Some checks failed
Self-hosted runner (nightly-past-ci-caller) / Get number (push) Has been cancelled
Self-hosted runner (nightly-past-ci-caller) / TensorFlow 2.11 (push) Has been cancelled
Self-hosted runner (nightly-past-ci-caller) / TensorFlow 2.10 (push) Has been cancelled
Self-hosted runner (nightly-past-ci-caller) / TensorFlow 2.9 (push) Has been cancelled
Self-hosted runner (nightly-past-ci-caller) / TensorFlow 2.8 (push) Has been cancelled
Self-hosted runner (nightly-past-ci-caller) / TensorFlow 2.7 (push) Has been cancelled
Self-hosted runner (nightly-past-ci-caller) / TensorFlow 2.6 (push) Has been cancelled
Self-hosted runner (nightly-past-ci-caller) / TensorFlow 2.5 (push) Has been cancelled
Self-hosted runner (benchmark) / Benchmark (aws-g5-4xlarge-cache) (push) Has been cancelled
Build documentation / build (push) Has been cancelled
Build documentation / build_other_lang (push) Has been cancelled
CodeQL Security Analysis / CodeQL Analysis (push) Has been cancelled
New model PR merged notification / Notify new model (push) Has been cancelled
PR CI / pr-ci (push) Has been cancelled
Slow tests on important models (on Push - A10) / Get all modified files (push) Has been cancelled
Secret Leaks / trufflehog (push) Has been cancelled
Update Transformers metadata / build_and_package (push) Has been cancelled
Slow tests on important models (on Push - A10) / Model CI (push) Has been cancelled
Check Tiny Models / Check tiny models (push) Has been cancelled
Self-hosted runner (Intel Gaudi3 scheduled CI caller) / Model CI (push) Has been cancelled
Self-hosted runner (Intel Gaudi3 scheduled CI caller) / Pipeline CI (push) Has been cancelled
Self-hosted runner (Intel Gaudi3 scheduled CI caller) / Example CI (push) Has been cancelled
Self-hosted runner (Intel Gaudi3 scheduled CI caller) / DeepSpeed CI (push) Has been cancelled
Self-hosted runner (Intel Gaudi3 scheduled CI caller) / Trainer/FSDP CI (push) Has been cancelled
Nvidia CI - Flash Attn / Setup (push) Has been cancelled
Nvidia CI - Flash Attn / Model CI (push) Has been cancelled
Nvidia CI / Setup (push) Has been cancelled
Nvidia CI / Model CI (push) Has been cancelled
Nvidia CI / Torch pipeline CI (push) Has been cancelled
Nvidia CI / Example CI (push) Has been cancelled
Nvidia CI / Trainer/FSDP CI (push) Has been cancelled
Nvidia CI / DeepSpeed CI (push) Has been cancelled
Nvidia CI / Quantization CI (push) Has been cancelled
Nvidia CI / Kernels CI (push) Has been cancelled
Doctests / Setup (push) Has been cancelled
Doctests / Call doctest jobs (push) Has been cancelled
Doctests / Send results to webhook (push) Has been cancelled
Extras Smoke Test / Get supported Python versions (push) Has been cancelled
Extras Smoke Test / Test extras on Python ${{ matrix.python-version }} (push) Has been cancelled
Extras Smoke Test / Check Slack token availability (push) Has been cancelled
Extras Smoke Test / Notify failures to Slack (push) Has been cancelled
Self-hosted runner (AMD scheduled CI caller) / Trigger Scheduled AMD CI (push) Has been cancelled
Stale Bot / Close Stale Issues (push) Has been cancelled
358 lines
12 KiB
Python
358 lines
12 KiB
Python
# Copyright 2026 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.
|
|
|
|
import sys
|
|
import threading
|
|
import unittest
|
|
|
|
from transformers import is_torch_available
|
|
from transformers.monkey_patching import (
|
|
apply_patches,
|
|
clear_patch_mapping,
|
|
get_patch_mapping,
|
|
patch_output_recorders,
|
|
register_patch_mapping,
|
|
unregister_patch_mapping,
|
|
)
|
|
from transformers.testing_utils import require_torch
|
|
from transformers.utils.output_capturing import OutputRecorder
|
|
|
|
|
|
if is_torch_available():
|
|
import torch.nn as nn
|
|
|
|
|
|
@require_torch
|
|
class MonkeyPatchTest(unittest.TestCase):
|
|
def setUp(self):
|
|
"""Clear any existing patches before each test."""
|
|
clear_patch_mapping()
|
|
|
|
def tearDown(self):
|
|
"""Clean up patches after each test."""
|
|
clear_patch_mapping()
|
|
|
|
def test_register_patch_mapping(self):
|
|
"""Test basic registration of monkey patches."""
|
|
|
|
class CustomModule(nn.Module):
|
|
pass
|
|
|
|
# Register a patch
|
|
register_patch_mapping(mapping={"TestModule": CustomModule})
|
|
|
|
# Verify it was registered
|
|
mapping = get_patch_mapping()
|
|
self.assertIn("TestModule", mapping)
|
|
self.assertEqual(mapping["TestModule"], CustomModule)
|
|
|
|
def test_register_duplicate_with_overwrite(self):
|
|
"""Test that registering a duplicate class with overwrite=True works."""
|
|
|
|
class CustomModule1(nn.Module):
|
|
pass
|
|
|
|
class CustomModule2(nn.Module):
|
|
pass
|
|
|
|
# Register initial patch
|
|
register_patch_mapping(mapping={"TestModule": CustomModule1})
|
|
|
|
# Overwrite with new patch
|
|
register_patch_mapping(mapping={"TestModule": CustomModule2}, overwrite=True)
|
|
|
|
# Verify the new patch is registered
|
|
mapping = get_patch_mapping()
|
|
self.assertEqual(mapping["TestModule"], CustomModule2)
|
|
|
|
def test_register_non_nn_module_raises_error(self):
|
|
"""Test that registering a non-nn.Module class raises TypeError."""
|
|
|
|
class NotAModule:
|
|
pass
|
|
|
|
# Try to register a non-nn.Module class
|
|
with self.assertRaises(TypeError) as context:
|
|
register_patch_mapping(mapping={"TestModule": NotAModule})
|
|
|
|
self.assertIn("must be a subclass of nn.Module", str(context.exception))
|
|
|
|
def test_unregister_patch_mapping(self):
|
|
"""Test unregistering monkey patches."""
|
|
|
|
class CustomModule(nn.Module):
|
|
pass
|
|
|
|
# Register and then unregister
|
|
register_patch_mapping(mapping={"TestModule": CustomModule})
|
|
unregister_patch_mapping(["TestModule"])
|
|
|
|
# Verify it was unregistered
|
|
mapping = get_patch_mapping()
|
|
self.assertNotIn("TestModule", mapping)
|
|
|
|
def test_unregister_nonexistent_class(self):
|
|
"""Test unregistering a class that doesn't exist raises an error."""
|
|
# This should raise an error
|
|
with self.assertRaises(ValueError) as context:
|
|
unregister_patch_mapping(["NonexistentModule"])
|
|
|
|
self.assertIn("not found in monkey patch mapping cache", str(context.exception))
|
|
self.assertIn("Cannot unregister", str(context.exception))
|
|
|
|
def test_clear_patch_mapping(self):
|
|
"""Test clearing all monkey patches."""
|
|
|
|
class CustomModule1(nn.Module):
|
|
pass
|
|
|
|
class CustomModule2(nn.Module):
|
|
pass
|
|
|
|
# Register multiple patches
|
|
register_patch_mapping(mapping={"TestModule1": CustomModule1, "TestModule2": CustomModule2})
|
|
|
|
# Clear all patches
|
|
clear_patch_mapping()
|
|
|
|
# Verify all were cleared
|
|
mapping = get_patch_mapping()
|
|
self.assertEqual(len(mapping), 0)
|
|
|
|
def test_get_patch_mapping_returns_copy(self):
|
|
"""Test that get_patch_mapping returns a copy, not the original."""
|
|
|
|
class CustomModule(nn.Module):
|
|
pass
|
|
|
|
register_patch_mapping(mapping={"TestModule": CustomModule})
|
|
|
|
# Get mapping and modify it
|
|
mapping = get_patch_mapping()
|
|
mapping["NewModule"] = CustomModule
|
|
|
|
# Verify the internal cache was not modified
|
|
internal_mapping = get_patch_mapping()
|
|
self.assertNotIn("NewModule", internal_mapping)
|
|
|
|
def test_apply_patches_context_manager(self):
|
|
"""Test that apply_patches context manager works correctly."""
|
|
|
|
class CustomLinear(nn.Linear):
|
|
pass
|
|
|
|
# Create a dummy module in transformers namespace for testing
|
|
import types
|
|
|
|
test_module = types.ModuleType("transformers.test_module")
|
|
test_module.Linear = nn.Linear
|
|
sys.modules["transformers.test_module"] = test_module
|
|
|
|
try:
|
|
# Register patch
|
|
register_patch_mapping(mapping={"Linear": CustomLinear})
|
|
|
|
# Outside context, original class should be used
|
|
self.assertEqual(test_module.Linear, nn.Linear)
|
|
|
|
# Inside context, patched class should be used
|
|
with apply_patches():
|
|
self.assertEqual(test_module.Linear, CustomLinear)
|
|
|
|
# Outside context again, original class should be restored
|
|
self.assertEqual(test_module.Linear, nn.Linear)
|
|
|
|
finally:
|
|
# Clean up the test module
|
|
del sys.modules["transformers.test_module"]
|
|
|
|
def test_thread_safety_concurrent_access(self):
|
|
"""Test that concurrent reads and writes are thread-safe."""
|
|
|
|
class CustomModule(nn.Module):
|
|
pass
|
|
|
|
results = []
|
|
|
|
def read_mapping():
|
|
for _ in range(100):
|
|
mapping = get_patch_mapping()
|
|
results.append(len(mapping))
|
|
|
|
def write_mapping():
|
|
for i in range(100):
|
|
mapping = {f"Module{i}": CustomModule}
|
|
register_patch_mapping(mapping=mapping)
|
|
|
|
# Create threads for reading and writing
|
|
read_thread = threading.Thread(target=read_mapping)
|
|
write_thread = threading.Thread(target=write_mapping)
|
|
|
|
read_thread.start()
|
|
write_thread.start()
|
|
|
|
read_thread.join()
|
|
write_thread.join()
|
|
|
|
# Test should complete without deadlocks or errors
|
|
self.assertEqual(len(results), 100)
|
|
|
|
def test_patch_output_recorders_with_output_recorder_instance(self):
|
|
"""Test patching output recorders that are OutputRecorder instances."""
|
|
|
|
class OriginalModule(nn.Module):
|
|
pass
|
|
|
|
class ReplacementModule(nn.Module):
|
|
pass
|
|
|
|
class TestModel(nn.Module):
|
|
# Simulate _can_record_outputs with OutputRecorder
|
|
_can_record_outputs = {"output": OutputRecorder(OriginalModule)}
|
|
|
|
def __init__(self):
|
|
super().__init__()
|
|
self.linear = nn.Linear(10, 10)
|
|
|
|
model = TestModel()
|
|
|
|
# Register patch
|
|
register_patch_mapping(mapping={"OriginalModule": ReplacementModule})
|
|
|
|
# Patch output recorders
|
|
patch_output_recorders(model)
|
|
|
|
# Verify the recorder's target_class was updated
|
|
recorder = model._can_record_outputs["output"]
|
|
self.assertEqual(recorder.target_class, ReplacementModule)
|
|
|
|
def test_patch_output_recorders_with_class_type(self):
|
|
"""Test patching output recorders that are class types directly."""
|
|
|
|
class OriginalModule(nn.Module):
|
|
pass
|
|
|
|
class ReplacementModule(nn.Module):
|
|
pass
|
|
|
|
class TestModel(nn.Module):
|
|
# Simulate _can_record_outputs with class type directly
|
|
_can_record_outputs = {"output": OriginalModule}
|
|
|
|
def __init__(self):
|
|
super().__init__()
|
|
self.linear = nn.Linear(10, 10)
|
|
|
|
model = TestModel()
|
|
|
|
# Register patch
|
|
register_patch_mapping(mapping={"OriginalModule": ReplacementModule})
|
|
|
|
# Patch output recorders
|
|
patch_output_recorders(model)
|
|
|
|
# Verify the class was updated
|
|
self.assertEqual(model._can_record_outputs["output"], ReplacementModule)
|
|
|
|
def test_pattern_matching_wildcard(self):
|
|
"""Test pattern matching with .* regex (matches any characters)."""
|
|
|
|
class CustomAttention(nn.Module):
|
|
pass
|
|
|
|
# Register with pattern
|
|
register_patch_mapping(mapping={".*Attention": CustomAttention})
|
|
|
|
# Create test modules with different attention classes
|
|
import types
|
|
|
|
test_module = types.ModuleType("transformers.test_pattern")
|
|
test_module.BertAttention = nn.Module
|
|
test_module.GPT2Attention = nn.Module
|
|
test_module.T5Attention = nn.Module
|
|
test_module.SomethingElse = nn.Module
|
|
sys.modules["transformers.test_pattern"] = test_module
|
|
|
|
try:
|
|
with apply_patches():
|
|
# All *Attention classes should be patched
|
|
self.assertEqual(test_module.BertAttention, CustomAttention)
|
|
self.assertEqual(test_module.GPT2Attention, CustomAttention)
|
|
self.assertEqual(test_module.T5Attention, CustomAttention)
|
|
# Non-matching class should not be patched
|
|
self.assertNotEqual(test_module.SomethingElse, CustomAttention)
|
|
finally:
|
|
del sys.modules["transformers.test_pattern"]
|
|
|
|
def test_exact_match_precedence_over_pattern(self):
|
|
"""Test that exact matches take precedence over patterns."""
|
|
|
|
class PatternReplacement(nn.Module):
|
|
pass
|
|
|
|
class ExactReplacement(nn.Module):
|
|
pass
|
|
|
|
# Register both pattern and exact match
|
|
register_patch_mapping(mapping={".*Attention": PatternReplacement})
|
|
register_patch_mapping(mapping={"BertAttention": ExactReplacement})
|
|
|
|
import types
|
|
|
|
test_module = types.ModuleType("transformers.test_precedence")
|
|
test_module.BertAttention = nn.Module
|
|
test_module.GPT2Attention = nn.Module
|
|
sys.modules["transformers.test_precedence"] = test_module
|
|
|
|
try:
|
|
with apply_patches():
|
|
# Exact match should take precedence
|
|
self.assertEqual(test_module.BertAttention, ExactReplacement)
|
|
# Pattern should still match others
|
|
self.assertEqual(test_module.GPT2Attention, PatternReplacement)
|
|
finally:
|
|
del sys.modules["transformers.test_precedence"]
|
|
|
|
def test_pattern_with_output_recorders(self):
|
|
"""Test pattern matching works with output recorders."""
|
|
|
|
class OriginalAttention(nn.Module):
|
|
pass
|
|
|
|
class ReplacementAttention(nn.Module):
|
|
pass
|
|
|
|
class TestModel(nn.Module):
|
|
_can_record_outputs = {"output": OutputRecorder(OriginalAttention)}
|
|
|
|
def __init__(self):
|
|
super().__init__()
|
|
self.linear = nn.Linear(10, 10)
|
|
|
|
model = TestModel()
|
|
|
|
# Register with pattern
|
|
register_patch_mapping(mapping={".*Attention": ReplacementAttention})
|
|
|
|
# Patch output recorders
|
|
patch_output_recorders(model)
|
|
|
|
# Verify the recorder's target_class was updated via pattern matching
|
|
recorder = model._can_record_outputs["output"]
|
|
self.assertEqual(recorder.target_class, ReplacementAttention)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|