remove one patch

Author: xadupre

_unittests/ut_torch_export_patches/test_patch_transformers.py

@@ -984,12 +984,12 @@ def test_prepare_inputs_for_generation_decoder_llm(self):
                 self.assertTrue(model_inputs["use_cache"] is True)
                 self.assertTrue(model_inputs["foo"] == "bar")
 
+            init_input_ids = input_ids[:, :2]
+            dynamic_cache = transformers.cache_utils.DynamicCache(config=config)
+            dynamic_cache = model(
+                init_input_ids, past_key_values=dynamic_cache
+            ).past_key_values
             with self.subTest(case="case5"):
-                init_input_ids = input_ids[:, :2]
-                dynamic_cache = transformers.cache_utils.DynamicCache(config=config)
-                dynamic_cache = model(
-                    init_input_ids, past_key_values=dynamic_cache
-                ).past_key_values
                 with self.assertRaises((AttributeError, TypeError)):
                     model_inputs = model.prepare_inputs_for_generation(
                         input_ids, past_key_values=dynamic_cache

onnx_diagnostic/torch_export_patches/patches/_patch_transformers_generation_mixin.py

@@ -1,6 +1,5 @@
 import inspect
-import os
-from typing import Optional, Tuple, Union
+from typing import Optional, Tuple
 import packaging.version as pv
 import torch
 import transformers
@@ -22,11 +21,11 @@ class patched_GenerationMixin:
             if pv.Version(transformers.__version__) >= pv.Version("4.56")
             else "prepare_inputs_for_generation"
         ),
-        (
-            "_sample"
-            if pv.Version(transformers.__version__) == pv.Version("4.57.0.dev0")
-            else None
-        ),
+        # (
+        #    "_sample"
+        #    if pv.Version(transformers.__version__) == pv.Version("4.57.0.dev0")
+        #    else None
+        # ),
     ]
     _PATCHED_CLASS_ = transformers.generation.utils.GenerationMixin
 
@@ -299,6 +298,8 @@ def prepare_inputs_for_generation(
         model_inputs.pop("labels", None)
         return model_inputs
 
+    '''
+    # drops a patch since it is for a very specific version.
     def _sample(
         self,
         input_ids: torch.LongTensor,
@@ -484,3 +485,4 @@ def _sample(
                 )
         else:
             return input_ids
+    '''