Files
xc-llm-ascend/.github/workflows/schedule_nightly_test_a3.yaml
Qiu cb7c419bc0 [Feat](sfa,dcp) support dcp for sfa (#6563)
### What this PR does / why we need it?
This PR adds DCP support to the SFA backend.

Please note that due to operator constraints, the current implementation
has to all-gather the entire KV cache and modify the block table to
satisfy the operator input requirements. This results in significantly
increased communication overhead and peak memory usage. Therefore, this
is only a temporary workaround and will be refactored once the operator
provides proper support.

Additionally, because of the above limitations,
`cp_kv_cache_interleave_size` is currently required to be equal to
`block_size`. This restriction will also be removed after the refactor.

#### Test
accuracy test using DeepSeek-V3.2-Exp-W8A8 with dp2tp8dcp8

| dataset | version | metric | mode | vllm-api-general-stream |
|----- | ----- | ----- | ----- | -----|
| gsm8kdataset | - | accuracy | gen | 96.35 |

- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0

---------

Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
2026-02-09 18:52:25 +08:00

192 lines
7.7 KiB
YAML

#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
#
# 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.
#
# This workflow related to the resources atlas 800 A3
# **Please note**: current A3 resource pool's maximum allowed concurrency is 5*16 NPUs
# We will limit the concurrency of jobs on A3 to avoid the risk of insufficient resources
name: Nightly-A3
on:
schedule:
# Run test at 24:00 Beijing time (UTC+8)
- cron: "0 16 * * *"
workflow_dispatch:
pull_request:
branches:
- 'main'
types: [ labeled ]
# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
# declared as "shell: bash -el {0}" on steps that need to be properly activated.
# It's used to activate ascend-toolkit environment variables.
defaults:
run:
shell: bash -el {0}
concurrency:
group: ascend-nightly-${{ github.ref }}-a3
cancel-in-progress: true
jobs:
multi-node-tests:
name: multi-node
if: always() && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch')
strategy:
fail-fast: false
max-parallel: 1
matrix:
test_config:
- name: multi-node-deepseek-pd
config_file_path: DeepSeek-V3.yaml
size: 2
- name: multi-node-qwen3-dp
config_file_path: Qwen3-235B-A22B.yaml
size: 2
# - name: multi-node-dpsk-4node-pd
# config_file_path: DeepSeek-R1-W8A8.yaml
# size: 4
- name: multi-node-qwenw8a8-2node
config_file_path: Qwen3-235B-W8A8.yaml
size: 2
# - name: multi-node-deepseek-r1-w8a8-eplb
# config_file_path: DeepSeek-R1-W8A8-EPLB.yaml
# size: 4
- name: multi-node-qwenw8a8-2node-eplb
config_file_path: Qwen3-235B-W8A8-EPLB.yaml
size: 2
- name: multi-node-dpsk3.2-2node
config_file_path: DeepSeek-V3_2-W8A8-A3-dual-nodes.yaml
size: 2
- name: multi-node-deepseek-r1-w8a8-longseq
config_file_path: DeepSeek-R1-W8A8-longseq.yaml
size: 2
- name: multi-node-qwenw8a8-2node-longseq
config_file_path: Qwen3-235B-W8A8-longseq.yaml
size: 2
- name: multi-node-deepseek-V3_2-W8A8-cp
config_file_path: DeepSeek-V3_2-W8A8-cp.yaml
size: 2
- name: multi-node-qwen-disagg-pd
config_file_path: Qwen3-235B-disagg-pd.yaml
size: 2
- name: multi-node-qwen-vl-disagg-pd
config_file_path: Qwen3-VL-235B-disagg-pd.yaml
size: 2
- name: multi-node-kimi-k2-instruct-w8a8
config_file_path: Kimi-K2-Instruct-W8A8.yaml
size: 2
- name: multi-node-deepseek-v3.1
config_file_path: DeepSeek-V3.1-BF16.yaml
size: 2
uses: ./.github/workflows/_e2e_nightly_multi_node.yaml
with:
soc_version: a3
runner: linux-aarch64-a3-0
image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:nightly-a3'
replicas: 1
size: ${{ matrix.test_config.size }}
config_file_path: ${{ matrix.test_config.config_file_path }}
secrets:
KUBECONFIG_B64: ${{ secrets.KUBECONFIG_B64 }}
single-node-tests:
name: single-node
if: always() && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch')
needs: multi-node-tests
strategy:
fail-fast: false
matrix:
test_config:
- name: qwen3-32b-in8-a3
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_qwen3_32b_int8.py
- name: qwen3-32b-int8-a3-feature-stack3
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_qwen3_32b_int8_a3_feature_stack3.py
- name: qwen3-235b-a22b-w8a8-eplb
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_qwen3_235b_a22b_w8a8_eplb.py
- name: deepseek-r1-w8a8-eplb
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_deepseek_r1_0528_w8a8_eplb.py
- name: deepseek-r1-w8a8-mtpx
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_mtpx_deepseek_r1_0528_w8a8.py
- name: qwen2-5-vl-7b
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_qwen2_5_vl_7b.py
- name: qwen2-5-vl-7b-epd
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_qwen2_5_vl_7b_epd.py
- name: qwen2-5-vl-32b
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_qwen2_5_vl_32b.py
- name: qwen3-32b-int8-prefix-cache
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_prefix_cache_qwen3_32b_int8.py
- name: deepseek-r1-0528-w8a8
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_deepseek_r1_0528_w8a8.py
- name: deepseek-r1-0528-w8a8-prefix-cache
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_prefix_cache_deepseek_r1_0528_w8a8.py
- name: qwq-32b-a3
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_qwq_32b.py
- name: qwen3-30b-w8a8
os: linux-aarch64-a3-2
tests: tests/e2e/nightly/single_node/models/test_qwen3_30b_w8a8.py
- name: qwen3-235b-w8a8
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_qwen3_235b_w8a8.py
- name: qwen3-next-w8a8
os: linux-aarch64-a3-4
tests: tests/e2e/nightly/single_node/models/test_qwen3_next_w8a8.py
- name: kimi-k2-thinking
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_kimi_k2_thinking.py
- name: deepseek-r1-w8a8-hbm
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_deepseek_r1_w8a8_hbm.py
- name: deepseek3_2-w8a8
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/models/test_deepseek_v3_2_w8a8.py
uses: ./.github/workflows/_e2e_nightly_single_node.yaml
with:
runner: ${{ matrix.test_config.os }}
image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:nightly-a3'
tests: ${{ matrix.test_config.tests }}
name: ${{ matrix.test_config.name }}
custom-ops-tests:
name: test ops
if: always() && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch')
needs: multi-node-tests
strategy:
fail-fast: false
matrix:
test_config:
- name: custom-multi-ops
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/ops/multicard_ops_a3/
uses: ./.github/workflows/_e2e_nightly_single_node.yaml
with:
runner: ${{ matrix.test_config.os }}
image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:nightly-a3'
tests: ${{ matrix.test_config.tests }}
name: ${{ matrix.test_config.name }}