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

[AMD][CI] Swap the AMD PR gate to ROCm 7.2 and demote ROCm 7.0 to a daily shadow (#34204)

Browse Source 
Co-authored-by: yctseng0211 <yctseng@smci355-ccs-aus-m12-17.cs-aus.dcgpu>
Co-authored-by: Chen <bingxche@amd.com>
This commit is contained in:
YC Yen-Ching Tseng
2026-08-12 03:31:03 -07:00
committed by GitHub
co-authored by yctseng0211 Chen
parent b3bffef70a
commit 00bdafe944
12 changed files with 515 additions and 432 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/pr-test-amd-extra.yml
+13 -12
View File
@@ -37,13 +37,13 @@ on:
- mi300
- mi325
rocm_version:
description: 'ROCm container variant (empty = Dockerfile default; rocm720 = ROCm 7.2.0)'
description: 'ROCm container variant (rocm720 = ROCm 7.2.0, the PR default; rocm700 = ROCm 7.0.0)'
required: false
type: choice
default: ''
default: rocm720
options:
- ''
- rocm720
- rocm700
aiter_ref:
description: 'Override AITER commit (optional, leave empty to use Dockerfile default)'
required: false
@@ -67,10 +67,10 @@ on:
type: string
default: mi300
rocm_version:
description: 'ROCm container variant (empty = Dockerfile default; rocm720 = ROCm 7.2.0)'
description: 'ROCm container variant (rocm720 = ROCm 7.2.0, the PR default; rocm700 = ROCm 7.0.0)'
required: false
type: string
default: ''
default: rocm720
aiter_ref:
description: 'Override AITER commit (optional, leave empty to use Dockerfile default)'
required: false
@@ -130,7 +130,7 @@ 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 && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi300') }}
name: ${{ format('extra-a-test-1-gpu-small-amd{0} (linux-{1}-1gpu-sglang)', inputs.rocm_version && inputs.rocm_version != 'rocm720' && 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 || 'mi300') }}
@@ -144,8 +144,9 @@ jobs:
run: bash scripts/ci/amd/ensure_vram_clear.sh rocm
- name: Start CI container
# `rocm_version` (e.g. rocm720) selects an alternate ROCm container; empty uses the Dockerfile default.
run: bash scripts/ci/amd/amd_ci_start_container.sh ${{ inputs.rocm_version && format('--rocm-version {0}', inputs.rocm_version) || '' }}
# `inputs` is empty on pull_request events, so fall back explicitly to
# the ROCm 7.2 PR default. Reusable callers can still request ROCm 7.0.
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm720' }}
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
@@ -162,7 +163,7 @@ 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 && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi300') }}
name: ${{ format('extra-a-test-1-gpu-large-amd{0} (linux-{1}-1gpu-sglang)', inputs.rocm_version && inputs.rocm_version != 'rocm720' && 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 || 'mi300') }}
@@ -176,7 +177,7 @@ 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 ${{ inputs.rocm_version && format('--rocm-version {0}', inputs.rocm_version) || '' }}
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm720' }}
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
@@ -192,7 +193,7 @@ 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 && format('-{0}', inputs.rocm_version) || '', inputs.runner_arch || 'mi300') }}
name: ${{ format('extra-a-test-2-gpu-large-amd{0} (linux-{1}-2gpu-sglang)', inputs.rocm_version && inputs.rocm_version != 'rocm720' && 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 || 'mi300') }}
@@ -206,7 +207,7 @@ 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 ${{ inputs.rocm_version && format('--rocm-version {0}', inputs.rocm_version) || '' }}
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm720' }}
env:
GITHUB_WORKSPACE: ${{ github.workspace }}