Gahow Wang
445e491123
third_party/vllm/ now tracked in git for direct patch management.
Based on vLLM v0.18.1 release with one patch applied:
vllm/v1/core/sched/scheduler.py:
Replace fatal assert with graceful skip when KV transfer callback
arrives for an already-aborted request during PD disaggregated serving.
Future vLLM modifications should be made directly in third_party/vllm/
and committed normally. The patches/ directory is kept as documentation
of what changed from upstream.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
42 lines
1.1 KiB
Python
42 lines
1.1 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
import pytest
|
|
|
|
from ...utils import check_logprobs_close
|
|
|
|
MODELS = [
|
|
# TODO(sang): Sliding window should be tested separately.
|
|
"ibm/PowerLM-3b",
|
|
"ibm/PowerMoE-3b",
|
|
]
|
|
|
|
|
|
@pytest.mark.parametrize("model", MODELS)
|
|
@pytest.mark.parametrize("dtype", ["bfloat16"])
|
|
@pytest.mark.parametrize("max_tokens", [64])
|
|
@pytest.mark.parametrize("num_logprobs", [5])
|
|
def test_models(
|
|
hf_runner,
|
|
vllm_runner,
|
|
example_prompts,
|
|
model: str,
|
|
dtype: str,
|
|
max_tokens: int,
|
|
num_logprobs: int,
|
|
) -> None:
|
|
with hf_runner(model, dtype=dtype) as hf_model:
|
|
hf_outputs = hf_model.generate_greedy_logprobs_limit(
|
|
example_prompts, max_tokens, num_logprobs
|
|
)
|
|
|
|
with vllm_runner(model, dtype=dtype) as vllm_model:
|
|
vllm_outputs = vllm_model.generate_greedy_logprobs(
|
|
example_prompts, max_tokens, num_logprobs
|
|
)
|
|
check_logprobs_close(
|
|
outputs_0_lst=hf_outputs,
|
|
outputs_1_lst=vllm_outputs,
|
|
name_0="hf",
|
|
name_1="vllm",
|
|
)
|