20231088/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[BugFix] Fix test_phi3v.py (#5725)

Browse Source
This commit is contained in:
Chang Su 2024-06-20 21:45:34 -07:00 committed by GitHub
parent 1f5674218f
commit c35e4a3dd7
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 9 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tests/conftest.py
View File

@ -233,11 +233,13 @@ class HfRunner:
prompts: List[str], prompts: List[str],
max_tokens: int, max_tokens: int,
images: Optional[List[Image.Image]] = None, images: Optional[List[Image.Image]] = None,
**kwargs,
) -> List[Tuple[List[int], str]]: ) -> List[Tuple[List[int], str]]:
outputs = self.generate(prompts, outputs = self.generate(prompts,
do_sample=False, do_sample=False,
max_new_tokens=max_tokens, max_new_tokens=max_tokens,
images=images) images=images,
**kwargs)
return [(output_ids[0], output_str[0]) return [(output_ids[0], output_str[0])
for output_ids, output_str in outputs] for output_ids, output_str in outputs]

10
tests/models/test_phi3v.py
View File

@ -77,7 +77,7 @@ if is_cpu():
# numeric difference for longer context and test can't pass # numeric difference for longer context and test can't pass
@pytest.mark.parametrize("model_and_config", model_and_vl_config) @pytest.mark.parametrize("model_and_config", model_and_vl_config)
@pytest.mark.parametrize("dtype", [target_dtype]) @pytest.mark.parametrize("dtype", [target_dtype])
@pytest.mark.parametrize("max_tokens", [8]) @pytest.mark.parametrize("max_tokens", [128])
def test_models(hf_runner, vllm_runner, hf_images, vllm_images, def test_models(hf_runner, vllm_runner, hf_images, vllm_images,
model_and_config, dtype: str, max_tokens: int) -> None: model_and_config, dtype: str, max_tokens: int) -> None:
"""Inference result should be the same between hf and vllm. """Inference result should be the same between hf and vllm.
@ -95,9 +95,11 @@ def test_models(hf_runner, vllm_runner, hf_images, vllm_images,
hf_model_kwargs = {"_attn_implementation": "eager"} hf_model_kwargs = {"_attn_implementation": "eager"}
with hf_runner(model_id, dtype=dtype, with hf_runner(model_id, dtype=dtype,
model_kwargs=hf_model_kwargs) as hf_model: model_kwargs=hf_model_kwargs) as hf_model:
hf_outputs = hf_model.generate_greedy(HF_IMAGE_PROMPTS, hf_outputs = hf_model.generate_greedy(
max_tokens, HF_IMAGE_PROMPTS,
images=hf_images) max_tokens,
images=hf_images,
eos_token_id=hf_model.processor.tokenizer.eos_token_id)
vllm_image_prompts = [ vllm_image_prompts = [
p.replace("<|image_1|>", p.replace("<|image_1|>",