EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[Test] Remove VLLM_USE_V1 in example and tests (#1733)

Browse Source 
V1 is enabled by default, no need to set it by hand now. This PR remove
the useless setting in example and tests

- vLLM version: v0.9.2
- vLLM main:
9ad0a4588b

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-07-15 12:49:57 +08:00
committed by GitHub
parent eb921d2b6f
commit 787010a637
29 changed files with 186 additions and 291 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
tests/e2e/singlecard/test_chunked.py
View File

@@ -20,8 +20,6 @@ Compare the outputs of vLLM with and without aclgraph.
Run `pytest tests/compile/test_aclgraph.py`.
"""
import os
import pytest
import torch
from vllm import LLM, SamplingParams
@@ -29,8 +27,6 @@ from vllm import LLM, SamplingParams
MODELS = ["deepseek-ai/DeepSeek-V2-Lite"]
@pytest.mark.skipif(os.getenv("VLLM_USE_V1") == "0",
reason="new chunked only support on v1")
@pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("max_tokens", [1])
def test_models(
@@ -39,36 +35,33 @@ def test_models(
monkeypatch: pytest.MonkeyPatch,
) -> None:
return
with monkeypatch.context() as m:
prompts = "The president of the United States is"
m.setenv("VLLM_USE_V1", "1")
prompts = "The president of the United States is"
sampling_params = SamplingParams(
max_tokens=max_tokens,
temperature=0.0,
)
sampling_params = SamplingParams(
max_tokens=max_tokens,
temperature=0.0,
)
vllm_model = LLM(model,
long_prefill_token_threshold=4,
enforce_eager=True)
output_chunked = vllm_model.generate(prompts, sampling_params)
logprobs_chunked = output_chunked.outputs[0].logprobs
del vllm_model
torch.npu.empty_cache()
vllm_model = LLM(model, long_prefill_token_threshold=4, enforce_eager=True)
output_chunked = vllm_model.generate(prompts, sampling_params)
logprobs_chunked = output_chunked.outputs[0].logprobs
del vllm_model
torch.npu.empty_cache()
vllm_model = LLM(model,
enforce_eager=True,
additional_config={
'ascend_scheduler_config': {
'enabled': True
},
})
output = vllm_model.generate(prompts, sampling_params)
logprobs = output.outputs[0].logprobs
del vllm_model
torch.npu.empty_cache()
vllm_model = LLM(model,
enforce_eager=True,
additional_config={
'ascend_scheduler_config': {
'enabled': True
},
})
output = vllm_model.generate(prompts, sampling_params)
logprobs = output.outputs[0].logprobs
del vllm_model
torch.npu.empty_cache()
logprobs_similarity = torch.cosine_similarity(
logprobs_chunked.flatten(), logprobs.flatten(), dim=0)
assert logprobs_similarity > 0.95
logprobs_similarity = torch.cosine_similarity(logprobs_chunked.flatten(),
logprobs.flatten(),
dim=0)
assert logprobs_similarity > 0.95