7bdc606677
Contains on #1111 for completeness. <!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> ### What this PR does / why we need it? Implement multi-stream parallelism for MoE layers with shared experts, where computation of shared experts will be overlapped with expert token dispatch and combine. Also, when multi-stream is enabled, weights of shared experts will be force to replicate across all cards, regardless of any tensor parallelism configurations, to avoid AllReduce operations. With the expected overlaping being: ``` | shared gate_up | shared act | | shared down | | dispatch | routed gate_up, act, down | combine | ``` <!-- - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Fixes # --> ### Does this PR introduce _any_ user-facing change? No. <!-- Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. Documentation-only updates are not considered user-facing changes. --> ### How was this patch tested? Tested on 1x16 910 node, with tailored 2 layer DSKv2. <!-- CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> --------- Signed-off-by: sdmyzlp <lrwei2@petalmail.com>
115 lines
3.7 KiB
Python
115 lines
3.7 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
|
|
#
|
|
"""Compare the short outputs of HF and vLLM when using greedy sampling.
|
|
|
|
Run `pytest tests/test_offline_inference.py`.
|
|
"""
|
|
import os
|
|
from unittest.mock import patch
|
|
|
|
from modelscope import snapshot_download # type: ignore
|
|
from vllm import SamplingParams
|
|
|
|
from tests.conftest import VllmRunner
|
|
|
|
os.environ["PYTORCH_NPU_ALLOC_CONF"] = "max_split_size_mb:256"
|
|
|
|
|
|
def test_models_distributed_QwQ():
|
|
example_prompts = [
|
|
"Hello, my name is",
|
|
]
|
|
dtype = "half"
|
|
max_tokens = 5
|
|
with VllmRunner(
|
|
"Qwen/QwQ-32B",
|
|
dtype=dtype,
|
|
tensor_parallel_size=4,
|
|
distributed_executor_backend="mp",
|
|
) as vllm_model:
|
|
vllm_model.generate_greedy(example_prompts, max_tokens)
|
|
|
|
|
|
def test_models_distributed_DeepSeek():
|
|
example_prompts = [
|
|
"Hello, my name is",
|
|
]
|
|
dtype = "half"
|
|
max_tokens = 5
|
|
with VllmRunner(
|
|
"deepseek-ai/DeepSeek-V2-Lite",
|
|
dtype=dtype,
|
|
tensor_parallel_size=4,
|
|
distributed_executor_backend="mp",
|
|
) as vllm_model:
|
|
vllm_model.generate_greedy(example_prompts, max_tokens)
|
|
|
|
|
|
@patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_TOPK_OPTIMIZE": "1"})
|
|
def test_models_distributed_topk() -> None:
|
|
example_prompts = [
|
|
"vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.",
|
|
"Briefly describe the major milestones in the development of artificial intelligence from 1950 to 2020.",
|
|
"Compare and contrast artificial intelligence with human intelligence in terms of processing information.",
|
|
]
|
|
dtype = "half"
|
|
sampling_params = SamplingParams(max_tokens=5,
|
|
temperature=0.0,
|
|
top_k=50,
|
|
top_p=0.9)
|
|
|
|
with VllmRunner(
|
|
"deepseek-ai/DeepSeek-V2-Lite",
|
|
dtype=dtype,
|
|
tensor_parallel_size=4,
|
|
distributed_executor_backend="mp",
|
|
) as vllm_model:
|
|
vllm_model.generate(example_prompts, sampling_params)
|
|
|
|
|
|
@patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_DBO": "1"})
|
|
def test_models_distributed_DeepSeek_dbo():
|
|
example_prompts = ["The president of the United States is"] * 41
|
|
dtype = "half"
|
|
sampling_params = SamplingParams(max_tokens=100, temperature=0.0)
|
|
with VllmRunner(
|
|
"deepseek-ai/DeepSeek-V2-Lite",
|
|
dtype=dtype,
|
|
tensor_parallel_size=4,
|
|
distributed_executor_backend="mp",
|
|
) as vllm_model:
|
|
vllm_model.generate(example_prompts, sampling_params)
|
|
|
|
|
|
def test_models_distributed_DeepSeek_W8A8():
|
|
example_prompts = [
|
|
"Hello, my name is",
|
|
]
|
|
max_tokens = 5
|
|
|
|
with VllmRunner(
|
|
snapshot_download("vllm-ascend/DeepSeek-V2-Lite-W8A8"),
|
|
max_model_len=8192,
|
|
enforce_eager=True,
|
|
dtype="auto",
|
|
tensor_parallel_size=4,
|
|
quantization="ascend",
|
|
) as vllm_model:
|
|
vllm_model.generate_greedy(example_prompts, max_tokens)
|