[Bugfix] Restore model info caching for package backends (#52690)

Signed-off-by: haoyang.qian <haoyangqian@gmail.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
hy123
2026-08-19 11:18:10 +00:00
committed by GitHub
co-authored by OpenAI Codex
parent cba06764d7
commit b160cab156
2 changed files with 32 additions and 1 deletions
+30
View File
@@ -173,6 +173,36 @@ def test_lazy_modelinfo_package_hash_includes_submodules(tmp_path):
assert first_hash != second_hash
def test_lazy_modelinfo_package_attempts_cache_load(monkeypatch):
cached_model_info = object()
loaded_hashes = []
def fake_load_cache(self, module_hash):
loaded_hashes.append(module_hash)
return cached_model_info
monkeypatch.setattr(
_LazyRegisteredModel,
"_load_modelinfo_from_cache",
fake_load_cache,
)
monkeypatch.setattr(
"vllm.model_executor.models.registry._run_in_subprocess",
lambda _: pytest.fail("Package-backed model should use the cache path"),
)
registered_model = _LazyRegisteredModel(
module_name="vllm.model_executor.models.transformers",
class_name="TransformersForCausalLM",
)
result = registered_model.inspect_model_cls()
assert result is cached_model_info
assert len(loaded_hashes) == 1
assert loaded_hashes[0]
def test_hf_registry_coverage():
untested_archs = (
ModelRegistry.get_supported_archs() - HF_EXAMPLE_MODELS.get_supported_archs()