset
+            )
+
+            # Evaluate stopping criteria
+            stop, reason = stopping_criteria(
+                next_token_id_squeezed,
+                next_token_text,
+            )
+
+            if not stop:
+                stopped = False
+
+            # Shard generations
+            # All generations will be appended in the rust sharded client
+            if i % self.world_size == self.rank:
+                if stop:
+                    # Decode generated tokens
+                    output_text = self.decode(
+                        all_input_ids[-stopping_criteria.current_tokens :, 0]
+                    )
+                    # Get seed
+                    if isinstance(next_token_chooser.choice, Sampling):
+                        seed = next_token_chooser.choice.seed
+                    else:
+                        seed = None
+
+                    generated_text = GeneratedText(
+                        output_text, stopping_criteria.current_tokens, reason, seed
+                    )
+                else:
+                    generated_text = None
+
+                # Prefill
+                if stopping_criteria.current_tokens == 1 and request.prefill_logprobs:
+                    # Remove generated token to only have prefill and add nan for first prompt token
+                    prefill_logprobs = [float("nan")] + torch.log_softmax(
+                        logits, -1
+                    ).gather(1, all_input_ids[1:]).squeeze(1)[
+                        -new_input_length:-1
+                    ].tolist()
+                    prefill_token_ids = all_input_ids[-new_input_length:-1]
+                    prefill_texts = self.tokenizer.batch_decode(
+                        prefill_token_ids,
+                        clean_up_tokenization_spaces=False,
+                        skip_special_tokens=False,
+                    )
+                    prefill_tokens = PrefillTokens(
+                        prefill_token_ids, prefill_logprobs, prefill_texts
+                    )
+                else:
+                    prefill_tokens = None
+
+                generation = Generation(
+                    request.id,
+                    prefill_tokens,
+                    next_token_id_squeezed,
+                    next_token_logprob,
+                    next_token_text,
+                    next_token_id_squeezed.item() in self.all_special_ids,
+                    generated_text,
+                )
+
+                generations.append(generation)
+
+            # Update values
+            batch.input_ids[i, 0] = next_token_id
+            batch.all_input_ids[i] = all_input_ids
+            batch.input_lengths[i] = new_input_length
+            batch.prefix_offsets[i] = prefix_offset
+            batch.read_offsets[i] = read_offset
+            batch.max_input_length = max(batch.max_input_length, new_input_length)
+
+        # We finished all generations in the batch; there is no next batch
+        if stopped:
+            return generations, None
+
+        # Slice unused values from prefill
+        batch.input_ids = batch.input_ids[:, :1]
+
+        # Update attention_mask as we added a new token to input_ids
+        batch.attention_mask[:, -batch.padding_right_offset] = 1
+        batch.image_attention_mask[:, -batch.padding_right_offset, :] = batch.image_attention_mask[:, -(batch.padding_right_offset+1), :]
+        # Decrease right offset
+        batch.padding_right_offset -= 1
+
+        # Update position_ids
+        batch.position_ids = batch.position_ids[:, -1:] + 1
+
+        # Update past key values
+        batch.past_key_values = past
+        batch.image_hidden_states = image_hidden_states
+
+        return generations, batch
diff --git a/server/text_generation_server/server.py b/server/text_generation_server/server.py
index 1cedc151..67137aaa 100644
--- a/server/text_generation_server/server.py
+++ b/server/text_generation_server/server.py
@@ -14,7 +14,7 @@ from text_generation_server.interceptor import ExceptionInterceptor
 from text_generation_server.models import Model, get_model
 from text_generation_server.pb import generate_pb2_grpc, generate_pb2
 from text_generation_server.tracing import UDSOpenTelemetryAioServerInterceptor
-
+from text_generation_server.models.idefics_causal_lm import IdeficsCausalLMBatch
 
 class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
     def __init__(self, model: Model, cache: Cache, server_urls: List[str]):
@@ -26,6 +26,7 @@ class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
             # Force inference mode for the lifetime of TextGenerationService
             self._inference_mode_raii_guard = torch._C._InferenceMode(True)
 
+
     async def Info(self, request, context):
         return self.model.info
 
@@ -54,9 +55,14 @@ class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
         return generate_pb2.FilterBatchResponse(batch=filtered_batch.to_pb())
 
     async def Warmup(self, request, context):
-        batch = self.model.batch_type.from_pb(
-            request.batch, self.model.tokenizer, self.model.dtype, self.model.device
-        )
+        if self.model.batch_type == IdeficsCausalLMBatch: #Hack, i would rather use kwargs in the `from_pb` call
+            batch = self.model.batch_type.from_pb(
+                request.batch, self.model.tokenizer, self.model.processor, self.model.dtype, self.model.device
+            )
+        else:
+            batch = self.model.batch_type.from_pb(
+                request.batch, self.model.tokenizer, self.model.dtype, self.model.device
+            )
         max_supported_total_tokens = self.model.warmup(batch)
 
         return generate_pb2.WarmupResponse(
@@ -64,9 +70,14 @@ class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
         )
 
     async def Prefill(self, request, context):
-        batch = self.model.batch_type.from_pb(
-            request.batch, self.model.tokenizer, self.model.dtype, self.model.device
-        )
+        if self.model.batch_type == IdeficsCausalLMBatch: #Hack, i would rather use kwargs in the `from_pb` call
+            batch = self.model.batch_type.from_pb(
+                request.batch, self.model.tokenizer, self.model.processor, self.model.dtype, self.model.device
+            )
+        else:
+            batch = self.model.batch_type.from_pb(
+                request.batch, self.model.tokenizer, self.model.dtype, self.model.device
+            )
 
         generations, next_batch = self.model.generate_token(batch)
         self.cache.set(next_batch)
diff --git a/server/text_generation_server/utils/layers.py b/server/text_generation_server/utils/layers.py
index 9688f6e0..745c1d2e 100644
--- a/server/text_generation_server/utils/layers.py
+++ b/server/text_generation_server/utils/layers.py
@@ -51,7 +51,31 @@ def load_layer_norm_no_bias(cls, prefix, weights, eps):
     ln.bias = None
     return ln
 
+@classmethod
+def load_conv2d(cls, prefix, weights, in_channels, out_channels, kernel_size, stride):
+    weight = weights.get_tensor(f"{prefix}.weight")
+    bias = weights.get_tensor(f"{prefix}.bias")
+    with init_empty_weights():
+        conv2d = cls(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride)
 
+    conv2d.weight = nn.Parameter(weight)
+    conv2d.bias = nn.Parameter(bias)
+    return conv2d
+
+
+@classmethod
+def load_conv2d_no_bias(cls, prefix, weights, in_channels, out_channels, kernel_size, stride):
+    weight = weights.get_tensor(f"{prefix}.weight")
+    with init_empty_weights():
+        conv2d = cls(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride)
+
+    conv2d.weight = nn.Parameter(weight)
+    conv2d.bias = None
+    return conv2d
+
+
+torch.nn.Conv2d.load = load_conv2d
+torch.nn.Conv2d.load_no_bias = load_conv2d_no_bias
 torch.nn.LayerNorm.load = load_layer_norm
 torch.nn.LayerNorm.load_no_bias = load_layer_norm_no_bias