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

[Misc] unify variable for LLM instance (#20996)

Browse Source 
Signed-off-by: Andy Xie <andy.xning@gmail.com>
This commit is contained in:
Ning Xie
2025-07-21 19:18:33 +08:00
committed by GitHub
parent e6b90a2805
commit d97841078b
53 changed files with 237 additions and 236 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
tests/conftest.py
View File

@@ -784,7 +784,7 @@ class VllmRunner:
enforce_eager: Optional[bool] = False,
**kwargs,
) -> None:
self.model = LLM(
self.llm = LLM(
model=model_name,
task=task,
tokenizer=tokenizer_name,
@@ -854,9 +854,9 @@ class VllmRunner:
videos=videos,
audios=audios)
req_outputs = self.model.generate(inputs,
sampling_params=sampling_params,
**kwargs)
req_outputs = self.llm.generate(inputs,
sampling_params=sampling_params,
**kwargs)
outputs: list[tuple[list[list[int]], list[str]]] = []
for req_output in req_outputs:
@@ -902,9 +902,9 @@ class VllmRunner:
videos=videos,
audios=audios)
req_outputs = self.model.generate(inputs,
sampling_params=sampling_params,
**kwargs)
req_outputs = self.llm.generate(inputs,
sampling_params=sampling_params,
**kwargs)
toks_str_logsprobs_prompt_logprobs = (
self._final_steps_generate_w_logprobs(req_outputs))
@@ -924,8 +924,8 @@ class VllmRunner:
'''
assert sampling_params.logprobs is not None
req_outputs = self.model.generate(encoder_decoder_prompts,
sampling_params=sampling_params)
req_outputs = self.llm.generate(encoder_decoder_prompts,
sampling_params=sampling_params)
toks_str_logsprobs_prompt_logprobs = (
self._final_steps_generate_w_logprobs(req_outputs))
# Omit prompt logprobs if not required by sampling params
@@ -1018,7 +1018,7 @@ class VllmRunner:
videos=videos,
audios=audios)
outputs = self.model.beam_search(
outputs = self.llm.beam_search(
inputs,
BeamSearchParams(beam_width=beam_width, max_tokens=max_tokens))
returned_outputs = []
@@ -1029,7 +1029,7 @@ class VllmRunner:
return returned_outputs
def classify(self, prompts: list[str]) -> list[list[float]]:
req_outputs = self.model.classify(prompts)
req_outputs = self.llm.classify(prompts)
return [req_output.outputs.probs for req_output in req_outputs]
def embed(self,
@@ -1044,11 +1044,11 @@ class VllmRunner:
videos=videos,
audios=audios)
req_outputs = self.model.embed(inputs, *args, **kwargs)
req_outputs = self.llm.embed(inputs, *args, **kwargs)
return [req_output.outputs.embedding for req_output in req_outputs]
def encode(self, prompts: list[str]) -> list[list[float]]:
req_outputs = self.model.encode(prompts)
req_outputs = self.llm.encode(prompts)
return [req_output.outputs.data for req_output in req_outputs]
def score(
@@ -1058,18 +1058,18 @@ class VllmRunner:
*args,
**kwargs,
) -> list[float]:
req_outputs = self.model.score(text_1, text_2, *args, **kwargs)
req_outputs = self.llm.score(text_1, text_2, *args, **kwargs)
return [req_output.outputs.score for req_output in req_outputs]
def apply_model(self, func: Callable[[nn.Module], _R]) -> list[_R]:
executor = self.model.llm_engine.model_executor
executor = self.llm.llm_engine.model_executor
return executor.apply_model(func)
def __enter__(self):
return self
def __exit__(self, exc_type, exc_value, traceback):
del self.model
del self.llm
cleanup_dist_env_and_memory()