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

[AMD][CI] Make ROCm 10 the Default for AMD PR and Nightly Tests (#38659)

Browse Source 
Co-authored-by: Chen Bingxu <bingxche@amd.com>
This commit is contained in:
YC Yen-Ching Tseng
2026-09-09 17:28:20 +08:00
committed by GitHub
co-authored by Chen Bingxu
parent 72d5c5bb73
commit 708f51e44b
8 changed files with 32 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/pr-test-amd-extra.yml
+12 -12
View File
@@ -37,10 +37,10 @@ on:
- mi300
- mi325
rocm_version:
description: 'ROCm container variant (ROCm 7.2.4 by default)'
description: 'ROCm container variant (ROCm 10 by default)'
required: false
type: choice
default: rocm724
default: rocm10
options:
- rocm10
- rocm724
@@ -74,10 +74,10 @@ on:
type: string
default: mi300
rocm_version:
description: 'ROCm container variant (ROCm 7.2.4 by default)'
description: 'ROCm container variant (ROCm 10 by default)'
required: false
type: string
default: rocm724
default: rocm10
aiter_ref:
description: 'Override AITER commit (optional, leave empty to use Dockerfile default)'
required: false
@@ -105,7 +105,7 @@ concurrency:
# scheduled reusable-workflow matrix calls cannot replace each other's
# pending runs. PR runs still share a per-branch group so pushes cancel stale
# runs. In a reusable workflow github.event_name is the originating event.
group: pr-test-amd-extra-${{ (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') && format('full-{0}-{1}', github.run_id, inputs.rocm_version || 'rocm724') || github.head_ref || github.ref_name || inputs.ref || 'default' }}
group: pr-test-amd-extra-${{ (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') && format('full-{0}-{1}', github.run_id, inputs.rocm_version || 'rocm10') || github.head_ref || github.ref_name || inputs.ref || 'default' }}
cancel-in-progress: ${{ github.event_name == 'pull_request' }}
permissions:
@@ -143,7 +143,7 @@ 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 || 'rocm724', inputs.runner_arch || 'mi300') }}
name: ${{ format('extra-a-test-1-gpu-small-amd ({0}, linux-{1}-1gpu-sglang)', inputs.rocm_version || 'rocm10', inputs.runner_arch || 'mi300') }}
needs: [call-gate]
if: ${{ !cancelled() && needs.call-gate.result == 'success' }}
runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -158,8 +158,8 @@ jobs:
- name: Start CI container
# `inputs` is empty on pull_request events, so fall back explicitly to
# the ROCm 7.2 PR default. Reusable callers can request another variant.
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm724' }}
# the ROCm 10 PR default. Reusable callers can request another variant.
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm10' }}
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
@@ -176,7 +176,7 @@ 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 || 'rocm724', inputs.runner_arch || 'mi300') }}
name: ${{ format('extra-a-test-1-gpu-large-amd ({0}, linux-{1}-1gpu-sglang)', inputs.rocm_version || 'rocm10', inputs.runner_arch || 'mi300') }}
needs: [call-gate]
if: ${{ !cancelled() && needs.call-gate.result == 'success' }}
runs-on: ${{ format('linux-{0}-1gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -190,7 +190,7 @@ jobs:
run: bash scripts/ci/amd/ensure_vram_clear.sh rocm
- name: Start CI container
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm724' }}
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm10' }}
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
@@ -206,7 +206,7 @@ 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 || 'rocm724', inputs.runner_arch || 'mi300') }}
name: ${{ format('extra-a-test-2-gpu-large-amd ({0}, linux-{1}-2gpu-sglang)', inputs.rocm_version || 'rocm10', inputs.runner_arch || 'mi300') }}
needs: [call-gate]
if: ${{ !cancelled() && needs.call-gate.result == 'success' }}
runs-on: ${{ format('linux-{0}-2gpu-sglang', inputs.runner_arch || 'mi300') }}
@@ -220,7 +220,7 @@ jobs:
run: bash scripts/ci/amd/ensure_vram_clear.sh rocm
- name: Start CI container
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm724' }}
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm10' }}
env:
GITHUB_WORKSPACE: ${{ github.workspace }}