a054f0f4ca
Previous, the DeepSeek V3 Pruning weight is not correct, the moe layer is not tested. We update a new Pruning model to enable moe layer compute. This PR fix the CI to address the new weight. --------- Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
82 lines
2.9 KiB
Python
82 lines
2.9 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.
|
|
#
|
|
"""
|
|
Execute the inference of fused_moe_allgather_ep and fused_moe_alltoall_ep.
|
|
|
|
Run 'pytest tests/multicard/test_fused_moe_allgather_ep.py'.
|
|
"""
|
|
|
|
import os
|
|
from unittest.mock import patch
|
|
|
|
from modelscope import snapshot_download # type: ignore
|
|
from vllm import SamplingParams
|
|
|
|
from tests.conftest import VllmRunner
|
|
|
|
|
|
@patch.dict(
|
|
os.environ, {
|
|
"VLLM_USE_V1": "1",
|
|
"VLLM_WORKER_MULTIPROC_METHOD": "spawn",
|
|
"TASK_QUEUE_ENABLE": "1",
|
|
"VLLM_ENABLE_FUSED_EXPERTS_ALLGATHER_EP": "1"
|
|
})
|
|
def test_generate_with_allgather():
|
|
example_prompts = ["Hello, my name is"]
|
|
sampling_params = SamplingParams(max_tokens=100, temperature=0.0)
|
|
|
|
with VllmRunner(snapshot_download("vllm-ascend/DeepSeek-V3-Pruning"),
|
|
tensor_parallel_size=4,
|
|
enforce_eager=True,
|
|
max_model_len=1024,
|
|
dtype="auto",
|
|
enable_expert_parallel=True,
|
|
additional_config={
|
|
"ascend_scheduler_config": {
|
|
"enabled": True,
|
|
"chunked_prefill_enabled": False,
|
|
},
|
|
"expert_tensor_parallel_size": 1
|
|
}) as vllm_model:
|
|
vllm_model.generate(example_prompts, sampling_params)
|
|
|
|
|
|
@patch.dict(
|
|
os.environ, {
|
|
"VLLM_USE_V1": "1",
|
|
"VLLM_WORKER_MULTIPROC_METHOD": "spawn",
|
|
"TASK_QUEUE_ENABLE": "1"
|
|
})
|
|
def test_generate_with_alltoall():
|
|
example_prompts = ["Hello, my name is"]
|
|
sampling_params = SamplingParams(max_tokens=100, temperature=0.0)
|
|
|
|
with VllmRunner(snapshot_download("vllm-ascend/DeepSeek-V3-Pruning"),
|
|
tensor_parallel_size=4,
|
|
enforce_eager=True,
|
|
max_model_len=1024,
|
|
dtype="auto",
|
|
enable_expert_parallel=True,
|
|
additional_config={
|
|
"ascend_scheduler_config": {
|
|
"enabled": True,
|
|
"chunked_prefill_enabled": False,
|
|
},
|
|
"expert_tensor_parallel_size": 1
|
|
}) as vllm_model:
|
|
vllm_model.generate(example_prompts, sampling_params) |