SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[Gemma3] compile ✨ (#37447)

Browse Source
This commit is contained in:
Joao Gante
2025-04-18 14:55:43 +01:00
committed by GitHub
parent a1b82563f1
commit e5ac23081e
10 changed files with 90 additions and 148 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/generation/test_utils.py
View File

@@ -2075,9 +2075,6 @@ class GenerationTesterMixin:
Tests that `.generate` is compatible with torch.compile without graph breaks, keeping the same results.
⚠️ Runs two sequential generations to ensure the cache doesn't get stuck after the first compiled run! ⚠️
"""
# Monkey-patching the HybridCache at test-time to continue testing compilation support
HybridCache.is_compileable = True
for model_class in self.all_generative_model_classes:
if not model_class._supports_static_cache:
self.skipTest("This model doesn't support static cache (= no expectations of compilation support)")
@@ -2174,9 +2171,6 @@ class GenerationTesterMixin:
Tests that all optional outputs are behaving as expected when compilation is triggered.
In essence, it's the same as `test_greedy_generate_dict_outputs`, but with automatic compilation triggered.
"""
# Monkey-patching the HybridCache at test-time to continue testing compilation support
HybridCache.is_compileable = True
for model_class in self.all_generative_model_classes:
if not model_class._supports_static_cache:
self.skipTest("This model doesn't support static cache (= no expectations of compilation support)")