diff --git a/integration-tests/models/__snapshots__/test_bloom_560m_sharded/test_bloom_560m_sharded.json b/integration-tests/models/__snapshots__/test_bloom_560m_sharded/test_bloom_560m_sharded.json
index dd8936af..a4f7269c 100644
--- a/integration-tests/models/__snapshots__/test_bloom_560m_sharded/test_bloom_560m_sharded.json
+++ b/integration-tests/models/__snapshots__/test_bloom_560m_sharded/test_bloom_560m_sharded.json
@@ -11,47 +11,47 @@
       },
       {
         "id": 49833,
-        "logprob": -10.5390625,
+        "logprob": -10.5859375,
         "text": " dég"
       },
       {
         "id": 21543,
-        "logprob": -0.14758301,
+        "logprob": -0.14794922,
         "text": "uster"
       },
       {
         "id": 447,
-        "logprob": -1.9296875,
+        "logprob": -1.9277344,
         "text": " un"
       },
       {
         "id": 46341,
-        "logprob": -15.4453125,
+        "logprob": -15.3203125,
         "text": " ort"
       },
       {
         "id": 35567,
-        "logprob": -7.59375,
+        "logprob": -7.5664062,
         "text": "olan"
       },
       {
         "id": 15,
-        "logprob": -1.3994141,
+        "logprob": -1.3974609,
         "text": ","
       },
       {
         "id": 1669,
-        "logprob": -1.578125,
+        "logprob": -1.5351562,
         "text": " il"
       },
       {
         "id": 11580,
-        "logprob": -0.9453125,
+        "logprob": -0.9423828,
         "text": " faut"
       },
       {
         "id": 3913,
-        "logprob": -3.7011719,
+        "logprob": -3.671875,
         "text": " tout"
       },
       {
@@ -60,69 +60,70 @@
         "text": " d'abord"
       }
     ],
-    "seed": 0,
+    "seed": 1,
     "tokens": [
       {
-        "id": 578,
-        "logprob": -1.6474609,
+        "id": 2997,
+        "logprob": -4.4101562,
         "special": false,
-        "text": " le"
+        "text": " vous"
       },
       {
-        "id": 5608,
-        "logprob": -2.5097656,
+        "id": 71256,
+        "logprob": -5.421875,
         "special": false,
-        "text": " faire"
+        "text": " retrouver"
       },
       {
-        "id": 159570,
-        "logprob": -6.65625,
+        "id": 693,
+        "logprob": -2.1738281,
         "special": false,
-        "text": " réch"
+        "text": " à"
       },
       {
-        "id": 810,
-        "logprob": 0.0,
+        "id": 366,
+        "logprob": -1.5322266,
         "special": false,
-        "text": "au"
+        "text": " la"
       },
       {
-        "id": 12736,
-        "logprob": 0.0,
+        "id": 221398,
+        "logprob": -2.6640625,
         "special": false,
-        "text": "ffer"
+        "text": " terrasse"
       },
       {
-        "id": 1742,
-        "logprob": -2.5859375,
+        "id": 1375,
+        "logprob": -4.375,
         "special": false,
-        "text": " au"
+        "text": " pour"
       },
       {
-        "id": 6105,
-        "logprob": -2.03125,
+        "id": 86887,
+        "logprob": -4.8203125,
         "special": false,
-        "text": " bain"
+        "text": " essayer"
       },
       {
-        "id": 88254,
-        "logprob": -0.12695312,
+        "id": 2155,
+        "logprob": -2.7421875,
         "special": false,
-        "text": "-mar"
+        "text": " ce"
       },
       {
-        "id": 641,
-        "logprob": 0.0,
+        "id": 5743,
+        "logprob": -3.1757812,
         "special": false,
-        "text": "ie"
+        "text": " jus"
       },
       {
-        "id": 2940,
-        "logprob": -3.5175781,
+        "id": 1479,
+        "logprob": -3.3554688,
         "special": false,
-        "text": " avec"
+        "text": " qui"
       }
-    ]
+    ],
+    "top_tokens": null
   },
-  "generated_text": " le faire réchauffer au bain-marie avec"
+  "generated_text": " vous retrouver à la terrasse pour essayer ce jus qui"
 }
diff --git a/integration-tests/models/__snapshots__/test_flash_gemma/test_flash_gemma_all_params.json b/integration-tests/models/__snapshots__/test_flash_gemma/test_flash_gemma_all_params.json
index 8253dc96..0286e3da 100644
--- a/integration-tests/models/__snapshots__/test_flash_gemma/test_flash_gemma_all_params.json
+++ b/integration-tests/models/__snapshots__/test_flash_gemma/test_flash_gemma_all_params.json
@@ -20,7 +20,7 @@
         "text": " request"
       }
     ],
-    "seed": 0,
+    "seed": 1,
     "tokens": [
       {
         "id": 7539,
@@ -35,20 +35,20 @@
         "text": " are"
       },
       {
-        "id": 671,
-        "logprob": -1.703125,
+        "id": 476,
+        "logprob": -0.453125,
         "special": false,
-        "text": " an"
+        "text": " a"
       },
       {
-        "id": 8727,
-        "logprob": 0.0,
+        "id": 2621,
+        "logprob": -2.078125,
         "special": false,
-        "text": " essential"
+        "text": " key"
       },
       {
         "id": 1702,
-        "logprob": 0.0,
+        "logprob": -0.20117188,
         "special": false,
         "text": " part"
       },
@@ -64,26 +64,26 @@
         "special": false,
         "text": " the"
       },
-      {
-        "id": 11859,
-        "logprob": -1.6953125,
-        "special": false,
-        "text": " lab"
-      },
       {
         "id": 2185,
-        "logprob": -1.3125,
+        "logprob": -0.765625,
         "special": false,
         "text": " process"
       },
       {
-        "id": 578,
-        "logprob": -1.5,
+        "id": 604,
+        "logprob": 0.0,
         "special": false,
-        "text": " and"
+        "text": " for"
+      },
+      {
+        "id": 17583,
+        "logprob": -2.609375,
+        "special": false,
+        "text": " sending"
       }
     ],
     "top_tokens": null
   },
-  "generated_text": "Test request forms are an essential part of the lab process and"
+  "generated_text": "Test request forms are a key part of the process for sending"
 }
diff --git a/integration-tests/models/__snapshots__/test_flash_phi/test_flash_phi_all_params.json b/integration-tests/models/__snapshots__/test_flash_phi/test_flash_phi_all_params.json
index 221ff13d..ac0ecab6 100644
--- a/integration-tests/models/__snapshots__/test_flash_phi/test_flash_phi_all_params.json
+++ b/integration-tests/models/__snapshots__/test_flash_phi/test_flash_phi_all_params.json
@@ -1,8 +1,8 @@
 {
   "details": {
     "best_of_sequences": null,
-    "finish_reason": "stop_sequence",
-    "generated_tokens": 6,
+    "finish_reason": "length",
+    "generated_tokens": 10,
     "prefill": [
       {
         "id": 14402,
@@ -15,7 +15,7 @@
         "text": " request"
       }
     ],
-    "seed": 0,
+    "seed": 1,
     "tokens": [
       {
         "id": 284,
@@ -24,37 +24,61 @@
         "text": " to"
       },
       {
-        "id": 3758,
-        "logprob": -0.62597656,
+        "id": 262,
+        "logprob": -2.0957031,
         "special": false,
-        "text": " send"
+        "text": " the"
       },
       {
-        "id": 1366,
-        "logprob": -0.87060547,
+        "id": 7824,
+        "logprob": 0.0,
         "special": false,
-        "text": " data"
+        "text": " API"
       },
       {
-        "id": 625,
-        "logprob": -0.88427734,
+        "id": 198,
+        "logprob": -1.4794922,
         "special": false,
-        "text": " over"
+        "text": "\n"
       },
       {
-        "id": 257,
-        "logprob": -1.0830078,
+        "id": 50280,
+        "logprob": 0.0,
         "special": false,
-        "text": " a"
+        "text": "        "
       },
       {
-        "id": 3127,
-        "logprob": -1.9462891,
+        "id": 37811,
+        "logprob": 0.0,
         "special": false,
-        "text": " network"
+        "text": "\"\"\""
+      },
+      {
+        "id": 628,
+        "logprob": 0.0,
+        "special": false,
+        "text": "\n\n"
+      },
+      {
+        "id": 50284,
+        "logprob": 0.0,
+        "special": false,
+        "text": "    "
+      },
+      {
+        "id": 4299,
+        "logprob": 0.0,
+        "special": false,
+        "text": "def"
+      },
+      {
+        "id": 1332,
+        "logprob": -1.7158203,
+        "special": false,
+        "text": " test"
       }
     ],
     "top_tokens": null
   },
-  "generated_text": "Test request to send data over a network"
+  "generated_text": "Test request to the API\n        \"\"\"\n\n    def test"
 }
diff --git a/integration-tests/models/__snapshots__/test_mamba/test_mamba_all_params.json b/integration-tests/models/__snapshots__/test_mamba/test_mamba_all_params.json
index 85e9a9e0..079bb13a 100644
--- a/integration-tests/models/__snapshots__/test_mamba/test_mamba_all_params.json
+++ b/integration-tests/models/__snapshots__/test_mamba/test_mamba_all_params.json
@@ -30,7 +30,7 @@
         "text": " "
       }
     ],
-    "seed": 0,
+    "seed": 1,
     "tokens": [
       {
         "id": 187,
@@ -39,22 +39,22 @@
         "text": "\n"
       },
       {
-        "id": 395,
-        "logprob": -0.3125,
+        "id": 11863,
+        "logprob": -1.3125,
         "special": false,
-        "text": "and"
+        "text": "blue"
       },
       {
-        "id": 4797,
+        "id": 285,
+        "logprob": -1.1015625,
+        "special": false,
+        "text": " and"
+      },
+      {
+        "id": 4759,
         "logprob": 0.0,
         "special": false,
-        "text": " blue"
-      },
-      {
-        "id": 9830,
-        "logprob": -1.65625,
-        "special": false,
-        "text": " colors"
+        "text": " green"
       },
       {
         "id": 15,
@@ -63,16 +63,16 @@
         "text": "."
       },
       {
-        "id": 329,
-        "logprob": -2.4375,
+        "id": 380,
+        "logprob": -1.65625,
         "special": false,
-        "text": " A"
+        "text": " The"
       },
       {
-        "id": 1180,
-        "logprob": -1.953125,
+        "id": 3295,
+        "logprob": -0.45117188,
         "special": false,
-        "text": " number"
+        "text": " color"
       },
       {
         "id": 273,
@@ -81,19 +81,19 @@
         "text": " of"
       },
       {
-        "id": 1027,
-        "logprob": -1.5546875,
+        "id": 253,
+        "logprob": 0.0,
         "special": false,
-        "text": " different"
+        "text": " the"
       },
       {
-        "id": 3295,
-        "logprob": -0.97265625,
+        "id": 2329,
+        "logprob": -3.125,
         "special": false,
-        "text": " color"
+        "text": " air"
       }
     ],
     "top_tokens": null
   },
-  "generated_text": "blue, red, yellow, \nand blue colors. A number of different color"
+  "generated_text": "blue, red, yellow, \nblue and green. The color of the air"
 }
diff --git a/integration-tests/models/test_flash_gemma.py b/integration-tests/models/test_flash_gemma.py
index 89d1b2e3..d167b813 100644
--- a/integration-tests/models/test_flash_gemma.py
+++ b/integration-tests/models/test_flash_gemma.py
@@ -3,7 +3,7 @@ import pytest
 
 @pytest.fixture(scope="module")
 def flash_gemma_handle(launcher):
-    with launcher("gg-hf/gemma-2b", num_shard=1) as handle:
+    with launcher("google/gemma-2b", num_shard=1) as handle:
         yield handle
 
 
@@ -13,7 +13,7 @@ async def flash_gemma(flash_gemma_handle):
     return flash_gemma_handle.client
 
 
-# @pytest.mark.skip
+@pytest.mark.skip
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_gemma(flash_gemma, response_snapshot):
@@ -25,7 +25,7 @@ async def test_flash_gemma(flash_gemma, response_snapshot):
     assert response == response_snapshot
 
 
-# @pytest.mark.skip
+@pytest.mark.skip
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_gemma_all_params(flash_gemma, response_snapshot):
diff --git a/integration-tests/models/test_flash_phi.py b/integration-tests/models/test_flash_phi.py
index 679e1ac8..2dbc5f04 100644
--- a/integration-tests/models/test_flash_phi.py
+++ b/integration-tests/models/test_flash_phi.py
@@ -42,8 +42,11 @@ async def test_flash_phi_all_params(flash_phi, response_snapshot):
         seed=1,
     )
 
-    assert response.details.generated_tokens == 6
-    assert response.generated_text == "Test request to send data over a network"
+    assert response.details.generated_tokens == 10
+    assert (
+        response.generated_text
+        == 'Test request to the API\n        """\n\n    def test'
+    )
     assert response == response_snapshot
 
 
diff --git a/integration-tests/models/test_mamba.py b/integration-tests/models/test_mamba.py
index f8a8b00f..e50a75ff 100644
--- a/integration-tests/models/test_mamba.py
+++ b/integration-tests/models/test_mamba.py
@@ -45,7 +45,7 @@ async def test_mamba_all_params(fused_kernel_mamba, response_snapshot):
     assert response.details.generated_tokens == 10
     assert (
         response.generated_text
-        == "blue, red, yellow, \nand blue colors. A number of different color"
+        == "blue, red, yellow, \nblue and green. The color of the air"
     )
     assert response == response_snapshot
 
diff --git a/integration-tests/models/test_mt0_base.py b/integration-tests/models/test_mt0_base.py
index 3a51be69..2ace787d 100644
--- a/integration-tests/models/test_mt0_base.py
+++ b/integration-tests/models/test_mt0_base.py
@@ -23,7 +23,7 @@ async def test_mt0_base(mt0_base, response_snapshot):
         seed=1,
     )
 
-    assert response.details.generated_tokens == 5
+    assert response.details.generated_tokens == 7
     assert response == response_snapshot