minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[AMD] Replace MI325 with MI300 CI Runners (#31409)

Browse Source
This commit is contained in:
Bingxu Chen
2026-07-29 08:18:07 -07:00
committed by GitHub
parent e1f2f9d1fa
commit bfc450248e
9 changed files with 200 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/pr-test-amd-extra.yml
+12 -7
View File
@@ -32,7 +32,7 @@ on:
description: 'AMD runner pool to dispatch GPU jobs to'
required: false
type: choice
default: mi325
default: mi300
options:
- mi300
- mi325
@@ -61,6 +61,11 @@ on:
required: false
type: string
default: ''
runner_arch:
description: 'AMD runner pool to dispatch GPU jobs to'
required: false
type: string
default: mi300
rocm_version:
description: 'ROCm container variant (empty = Dockerfile default; rocm720 = ROCm 7.2.0)'
required: false
@@ -125,10 +130,10 @@ jobs:
# setup across scarce AMD GPUs to shave only a couple minutes of test time,
# so one GPU running the whole suite sequentially is the better trade.
extra-a-test-1-gpu-small-amd:
name: ${{ format('extra-a-test-1-gpu-small-amd{0} (linux-{1}-1gpu-sglang)', inputs.rocm_version && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi325') }}
name: ${{ format('extra-a-test-1-gpu-small-amd{0} (linux-{1}-1gpu-sglang)', inputs.rocm_version && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi300') }}
needs: [call-gate]
if: ${{ !cancelled() && needs.call-gate.result == 'success' }}
runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi325') }}
runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi300') }}
steps:
- name: Checkout code
uses: actions/checkout@v4
@@ -157,10 +162,10 @@ jobs:
# pool as small (AMD GPUs carry enough VRAM that "large" here is a CUDA
# memory-tier label, not a separate AMD runner pool).
extra-a-test-1-gpu-large-amd:
name: ${{ format('extra-a-test-1-gpu-large-amd{0} (linux-{1}-1gpu-sglang)', inputs.rocm_version && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi325') }}
name: ${{ format('extra-a-test-1-gpu-large-amd{0} (linux-{1}-1gpu-sglang)', inputs.rocm_version && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi300') }}
needs: [call-gate]
if: ${{ !cancelled() && needs.call-gate.result == 'success' }}
runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi325') }}
runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi300') }}
steps:
- name: Checkout code
uses: actions/checkout@v4
@@ -187,10 +192,10 @@ jobs:
# Multi-GPU TP / PP / PD mock-model + kv_canary e2e tests. Mirrors CUDA's
# extra-a 2-gpu-large stage; runs on the 2-GPU AMD pool.
extra-a-test-2-gpu-large-amd:
name: ${{ format('extra-a-test-2-gpu-large-amd{0} (linux-{1}-2gpu-sglang)', inputs.rocm_version && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi325') }}
name: ${{ format('extra-a-test-2-gpu-large-amd{0} (linux-{1}-2gpu-sglang)', inputs.rocm_version && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi300') }}
needs: [call-gate]
if: ${{ !cancelled() && needs.call-gate.result == 'success' }}
runs-on: ${{ format('linux-{0}-2gpu-sglang', inputs.runner_arch || 'mi325') }}
runs-on: ${{ format('linux-{0}-2gpu-sglang', inputs.runner_arch || 'mi300') }}
steps:
- name: Checkout code
uses: actions/checkout@v4