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 ef88926c..93724fe4 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
@@ -11,22 +11,22 @@
       },
       {
         "id": 13,
-        "logprob": -2.734375,
+        "logprob": -2.59375,
         "text": ","
       },
       {
         "id": 8862,
-        "logprob": -3.6875,
+        "logprob": -3.5625,
         "text": " yellow"
       },
       {
         "id": 13,
-        "logprob": -0.40234375,
+        "logprob": -0.44726562,
         "text": ","
       },
       {
         "id": 209,
-        "logprob": -8.25,
+        "logprob": -8.0,
         "text": " "
       }
     ],
@@ -52,7 +52,7 @@
       },
       {
         "id": 9830,
-        "logprob": -2.25,
+        "logprob": -2.03125,
         "special": false,
         "text": " colors"
       },
@@ -64,13 +64,13 @@
       },
       {
         "id": 329,
-        "logprob": -2.171875,
+        "logprob": -2.734375,
         "special": false,
         "text": " A"
       },
       {
         "id": 1180,
-        "logprob": -2.046875,
+        "logprob": -2.0,
         "special": false,
         "text": " number"
       },
@@ -81,19 +81,19 @@
         "text": " of"
       },
       {
-        "id": 1027,
-        "logprob": -1.5546875,
+        "id": 253,
+        "logprob": -0.69140625,
         "special": false,
-        "text": " different"
+        "text": " the"
       },
       {
         "id": 3295,
-        "logprob": -0.97265625,
+        "logprob": -0.8203125,
         "special": false,
         "text": " color"
       }
     ],
     "top_tokens": null
   },
-  "generated_text": "blue, red, yellow, \nand blue colors. A number of different color"
+  "generated_text": "blue, red, yellow, \nand blue colors. A number of the color"
 }
diff --git a/integration-tests/models/test_flash_llama_exl2.py b/integration-tests/models/test_flash_llama_exl2.py
index 7169c999..1ee67695 100644
--- a/integration-tests/models/test_flash_llama_exl2.py
+++ b/integration-tests/models/test_flash_llama_exl2.py
@@ -21,7 +21,7 @@ async def flash_llama_exl2(flash_llama_exl2_handle):
     return flash_llama_exl2_handle.client
 
 
-@pytest.mark.release
+# @pytest.mark.release
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_llama_exl2(flash_llama_exl2, ignore_logprob_response_snapshot):
@@ -33,7 +33,7 @@ async def test_flash_llama_exl2(flash_llama_exl2, ignore_logprob_response_snapsh
     assert response == ignore_logprob_response_snapshot
 
 
-@pytest.mark.release
+# @pytest.mark.release
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_llama_exl2_all_params(
@@ -60,7 +60,7 @@ async def test_flash_llama_exl2_all_params(
     assert response == ignore_logprob_response_snapshot
 
 
-@pytest.mark.release
+# @pytest.mark.release
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_llama_exl2_load(
diff --git a/integration-tests/models/test_flash_llama_fp8.py b/integration-tests/models/test_flash_llama_fp8.py
index fe5df590..bc7458b7 100644
--- a/integration-tests/models/test_flash_llama_fp8.py
+++ b/integration-tests/models/test_flash_llama_fp8.py
@@ -13,7 +13,7 @@ async def flash_llama_fp8(flash_llama_fp8_handle):
     return flash_llama_fp8_handle.client
 
 
-@pytest.mark.release
+# @pytest.mark.release
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_llama_fp8(flash_llama_fp8, response_snapshot):
@@ -25,7 +25,7 @@ async def test_flash_llama_fp8(flash_llama_fp8, response_snapshot):
     assert response == response_snapshot
 
 
-@pytest.mark.release
+# @pytest.mark.release
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_llama_fp8_all_params(flash_llama_fp8, response_snapshot):
@@ -48,7 +48,7 @@ async def test_flash_llama_fp8_all_params(flash_llama_fp8, response_snapshot):
     assert response == response_snapshot
 
 
-@pytest.mark.release
+# @pytest.mark.release
 @pytest.mark.asyncio
 @pytest.mark.private
 async def test_flash_llama_fp8_load(flash_llama_fp8, generate_load, response_snapshot):