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>
35 lines
951 B
Python
35 lines
951 B
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
import pytest
|
|
import torch
|
|
|
|
from vllm import LLM
|
|
|
|
|
|
def test_empty_prompt():
|
|
llm = LLM(model="openai-community/gpt2", enforce_eager=True)
|
|
with pytest.raises(ValueError, match="decoder prompt cannot be empty"):
|
|
llm.generate([""])
|
|
|
|
|
|
def test_out_of_vocab_token():
|
|
llm = LLM(model="openai-community/gpt2", enforce_eager=True)
|
|
with pytest.raises(ValueError, match="out of vocabulary"):
|
|
llm.generate({"prompt_token_ids": [999999]})
|
|
|
|
|
|
def test_require_mm_embeds():
|
|
llm = LLM(
|
|
model="llava-hf/llava-1.5-7b-hf",
|
|
enforce_eager=True,
|
|
enable_mm_embeds=False,
|
|
)
|
|
with pytest.raises(ValueError, match="--enable-mm-embeds"):
|
|
llm.generate(
|
|
{
|
|
"prompt": "<image>",
|
|
"multi_modal_data": {"image": torch.empty(1, 1, 1)},
|
|
}
|
|
)
|