fbae41697e
### What this PR does / why we need it?
* Refactor the LayerNorm and activation operator classes to decouple the
310P device implementation from the main branch.
* Refactor `mm_encoder_attention` on 310P to use the
`torch_npu._npu_flash_attention_unpad` operator.
* Refactor the QKV inputs in the prefill stage of `attention_v1` on 310P
so they are no longer padded to 16× alignment.
* Refactor `model_runner` on 310P to align the KV-cache initialization
logic with the mainline implementation.
### Does this PR introduce _any_ user-facing change?
NO
### How was this patch tested?
use the e2e tests.
- vLLM version: v0.13.0
- vLLM main:
d68209402d
---------
Signed-off-by: Tflowers-0129 <2906339855@qq.com>
59 lines
1.9 KiB
YAML
59 lines
1.9 KiB
YAML
# This is a docker build check and publish job:
|
|
# 1. PR Triggered docker image build check
|
|
# - is for image build check
|
|
# - Enable on main/*-dev branch
|
|
# - push: ${{ github.event_name != 'pull_request' }} ==> false
|
|
# 2. branches push trigger image publish
|
|
# - is for branch/dev/nightly image
|
|
# - commits are merge into main/*-dev ==> vllm-ascend:main / vllm-ascend:*-dev
|
|
# 3. tags push trigger image publish
|
|
# - is for final release image
|
|
# - Publish when tag with v* (pep440 version) ===> vllm-ascend:v1.2.3 / vllm-ascend:v1.2.3rc1
|
|
name: Image Build and Push
|
|
on:
|
|
schedule:
|
|
# UTC+8: 8am, 12pm, 16pm, 22pm
|
|
- cron: '0 0,4,8,14 * * *'
|
|
push:
|
|
tags:
|
|
- 'v*'
|
|
workflow_dispatch:
|
|
inputs:
|
|
tag:
|
|
description: 'Docker tag for build results'
|
|
type: string
|
|
default: main
|
|
required: true
|
|
|
|
jobs:
|
|
image_build:
|
|
name: Image Build and Push
|
|
strategy:
|
|
matrix:
|
|
build_meta:
|
|
- name: A2 Ubuntu
|
|
dockerfile: Dockerfile
|
|
suffix: ''
|
|
- name: A2 openeuler
|
|
dockerfile: Dockerfile.openEuler
|
|
suffix: 'openeuler'
|
|
- name: A3 Ubuntu
|
|
dockerfile: Dockerfile.a3
|
|
suffix: 'a3'
|
|
- name: A3 openEuler
|
|
dockerfile: Dockerfile.a3.openEuler
|
|
suffix: 'a3-openeuler'
|
|
- name: 310P Ubuntu
|
|
dockerfile: Dockerfile.310p
|
|
- name: 310P openEuler
|
|
dockerfile: Dockerfile.310p.openEuler
|
|
uses: ./.github/workflows/_schedule_image_build.yaml
|
|
with:
|
|
dockerfile: ${{ matrix.build_meta.dockerfile }}
|
|
suffix: ${{ matrix.build_meta.suffix }}
|
|
quay_username: ${{ vars.QUAY_USERNAME }}
|
|
should_push: ${{ github.repository_owner == 'vllm-project' }}
|
|
workflow_dispatch_tag: ${{ inputs.tag }}
|
|
secrets:
|
|
QUAY_PASSWORD: ${{ secrets.QUAY_PASSWORD }}
|