From 060720c5733d3b0dce208f978fb424c851fe109a Mon Sep 17 00:00:00 2001 From: YC Tseng Date: Tue, 3 Mar 2026 06:17:12 +0300 Subject: [PATCH] [AMD] AMD new CI runner (#19739) Co-authored-by: bingxche --- .../workflows/nightly-test-amd-rocm720.yml | 30 +++++++++---------- .github/workflows/nightly-test-amd.yml | 30 +++++++++---------- .github/workflows/pr-test-amd-rocm720.yml | 24 +++++++-------- .github/workflows/pr-test-amd.yml | 24 +++++++-------- 4 files changed, 54 insertions(+), 54 deletions(-) diff --git a/.github/workflows/nightly-test-amd-rocm720.yml b/.github/workflows/nightly-test-amd-rocm720.yml index fba760527..f4a950c39 100644 --- a/.github/workflows/nightly-test-amd-rocm720.yml +++ b/.github/workflows/nightly-test-amd-rocm720.yml @@ -97,7 +97,7 @@ jobs: # 1-GPU Unit Tests - LoRA, debug utils, scheduler, etc. (MI30x ROCm 7.2) nightly-test-1-gpu-unit-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-1-gpu-unit-rocm720') - runs-on: linux-mi325-gpu-1 + runs-on: linux-mi325-1gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -126,7 +126,7 @@ jobs: # 2-GPU Accuracy Tests - GSM8K eval (MI30x ROCm 7.2) nightly-accuracy-2-gpu-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu-rocm720') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -154,7 +154,7 @@ jobs: # 2-GPU VLM Accuracy Tests - Vision-Language Models MMMU evaluation (ROCm 7.2) nightly-accuracy-2-gpu-vlm-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu-vlm-rocm720') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -183,7 +183,7 @@ jobs: # 2-GPU Text Models Performance Tests (ROCm 7.2) nightly-perf-2-gpu-text-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-text-rocm720') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -213,7 +213,7 @@ jobs: # 2-GPU VLM Performance Tests (ROCm 7.2) nightly-perf-2-gpu-vlm-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-vlm-rocm720') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -243,7 +243,7 @@ jobs: # 8-GPU Accuracy Tests - GPT-OSS, Grok1-FP8 (ROCm 7.2) nightly-accuracy-8-gpu-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -283,7 +283,7 @@ jobs: # 8-GPU Grok1-INT4 (Accuracy + Performance) ROCm 7.2 nightly-8-gpu-grok1-int4-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok1-int4-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -326,7 +326,7 @@ jobs: # 8-GPU Grok2 (Accuracy + Performance) ROCm 7.2 nightly-8-gpu-grok2-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok2-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -369,7 +369,7 @@ jobs: # 8-GPU DeepSeek-V3.1 (Accuracy + Performance) ROCm 7.2 nightly-8-gpu-deepseek-v31-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v31-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -412,7 +412,7 @@ jobs: # 8-GPU DeepSeek-V3.2 (Basic Accuracy + Perf) ROCm 7.2 nightly-8-gpu-deepseek-v32-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -453,7 +453,7 @@ jobs: # 8-GPU DeepSeek-V3.2 MTP (MTP Accuracy + Perf) ROCm 7.2 nightly-8-gpu-deepseek-v32-mtp-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32-mtp-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -494,7 +494,7 @@ jobs: # 8-GPU Kimi-K2.5 (Accuracy) ROCm 7.2 nightly-8-gpu-kimi-k25-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-kimi-k25-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -524,7 +524,7 @@ jobs: # 8-GPU Qwen3-235B (Accuracy + Performance) ROCm 7.2 nightly-8-gpu-qwen3-235b-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen3-235b-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -554,7 +554,7 @@ jobs: # 8-GPU GLM-5 (Accuracy) ROCm 7.2 nightly-8-gpu-glm5-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-glm5-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -587,7 +587,7 @@ jobs: # 8-GPU MiniMax-M2.5 (Accuracy) ROCm 7.2 nightly-8-gpu-minimax-m25-rocm720: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-minimax-m25-rocm720') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 diff --git a/.github/workflows/nightly-test-amd.yml b/.github/workflows/nightly-test-amd.yml index eee4fc864..d39247360 100644 --- a/.github/workflows/nightly-test-amd.yml +++ b/.github/workflows/nightly-test-amd.yml @@ -97,7 +97,7 @@ jobs: # 1-GPU Unit Tests - LoRA, debug utils, scheduler, etc. (MI30x only) nightly-test-1-gpu-unit: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-1-gpu-unit') - runs-on: linux-mi325-gpu-1 + runs-on: linux-mi325-1gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -127,7 +127,7 @@ jobs: # 2-GPU Accuracy Tests - GSM8K eval (MI30x only) nightly-accuracy-2-gpu: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -156,7 +156,7 @@ jobs: # 2-GPU VLM Accuracy Tests - Vision-Language Models MMMU evaluation nightly-accuracy-2-gpu-vlm: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu-vlm') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -186,7 +186,7 @@ jobs: # 2-GPU Text Models Performance Tests nightly-perf-2-gpu-text: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-text') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -217,7 +217,7 @@ jobs: # 2-GPU VLM Performance Tests nightly-perf-2-gpu-vlm: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-vlm') - runs-on: linux-mi325-gpu-2 + runs-on: linux-mi325-2gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -248,7 +248,7 @@ jobs: # 8-GPU Accuracy Tests - GPT-OSS, Grok1-FP8 (accuracy only) nightly-accuracy-8-gpu: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -288,7 +288,7 @@ jobs: # 8-GPU Grok1-INT4 (Accuracy + Performance combined) nightly-8-gpu-grok1-int4: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok1-int4') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -331,7 +331,7 @@ jobs: # 8-GPU Grok2 (Accuracy + Performance combined) nightly-8-gpu-grok2: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok2') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -374,7 +374,7 @@ jobs: # 8-GPU DeepSeek-V3.1 (Accuracy + Performance combined) nightly-8-gpu-deepseek-v31: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v31') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -417,7 +417,7 @@ jobs: # 8-GPU DeepSeek-V3.2 (Basic Accuracy + Perf) nightly-8-gpu-deepseek-v32: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -458,7 +458,7 @@ jobs: # 8-GPU DeepSeek-V3.2 MTP (MTP Accuracy + Perf) nightly-8-gpu-deepseek-v32-mtp: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32-mtp') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -499,7 +499,7 @@ jobs: # 8-GPU Kimi-K2.5 (Accuracy) nightly-8-gpu-kimi-k25: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-kimi-k25') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -528,7 +528,7 @@ jobs: nightly-8-gpu-qwen3-235b: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen3-235b') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -557,7 +557,7 @@ jobs: nightly-8-gpu-glm5: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-glm5') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 @@ -590,7 +590,7 @@ jobs: # 8-GPU MiniMax-M2.5 (Accuracy) nightly-8-gpu-minimax-m25: if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-minimax-m25') - runs-on: linux-mi325-gpu-8 + runs-on: linux-mi325-8gpu-sglang steps: - name: Checkout code uses: actions/checkout@v4 diff --git a/.github/workflows/pr-test-amd-rocm720.yml b/.github/workflows/pr-test-amd-rocm720.yml index c0216b017..b566e1d84 100644 --- a/.github/workflows/pr-test-amd-rocm720.yml +++ b/.github/workflows/pr-test-amd-rocm720.yml @@ -153,7 +153,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -199,7 +199,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-2] + runner: [linux-mi325-2gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -241,7 +241,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -279,7 +279,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -318,7 +318,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] part: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12] runs-on: ${{matrix.runner}} steps: @@ -357,7 +357,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -433,7 +433,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] part: [0, 1] runs-on: ${{matrix.runner}} steps: @@ -472,7 +472,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-2] + runner: [linux-mi325-2gpu-sglang] part: [0, 1] runs-on: ${{matrix.runner}} steps: @@ -512,7 +512,7 @@ jobs: fail-fast: false max-parallel: 1 # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] part: [0, 1] # 2 partitions: 11 tests ÷ 2 = ~5-6 tests each runs-on: ${{matrix.runner}} steps: @@ -642,7 +642,7 @@ jobs: fail-fast: false max-parallel: 1 # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT matrix: - runner: [linux-mi325-gpu-2] + runner: [linux-mi325-2gpu-sglang] part: [0, 1] # 2 partitions: 9 tests ÷ 2 = ~4-5 tests each runs-on: ${{matrix.runner}} steps: @@ -768,11 +768,11 @@ jobs: ) ) env: - RUNNER_LABELS: linux-mi325-gpu-8 + RUNNER_LABELS: linux-mi325-8gpu-sglang strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-8] + runner: [linux-mi325-8gpu-sglang] part: [0, 1, 2] runs-on: ${{matrix.runner}} steps: diff --git a/.github/workflows/pr-test-amd.yml b/.github/workflows/pr-test-amd.yml index 867808feb..a803a1ed1 100644 --- a/.github/workflows/pr-test-amd.yml +++ b/.github/workflows/pr-test-amd.yml @@ -150,7 +150,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -197,7 +197,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-2] + runner: [linux-mi325-2gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -240,7 +240,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -279,7 +279,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -319,7 +319,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] part: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13] runs-on: ${{matrix.runner}} steps: @@ -359,7 +359,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] runs-on: ${{matrix.runner}} steps: - name: Checkout code @@ -437,7 +437,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] part: [0, 1] runs-on: ${{matrix.runner}} steps: @@ -477,7 +477,7 @@ jobs: strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-2] + runner: [linux-mi325-2gpu-sglang] part: [0, 1] runs-on: ${{matrix.runner}} steps: @@ -509,7 +509,7 @@ jobs: fail-fast: false max-parallel: 1 # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT matrix: - runner: [linux-mi325-gpu-1] + runner: [linux-mi325-1gpu-sglang] part: [0, 1] # 2 partitions: 11 tests ÷ 2 = ~5-6 tests each runs-on: ${{matrix.runner}} steps: @@ -629,7 +629,7 @@ jobs: fail-fast: false max-parallel: 1 # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT matrix: - runner: [linux-mi325-gpu-2] + runner: [linux-mi325-2gpu-sglang] part: [0, 1] # 2 partitions: 9 tests ÷ 2 = ~4-5 tests each runs-on: ${{matrix.runner}} steps: @@ -754,11 +754,11 @@ jobs: ) ) env: - RUNNER_LABELS: linux-mi325-gpu-8 + RUNNER_LABELS: linux-mi325-8gpu-sglang strategy: fail-fast: false matrix: - runner: [linux-mi325-gpu-8] + runner: [linux-mi325-8gpu-sglang] part: [0, 1, 2] runs-on: ${{matrix.runner}} steps: