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

22
tests/prefix_caching/test_disable_sliding_window.py
View File

@@ -25,25 +25,25 @@ MODEL_LEN_LEN = [
@pytest.mark.parametrize("model_len_len", MODEL_LEN_LEN)
def test_disable_sliding_window(model_len_len, ):
model, sliding_len, full_len = model_len_len
vllm_disabled_model = LLM(model, disable_sliding_window=True)
vllm_disabled_model.generate("Hi my name is")
model_config = vllm_disabled_model.llm_engine.model_config
disabled_llm = LLM(model, disable_sliding_window=True)
disabled_llm.generate("Hi my name is")
model_config = disabled_llm.llm_engine.model_config
assert model_config.max_model_len == sliding_len, (
"Max len expected to equal sliding_len of %s, but got %s", sliding_len,
model_config.max_model_len)
del vllm_disabled_model
del disabled_llm
cleanup_dist_env_and_memory()
vllm_enabled_model = LLM(model,
enforce_eager=True,
disable_sliding_window=False,
enable_prefix_caching=False)
vllm_enabled_model.generate("Hi my name is")
model_config = vllm_enabled_model.llm_engine.model_config
enabled_llm = LLM(model,
enforce_eager=True,
disable_sliding_window=False,
enable_prefix_caching=False)
enabled_llm.generate("Hi my name is")
model_config = enabled_llm.llm_engine.model_config
assert model_config.max_model_len == full_len, (
"Max len expected to equal full_len of %s, but got %s", full_len,
model_config.max_model_len)
del vllm_enabled_model
del enabled_llm
cleanup_dist_env_and_memory()

6
tests/prefix_caching/test_prefix_caching.py
View File

@@ -93,8 +93,8 @@ def test_mixed_requests(
# Run all the promopts
greedy_params = SamplingParams(temperature=0.0,
max_tokens=max_tokens)
req_outputs = vllm_model.model.generate(example_prompts,
greedy_params)
req_outputs = vllm_model.llm.generate(example_prompts,
greedy_params)
# Verify number of cached tokens
for i in range(len(req_outputs)):
@@ -161,7 +161,7 @@ def test_fully_cached_prefill_needs_uncached_token(model):
max_num_batched_tokens=max_num_batched_tokens,
max_num_seqs=max_num_batched_tokens,
)
engine: LLMEngine = runner.model.llm_engine
engine: LLMEngine = runner.llm.llm_engine
scheduler: Scheduler = SchedulerProxy(engine.scheduler[0]) # type: ignore
engine.scheduler[0] = scheduler