|
|
|
@@ -37,13 +37,13 @@ on:
|
|
|
|
|
- mi300
|
|
|
|
|
- mi325
|
|
|
|
|
rocm_version:
|
|
|
|
|
description: 'ROCm container variant (ROCm 7.2.0 by default)'
|
|
|
|
|
description: 'ROCm container variant (ROCm 7.2.4 by default)'
|
|
|
|
|
required: false
|
|
|
|
|
type: choice
|
|
|
|
|
default: rocm720
|
|
|
|
|
default: rocm724
|
|
|
|
|
options:
|
|
|
|
|
- rocm720
|
|
|
|
|
- rocm724
|
|
|
|
|
- rocm720
|
|
|
|
|
- rocm700
|
|
|
|
|
aiter_ref:
|
|
|
|
|
description: 'Override AITER commit (optional, leave empty to use Dockerfile default)'
|
|
|
|
@@ -73,10 +73,10 @@ on:
|
|
|
|
|
type: string
|
|
|
|
|
default: mi300
|
|
|
|
|
rocm_version:
|
|
|
|
|
description: 'ROCm container variant (ROCm 7.2.0 by default)'
|
|
|
|
|
description: 'ROCm container variant (ROCm 7.2.4 by default)'
|
|
|
|
|
required: false
|
|
|
|
|
type: string
|
|
|
|
|
default: rocm720
|
|
|
|
|
default: rocm724
|
|
|
|
|
aiter_ref:
|
|
|
|
|
description: 'Override AITER commit (optional, leave empty to use Dockerfile default)'
|
|
|
|
|
required: false
|
|
|
|
@@ -142,7 +142,7 @@ 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 && inputs.rocm_version != 'rocm720' && 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 != 'rocm724' && 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') }}
|
|
|
|
@@ -158,7 +158,7 @@ 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 || 'rocm720' }}
|
|
|
|
|
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm724' }}
|
|
|
|
|
env:
|
|
|
|
|
GITHUB_WORKSPACE: ${{ github.workspace }}
|
|
|
|
|
|
|
|
|
@@ -175,7 +175,7 @@ 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 && inputs.rocm_version != 'rocm720' && 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 != 'rocm724' && 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') }}
|
|
|
|
@@ -189,7 +189,7 @@ 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 || 'rocm720' }}
|
|
|
|
|
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm724' }}
|
|
|
|
|
env:
|
|
|
|
|
GITHUB_WORKSPACE: ${{ github.workspace }}
|
|
|
|
|
|
|
|
|
@@ -205,7 +205,7 @@ 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 && inputs.rocm_version != 'rocm720' && 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 != 'rocm724' && 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') }}
|
|
|
|
@@ -219,7 +219,7 @@ 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 || 'rocm720' }}
|
|
|
|
|
run: bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version ${{ inputs.rocm_version || 'rocm724' }}
|
|
|
|
|
env:
|
|
|
|
|
GITHUB_WORKSPACE: ${{ github.workspace }}
|
|
|
|
|
|
|
|
|
|