a88937f5cb
### What this PR does / why we need it?
When there is no kv cache in some devices, the `_compute_prefill_context
func` will return `None`, which is unexecpted. This PR replaces None
with full zeros/-inf tensors to avoid TypeError.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
```bash
pytest tests/e2e/multicard/4-cards/long_sequence/test_chunked_prefill.py -k test_models_chunked_prefill_with_empty_kvcache
```
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
---------
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
123 lines
3.8 KiB
Python
123 lines
3.8 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
# Copyright 2023 The vLLM team.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
# This file is a part of the vllm-ascend project.
|
|
# Adapted from vllm/tests/basic_correctness/test_basic_correctness.py
|
|
#
|
|
import os
|
|
import random
|
|
import string
|
|
from unittest.mock import patch
|
|
|
|
import pytest
|
|
from vllm import SamplingParams
|
|
|
|
from tests.e2e.conftest import VllmRunner
|
|
|
|
MODELS = [
|
|
"vllm-ascend/Qwen3-30B-A3B-W8A8",
|
|
"vllm-ascend/DeepSeek-V2-Lite-W8A8",
|
|
]
|
|
|
|
|
|
def generate_prompts(input_len, batchsize):
|
|
prompts = [
|
|
" ".join([
|
|
f"{random.choice(string.ascii_letters)}" for _ in range(input_len)
|
|
]) for _ in range(batchsize)
|
|
]
|
|
return prompts
|
|
|
|
|
|
@patch.dict(
|
|
os.environ, {
|
|
"HCCL_BUFFSIZE": "768",
|
|
"VLLM_ASCEND_ENABLE_FLASHCOMM1": "1",
|
|
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": "1"
|
|
})
|
|
@pytest.mark.parametrize("model", MODELS)
|
|
def test_models_chunked_prefill_mixed_length_prompts_including_1_token(
|
|
model: str):
|
|
TEST_ROPE_PARAMETERS = {
|
|
"rope_theta": 1000000,
|
|
"rope_type": "yarn",
|
|
"factor": 4,
|
|
"original_max_position_embeddings": 32768
|
|
}
|
|
prompts = [
|
|
generate_prompts(128 * 1024, 1)[0],
|
|
generate_prompts(1, 1)[0],
|
|
generate_prompts(9104, 1)[0],
|
|
]
|
|
sampling_params = SamplingParams(max_tokens=1, temperature=0.0)
|
|
|
|
with VllmRunner(
|
|
model,
|
|
enforce_eager=True,
|
|
max_num_seqs=2,
|
|
max_num_batched_tokens=131000,
|
|
max_model_len=132000,
|
|
tensor_parallel_size=2,
|
|
prefill_context_parallel_size=2,
|
|
decode_context_parallel_size=1,
|
|
enable_expert_parallel=True,
|
|
block_size=128,
|
|
quantization="ascend",
|
|
hf_overrides={"rope_parameters": TEST_ROPE_PARAMETERS},
|
|
) as runner:
|
|
runner.model.generate(prompts, sampling_params)
|
|
|
|
|
|
@patch.dict(
|
|
os.environ, {
|
|
"HCCL_BUFFSIZE": "768",
|
|
"VLLM_ASCEND_ENABLE_FLASHCOMM1": "1",
|
|
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": "1"
|
|
})
|
|
@pytest.mark.parametrize("model", MODELS)
|
|
def test_models_chunked_prefill_with_empty_kvcache(model: str):
|
|
TEST_ROPE_PARAMETERS = {
|
|
"rope_theta": 1000000,
|
|
"rope_type": "yarn",
|
|
"factor": 4,
|
|
"original_max_position_embeddings": 32768
|
|
}
|
|
# Note(qcs): we use chunk_size=50, kv_cache_interleave_size=128
|
|
# to simulate certain edge cases.
|
|
prompts = [
|
|
generate_prompts(128, 1)[0],
|
|
generate_prompts(1, 1)[0],
|
|
generate_prompts(130, 1)[0],
|
|
generate_prompts(51, 1)[0],
|
|
generate_prompts(129, 1)[0],
|
|
]
|
|
sampling_params = SamplingParams(max_tokens=1, temperature=0.0)
|
|
|
|
with VllmRunner(
|
|
model,
|
|
enforce_eager=True,
|
|
max_num_seqs=2,
|
|
tensor_parallel_size=2,
|
|
prefill_context_parallel_size=2,
|
|
decode_context_parallel_size=1,
|
|
enable_expert_parallel=True,
|
|
long_prefill_token_threshold=50,
|
|
block_size=128,
|
|
cp_kv_cache_interleave_size=128,
|
|
quantization="ascend",
|
|
hf_overrides={"rope_parameters": TEST_ROPE_PARAMETERS},
|
|
) as runner:
|
|
runner.model.generate(prompts, sampling_params)
|