biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Fix handling of Tensorizer arguments for LoadConfig (#20643)

Browse Source 
Signed-off-by: Sanger Steel <sangersteel@gmail.com>
This commit is contained in:
Sanger Steel
2025-07-09 11:36:37 -04:00
committed by GitHub
parent efe73d0575
commit 4ac9c33f78
4 changed files with 21 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
tests/tensorizer_loader/test_tensorizer.py
View File

@@ -103,25 +103,6 @@ def write_keyfile(keyfile_path: str):
f.write(encryption_params.key)
@pytest.mark.skipif(not is_curl_installed(), reason="cURL is not installed")
def test_can_deserialize_s3(vllm_runner):
model_ref = "EleutherAI/pythia-1.4b"
tensorized_path = f"s3://tensorized/{model_ref}/fp16/model.tensors"
with vllm_runner(model_ref,
load_format="tensorizer",
model_loader_extra_config=TensorizerConfig(
tensorizer_uri=tensorized_path,
num_readers=1,
s3_endpoint="object.ord1.coreweave.com",
)) as loaded_hf_model:
deserialized_outputs = loaded_hf_model.generate(
prompts, sampling_params)
# noqa: E501
assert deserialized_outputs
@pytest.mark.skipif(not is_curl_installed(), reason="cURL is not installed")
def test_deserialized_encrypted_vllm_model_has_same_outputs(
model_ref, vllm_runner, tmp_path, model_path):