Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update whisper transformer module to 4.48.0 #24382

Draft
wants to merge 30 commits into
base: main
Choose a base branch
from
Draft

Conversation

jchen351
Copy link
Contributor

@jchen351 jchen351 commented Apr 10, 2025

Description

Motivation and Context

Branched off from #24291

@@ -7,6 +7,7 @@

import numpy as np
import torch
import transformers

Check notice

Code scanning / CodeQL

Module is imported with 'import' and 'import from' Note

Module 'transformers' is imported with both 'import' and 'import from'.
Module 'onnxruntime.test.python.transformers' is imported with both 'import' and 'import from'.

Copilot Autofix

AI 4 days ago

The best way to fix the problem is to remove the from transformers import AutoConfig, AutoTokenizer statement and use the transformers.AutoConfig and transformers.AutoTokenizer directly in the code. This approach maintains the existing functionality while eliminating the confusion caused by the dual import.

  • Remove the from transformers import AutoConfig, AutoTokenizer statement.
  • Replace all instances of AutoConfig and AutoTokenizer with transformers.AutoConfig and transformers.AutoTokenizer, respectively.
Suggested changeset 1
onnxruntime/python/tools/transformers/models/llama/llama_inputs.py

Autofix patch

Autofix patch
Run the following command in your local git repository to apply this patch
cat << 'EOF' | git apply
diff --git a/onnxruntime/python/tools/transformers/models/llama/llama_inputs.py b/onnxruntime/python/tools/transformers/models/llama/llama_inputs.py
--- a/onnxruntime/python/tools/transformers/models/llama/llama_inputs.py
+++ b/onnxruntime/python/tools/transformers/models/llama/llama_inputs.py
@@ -10,3 +10,2 @@
 import transformers
-from transformers import AutoConfig, AutoTokenizer
 
@@ -32,3 +31,3 @@
 def get_sample_inputs(
-    config: AutoConfig,
+    config: transformers.AutoConfig,
     device: torch.device,
@@ -67,3 +66,3 @@
 def get_sample_with_past_kv_inputs(
-    config: AutoConfig,
+    config: transformers.AutoConfig,
     device: torch.device,
EOF
@@ -10,3 +10,2 @@
import transformers
from transformers import AutoConfig, AutoTokenizer

@@ -32,3 +31,3 @@
def get_sample_inputs(
config: AutoConfig,
config: transformers.AutoConfig,
device: torch.device,
@@ -67,3 +66,3 @@
def get_sample_with_past_kv_inputs(
config: AutoConfig,
config: transformers.AutoConfig,
device: torch.device,
Copilot is powered by AI and may make mistakes. Always verify output.
import torch
import transformers

Check notice

Code scanning / CodeQL

Module is imported with 'import' and 'import from' Note

Module 'transformers' is imported with both 'import' and 'import from'.
Module 'onnxruntime.test.python.transformers' is imported with both 'import' and 'import from'.

Copilot Autofix

AI 4 days ago

To fix the problem, we should remove the from transformers import AutoConfig statement and use transformers.AutoConfig instead. This will ensure that the transformers module is only imported once, reducing confusion and potential namespace conflicts.

  • Remove the from transformers import AutoConfig statement.
  • Replace all instances of AutoConfig with transformers.AutoConfig.
Suggested changeset 1
onnxruntime/python/tools/transformers/models/llama/llama_parity.py

Autofix patch

Autofix patch
Run the following command in your local git repository to apply this patch
cat << 'EOF' | git apply
diff --git a/onnxruntime/python/tools/transformers/models/llama/llama_parity.py b/onnxruntime/python/tools/transformers/models/llama/llama_parity.py
--- a/onnxruntime/python/tools/transformers/models/llama/llama_parity.py
+++ b/onnxruntime/python/tools/transformers/models/llama/llama_parity.py
@@ -28,3 +28,3 @@
 from models.torch_export_patches.cache_helper import make_dynamic_cache
-from transformers import AutoConfig
+
 
@@ -35,3 +35,3 @@
 
-def get_sequence_lengths(args: argparse.Namespace, config: AutoConfig):
+def get_sequence_lengths(args: argparse.Namespace, config: transformers.AutoConfig):
     past_sequence_length, curr_sequence_length = (8, 1) if args.use_past_kv else (0, 8)
@@ -41,3 +41,3 @@
 
-def get_inputs(args: argparse.Namespace, config: AutoConfig):
+def get_inputs(args: argparse.Namespace, config: transformers.AutoConfig):
     # Dummy values for parity
@@ -104,3 +104,3 @@
     pytorch_model: None | torch.nn.Module = None,
-    config: None | AutoConfig = None,
+    config: None | transformers.AutoConfig = None,
 ):
EOF
@@ -28,3 +28,3 @@
from models.torch_export_patches.cache_helper import make_dynamic_cache
from transformers import AutoConfig


@@ -35,3 +35,3 @@

def get_sequence_lengths(args: argparse.Namespace, config: AutoConfig):
def get_sequence_lengths(args: argparse.Namespace, config: transformers.AutoConfig):
past_sequence_length, curr_sequence_length = (8, 1) if args.use_past_kv else (0, 8)
@@ -41,3 +41,3 @@

def get_inputs(args: argparse.Namespace, config: AutoConfig):
def get_inputs(args: argparse.Namespace, config: transformers.AutoConfig):
# Dummy values for parity
@@ -104,3 +104,3 @@
pytorch_model: None | torch.nn.Module = None,
config: None | AutoConfig = None,
config: None | transformers.AutoConfig = None,
):
Copilot is powered by AI and may make mistakes. Always verify output.
Comment on lines +10 to +19
def _catch_produce_guards_and_solve_constraints(
previous_function: Callable,
fake_mode: "FakeTensorMode",
gm: "torch.fx.GraphModule",
dynamic_shapes: dict[str, Any] | tuple[Any] | list[Any] | None,
equalities_inputs: "EqualityConstraint", # noqa: F821
original_signature: inspect.Signature,
_is_torch_jit_trace: bool = False,
verbose: int = 0,
):

Check notice

Code scanning / CodeQL

Explicit returns mixed with implicit (fall through) returns Note

Mixing implicit and explicit returns may indicate an error as implicit returns always return None.

Copilot Autofix

AI 4 days ago

To fix the problem, we need to add an explicit return statement at the end of the _catch_produce_guards_and_solve_constraints function. This will ensure that the function always returns a value, even when an exception is caught and the if conditions are not met. The explicit return statement should return None to maintain the existing functionality.

Suggested changeset 1
onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py

Autofix patch

Autofix patch
Run the following command in your local git repository to apply this patch
cat << 'EOF' | git apply
diff --git a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
--- a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
+++ b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
@@ -43,3 +43,3 @@
             )
-
+        return None
 
EOF
@@ -43,3 +43,3 @@
)

return None

Copilot is powered by AI and may make mistakes. Always verify output.
Comment on lines +46 to +52
def patch__check_input_constraints_for_graph(
previous_function: Callable,
input_placeholders: list[torch.fx.Node],
flat_args_with_path,
range_constraints,
verbose: int = 0,
) -> None:

Check notice

Code scanning / CodeQL

Explicit returns mixed with implicit (fall through) returns Note

Mixing implicit and explicit returns may indicate an error as implicit returns always return None.

Copilot Autofix

AI 4 days ago

To fix the problem, we need to add an explicit return statement at the end of the function patch__check_input_constraints_for_graph. This ensures that the function consistently returns a value, making the code easier to read and understand. The explicit return value should be None to maintain the existing functionality.

Suggested changeset 1
onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py

Autofix patch

Autofix patch
Run the following command in your local git repository to apply this patch
cat << 'EOF' | git apply
diff --git a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
--- a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
+++ b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
@@ -66,3 +66,3 @@
             )
-
+        return None
 
EOF
@@ -66,3 +66,3 @@
)

return None

Copilot is powered by AI and may make mistakes. Always verify output.
Comment on lines +304 to +306
# if config.print_specializations:
# self.log.warning(
# "Specializing %s to %s", self.var_to_sources[a][0].name(), tgt

Check notice

Code scanning / CodeQL

Commented-out code Note

This comment appears to contain commented-out code.

Copilot Autofix

AI 4 days ago

To fix the problem, we should remove the commented-out code. This will make the code cleaner and reduce potential confusion for future developers. If the logging statement is needed in the future, it can be reintroduced with proper documentation.

  • Remove the commented-out logging statement on lines 304-308.
  • Ensure that the removal does not affect the existing functionality of the code.
Suggested changeset 1
onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py

Autofix patch

Autofix patch
Run the following command in your local git repository to apply this patch
cat << 'EOF' | git apply
diff --git a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
--- a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
+++ b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_torch.py
@@ -303,7 +303,7 @@
 
-            # if config.print_specializations:
-            #    self.log.warning(
-            #         "Specializing %s to %s", self.var_to_sources[a][0].name(), tgt
-            #     )
-            #     self.log.debug("SPECIALIZATION", stack_info=True)
+
+
+
+
+
         assert msg != "range_refined_to_singleton", (
EOF
@@ -303,7 +303,7 @@

# if config.print_specializations:
# self.log.warning(
# "Specializing %s to %s", self.var_to_sources[a][0].name(), tgt
# )
# self.log.debug("SPECIALIZATION", stack_info=True)





assert msg != "range_refined_to_singleton", (
Copilot is powered by AI and may make mistakes. Always verify output.
Comment on lines +281 to +288
# if input_ids.shape[1] == 0:
# inputs_embeds = inputs_embeds[:, -cache_position.shape[0] :]
# else:
# if cache_position[-1] >= input_ids.shape[1]:
# input_ids = input_ids[:, -cache_position.shape[0] :]
# else:
# if input_ids.shape[1] != cache_position.shape[0]:
# input_ids = input_ids[:, cache_position]

Check notice

Code scanning / CodeQL

Commented-out code Note

This comment appears to contain commented-out code.

Copilot Autofix

AI 4 days ago

To fix the problem, we should remove the commented-out code. This will make the code cleaner and less confusing for future developers. The removal should be done in the _cache_dependant_input_preparation_exporting method, specifically lines 280 to 288.

Suggested changeset 1
onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_transformers.py

Autofix patch

Autofix patch
Run the following command in your local git repository to apply this patch
cat << 'EOF' | git apply
diff --git a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_transformers.py b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_transformers.py
--- a/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_transformers.py
+++ b/onnxruntime/python/tools/transformers/models/torch_export_patches/patches/patch_transformers.py
@@ -279,11 +279,3 @@
         else:
-            # This is the code we need to implemented with torch.cond.
-            # if input_ids.shape[1] == 0:
-            #     inputs_embeds = inputs_embeds[:, -cache_position.shape[0] :]
-            # else:
-            #     if cache_position[-1] >= input_ids.shape[1]:
-            #         input_ids = input_ids[:, -cache_position.shape[0] :]
-            #     else:
-            #         if input_ids.shape[1] != cache_position.shape[0]:
-            #             input_ids = input_ids[:, cache_position]
+            
             def branch_1(inputs_embeds, cache_position):
EOF
@@ -279,11 +279,3 @@
else:
# This is the code we need to implemented with torch.cond.
# if input_ids.shape[1] == 0:
# inputs_embeds = inputs_embeds[:, -cache_position.shape[0] :]
# else:
# if cache_position[-1] >= input_ids.shape[1]:
# input_ids = input_ids[:, -cache_position.shape[0] :]
# else:
# if input_ids.shape[1] != cache_position.shape[0]:
# input_ids = input_ids[:, cache_position]

def branch_1(inputs_embeds, cache_position):
Copilot is powered by AI and may make mistakes. Always verify output.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants