EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[CI] cleanup single/multi-card test (#5623)

Browse Source 
1. speed up e2e light test.
2. create `2-cards` and `4-cards` folder in multicard
3. move ops to nightly
4. run test in Alphabetical Order

- vLLM version: v0.13.0
- vLLM main:
8be6432bda

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2026-01-07 14:13:34 +08:00
committed by GitHub
parent 1afbc01ed4
commit 6f7a81cd9f
30 changed files with 114 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files

248
tests/e2e/multicard/4-cards/long_sequence/test_basic.py Normal file
View File

@@ -0,0 +1,248 @@
#
# 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
from vllm import SamplingParams
from tests.e2e.conftest import VllmRunner
os.environ["HCCL_BUFFSIZE"] = "768"
def test_models_pcp_dcp_basic():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
enforce_eager=True,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=2,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128) as runner:
runner.model.generate(prompts, sampling_params)
model = "vllm-ascend/Qwen3-30B-A3B-W8A8"
with VllmRunner(
model,
enforce_eager=True,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=1,
enable_expert_parallel=True,
block_size=128,
quantization="ascend",
) as runner:
runner.model.generate(prompts, sampling_params)
def test_models_pcp_dcp_full_graph():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=2,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128,
compilation_config={
"cudagraph_mode": "FULL_DECODE_ONLY",
"cudagraph_capture_sizes": [4, 8, 24, 48, 60]
}) as runner:
runner.model.generate(prompts, sampling_params)
model = "vllm-ascend/Qwen3-30B-A3B-W8A8"
with VllmRunner(model,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=1,
enable_expert_parallel=True,
block_size=128,
quantization="ascend",
compilation_config={
"cudagraph_mode": "FULL_DECODE_ONLY",
"cudagraph_capture_sizes": [4, 8, 24, 48, 60]
}) as runner:
runner.model.generate(prompts, sampling_params)
def test_models_pcp_dcp_piece_wise():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=2,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
cudagraph_capture_sizes=[1, 2, 4, 8],
block_size=128) as runner:
runner.model.generate(prompts, sampling_params)
model = "vllm-ascend/Qwen3-30B-A3B-W8A8"
with VllmRunner(model,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=1,
enable_expert_parallel=True,
cudagraph_capture_sizes=[1, 2, 4, 8],
block_size=128,
quantization="ascend") as runner:
runner.model.generate(prompts, sampling_params)
def test_pcp_basic():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
enforce_eager=True,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=1,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128) as runner:
runner.model.generate(prompts, sampling_params)
def test_pcp_full_graph():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
enforce_eager=False,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=1,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128,
compilation_config={
"cudagraph_mode": "FULL_DECODE_ONLY",
"cudagraph_capture_sizes": [4, 8, 24, 48, 60]
}) as runner:
runner.model.generate(prompts, sampling_params)
def test_pcp_piece_wise():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
enforce_eager=False,
max_model_len=1024,
tensor_parallel_size=2,
prefill_context_parallel_size=2,
decode_context_parallel_size=1,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128) as runner:
runner.model.generate(prompts, sampling_params)
def test_dcp_basic():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
enforce_eager=True,
max_model_len=1024,
tensor_parallel_size=4,
prefill_context_parallel_size=1,
decode_context_parallel_size=2,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128) as runner:
runner.model.generate(prompts, sampling_params)
def test_dcp_full_graph():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
enforce_eager=False,
max_model_len=1024,
tensor_parallel_size=4,
prefill_context_parallel_size=1,
decode_context_parallel_size=2,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128,
compilation_config={
"cudagraph_mode": "FULL_DECODE_ONLY",
"cudagraph_capture_sizes": [4, 8, 24, 48, 60]
}) as runner:
runner.model.generate(prompts, sampling_params)
def test_dcp_piece_wise():
prompts = [
"The capital of France is", "Hello, my name is Tom, I am",
"The president of United States is", "AI future is"
]
model = "deepseek-ai/DeepSeek-V2-Lite-Chat"
sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
with VllmRunner(model,
enforce_eager=False,
max_model_len=1024,
tensor_parallel_size=4,
prefill_context_parallel_size=1,
decode_context_parallel_size=2,
max_num_batched_tokens=1024,
enable_expert_parallel=True,
block_size=128) as runner:
runner.model.generate(prompts, sampling_params)