EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
Files
e17006077ab1e1cf52a659952e3e187647e4be32
xc-llm-ascend/.github/workflows/schedule_nightly_test_a3.yaml
weiguihua2 59a7526339 [CI][Misc] modify ds3.2+dcp ci (#7841) 
### What this PR does / why we need it?

Due to the current dcp solution of allgathering the KV cache, the
performance deteriorates significantly, and the CI may get stuck. This
PR temporarily removes the performance and accuracy benchmarks for
DeepSeek-V3.2-W8A8-cp to prevent CI hangs until optimization is
complete.

pcik-from:https://github.com/vllm-project/vllm-ascend/pull/7842

### Does this PR introduce _any_ user-facing change?

No

### How was this patch tested?

Verified that the configuration file remains valid and that the CI no
longer attempts to run the problematic benchmarks.

pick-from: https://github.com/vllm-project/vllm-ascend/pull/7842

---------

Signed-off-by: weiguihua2 <weiguihua2@huawei.com>
2026-04-01 08:58:21 +08:00

237 lines
8.8 KiB
YAML
Raw Blame History

#
# 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 23:45 Beijing time (UTC+8)
- cron: "45 15 * * *"
workflow_dispatch:
pull_request:
branches:
- 'main'
- '*-dev'
- 'releases/v*'
types: [ labeled, synchronize ]
permissions:
contents: read
pull-requests: read
issues: read
# 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:
parse-trigger:
name: Parse trigger and determine test scope
if: >-
github.event_name == 'schedule' ||
github.event_name == 'workflow_dispatch' ||
contains(github.event.pull_request.labels.*.name, 'nightly-test')
uses: ./.github/workflows/_parse_trigger.yaml
build-image:
name: Build nightly-a3 image
if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
uses: ./.github/workflows/_nightly_image_build.yaml
with:
target: a3
secrets:
HW_USERNAME: ${{ secrets.HW_USERNAME }}
HW_TOKEN: ${{ secrets.HW_TOKEN }}
GITEE_TOKEN: ${{ secrets.GITEE_TOKEN }}
multi-node-tests:
name: multi-node
needs: [parse-trigger, build-image]
if: >-
always() &&
needs.parse-trigger.outputs.run == 'true' &&
(needs.build-image.result == 'success' || needs.build-image.result == 'skipped')
strategy:
fail-fast: false
max-parallel: 2
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-qwenw8a8-2node
config_file_path: Qwen3-235B-W8A8.yaml
size: 2
- 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-qwen3-dp-mooncake-layerwise
config_file_path: Qwen3-235B-A22B-Mooncake-Layerwise.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-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
- name: multi-node-deepseek-v3.2-W8A8-EP
config_file_path: DeepSeek-V3_2-W8A8-EP.yaml
size: 4
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 }}
vllm_ascend_ref: ${{ needs.parse-trigger.outputs.ref }}
should_run: >-
${{
needs.parse-trigger.outputs.run == 'true' && (
needs.parse-trigger.outputs.filter == 'all' ||
contains(needs.parse-trigger.outputs.filter, format(',{0},', matrix.test_config.name))
)
}}
secrets:
KUBECONFIG_B64: ${{ secrets.KUBECONFIG_B64 }}
single-node-tests:
name: single-node
needs: [parse-trigger, build-image, multi-node-tests]
if: >-
always() &&
needs.parse-trigger.outputs.run == 'true' &&
(needs.build-image.result == 'success' || needs.build-image.result == 'skipped')
strategy:
fail-fast: false
matrix:
test_config:
# pytest-driven tests
- name: qwen3-30b-acc
os: linux-aarch64-a3-4
tests: tests/e2e/weekly/single_node/models/test_qwen3_30b_acc.py
- name: custom-multi-ops
os: linux-aarch64-a3-16
tests: tests/e2e/nightly/single_node/ops/multicard_ops_a3/
# YAML-driven tests
- name: deepseek-r1-0528-w8a8
os: linux-aarch64-a3-16
config_file_path: DeepSeek-R1-0528-W8A8.yaml
- name: deepseek-r1-w8a8-hbm
os: linux-aarch64-a3-16
config_file_path: DeepSeek-R1-W8A8-HBM.yaml
- name: deepseek-v3-2-w8a8
os: linux-aarch64-a3-16
config_file_path: DeepSeek-V3.2-W8A8.yaml
- name: glm-5-w4a8
os: linux-aarch64-a3-16
config_file_path: GLM-5.yaml
- name: glm-4.7-w8a8
os: linux-aarch64-a3-16
config_file_path: GLM-4.7.yaml
- name: kimi-k2-thinking
os: linux-aarch64-a3-16
config_file_path: Kimi-K2-Thinking.yaml
- name: kimi-k2.5
os: linux-aarch64-a3-16
config_file_path: Kimi-K2.5.yaml
- name: minimax-m2-5
os: linux-aarch64-a3-16
config_file_path: MiniMax-M2.5-A3.yaml
- name: mtpx-deepseek-r1-0528-w8a8
os: linux-aarch64-a3-16
config_file_path: MTPX-DeepSeek-R1-0528-W8A8.yaml
- name: qwen3-235b-a22b-w8a8
os: linux-aarch64-a3-16
config_file_path: Qwen3-235B-A22B-W8A8.yaml
- name: qwen3-30b-a3b-w8a8
os: linux-aarch64-a3-4
config_file_path: Qwen3-30B-A3B-W8A8.yaml
- name: qwen3-next-80b-a3b-instruct
os: linux-aarch64-a3-4
config_file_path: Qwen3-Next-80B-A3B-Instruct.yaml
- name: qwen3-next-80b-a3b-instruct-w8a8
os: linux-aarch64-a3-4
config_file_path: Qwen3-Next-80B-A3B-Instruct-W8A8.yaml
- name: qwq-32b
os: linux-aarch64-a3-4
config_file_path: QwQ-32B.yaml
- name: qwen3-32b-int8
os: linux-aarch64-a3-4
config_file_path: Qwen3-32B-Int8.yaml
- name: qwen2-5-vl-7b
os: linux-aarch64-a3-4
config_file_path: Qwen2.5-VL-7B-Instruct.yaml
- name: qwen2-5-vl-7b-epd
os: linux-aarch64-a3-4
config_file_path: Qwen2.5-VL-7B-Instruct-EPD.yaml
- name: qwen2-5-vl-32b
os: linux-aarch64-a3-4
config_file_path: Qwen2.5-VL-32B-Instruct.yaml
- name: qwen3-32b-int8-a3-feature-stack3
os: linux-aarch64-a3-4
config_file_path: Qwen3-32B-Int8-A3-Feature-Stack3.yaml
- name: qwen3-32b-int8-prefix-cache
os: linux-aarch64-a3-4
config_file_path: Prefix-Cache-Qwen3-32B-Int8.yaml
- name: deepseek-r1-0528-w8a8-prefix-cache
os: linux-aarch64-a3-16
config_file_path: Prefix-Cache-DeepSeek-R1-0528-W8A8.yaml
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 }}
config_file_path: ${{ matrix.test_config.config_file_path }}
name: ${{ matrix.test_config.name }}
should_run: >-
${{
needs.parse-trigger.outputs.run == 'true' && (
needs.parse-trigger.outputs.filter == 'all' ||
contains(needs.parse-trigger.outputs.filter, format(',{0},', matrix.test_config.name))
)
}}
Reference in New Issue View Git Blame Copy Permalink