[AMD] Add AMD CI registration (1-gpu unit test) to nightly CI. (#16941)

.github/workflows/nightly-test-amd.yml

@@ -17,22 +17,24 @@ on:
         default: 'all'
         options:
           - 'all'
-          - 'nightly-test-2-gpu'
-          - 'nightly-test-2-gpu-vlm'
-          - 'nightly-test-8-gpu-gpt-oss'
-          - 'nightly-test-8-gpu-grok'
-          - 'nightly-test-8-gpu-deepseek-r1'
-          - 'nightly-perf-8-gpu-grok'
-          - 'nightly-perf-8-gpu-deepseek-v3'
-          - 'nightly-perf-8-gpu-deepseek-v31'
+          # MI30x Unit Tests
+          - 'nightly-test-1-gpu-unit'
+          # MI30x Accuracy Tests (GSM8K / MMMU)
+          - 'nightly-accuracy-2-gpu'
+          - 'nightly-accuracy-2-gpu-vlm'
+          - 'nightly-accuracy-8-gpu'
+          - 'nightly-accuracy-8-gpu-deepseek-r1'
+          # MI30x Accuracy + Performance Tests (combined)
+          - 'nightly-8-gpu-grok1-int4'
+          - 'nightly-8-gpu-grok2'
+          - 'nightly-8-gpu-deepseek-v31'
           # MI35x jobs
-          - 'nightly-test-2-gpu-mi35x'
-          - 'nightly-test-2-gpu-vlm-mi35x'
-          - 'nightly-test-8-gpu-mi35x-gpt-oss'
-          - 'nightly-test-8-gpu-mi35x-grok'
-          - 'nightly-test-8-gpu-mi35x-deepseek-r1'
-          - 'nightly-perf-8-gpu-mi35x-grok'
-          - 'nightly-perf-8-gpu-mi35x-deepseek-r1-mxfp4'
+          - 'nightly-test-1-gpu-mi35x'
+          - 'nightly-accuracy-8-gpu-mi35x'
+          - 'nightly-accuracy-8-gpu-mi35x-deepseek-r1'
+          - 'nightly-8-gpu-mi35x-grok1-int4'
+          - 'nightly-8-gpu-mi35x-grok2'
+          - 'nightly-8-gpu-mi35x-deepseek-r1-mxfp4'
   workflow_call:
     inputs:
       ref:
@@ -51,9 +53,40 @@ concurrency:
   cancel-in-progress: ${{ github.event_name != 'workflow_call' }}
 
 jobs:
-  # 2-GPU tests (TP=2)
-  nightly-test-2-gpu:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-2-gpu')
+  # ============================================== MI30x Unit Tests ==============================================
+  # 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
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: bash scripts/ci/amd_ci_install_dependency.sh
+
+      - name: Nightly Unit Test (1-GPU)
+        timeout-minutes: 60
+        run: |
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-1-gpu --nightly --timeout-per-file 600 --continue-on-error || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+  # ============================================== MI30x Accuracy Tests ==============================================
+  # 2-GPU Accuracy Tests - GSM8K eval (MI30x only)
+  nightly-accuracy-2-gpu:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu')
     runs-on: linux-mi325-gpu-2
     steps:
       - name: Checkout code
@@ -79,9 +112,9 @@ jobs:
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
-  # 2-GPU VLM tests - Vision-Language Models MMMU evaluation
-  nightly-test-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-test-2-gpu-vlm')
+  # 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
     steps:
       - name: Checkout code
@@ -99,18 +132,18 @@ jobs:
       - name: Install dependencies
         run: bash scripts/ci/amd_ci_install_dependency.sh
 
-      - name: Nightly Test (2-GPU VLM MMMU)
+      - name: Nightly Accuracy Test (2-GPU VLM MMMU)
         timeout-minutes: 180
         run: |
           bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
             -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-vlm --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-2-gpu-vlm --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
-  # 8-GPU tests (TP=8) - GPT-OSS models
-  nightly-test-8-gpu-gpt-oss:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-8-gpu-gpt-oss')
+  # 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
     steps:
       - name: Checkout code
@@ -128,103 +161,28 @@ jobs:
       - name: Install dependencies
         run: bash scripts/ci/amd_ci_install_dependency.sh
 
-      - name: Nightly Test (8-GPU GPT-OSS)
+      - name: Accuracy Test (8-GPU GPT-OSS)
+        timeout-minutes: 180
         run: |
           bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e AMD_TEST_MODEL_GROUP=gpt-oss \
             -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-gpt-oss --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
-  # 8-GPU tests (TP=8) - GROK models (GROK1-FP8, GROK1-IN4, GROK2.5)
-  nightly-test-8-gpu-grok:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-8-gpu-grok')
-    runs-on: linux-mi325-gpu-8
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: bash scripts/ci/amd_ci_install_dependency.sh
-
-      - name: Nightly Test (8-GPU GROK)
-        run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e AMD_TEST_MODEL_GROUP=grok \
-            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
-          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
-          exit ${TEST_EXIT_CODE:-0}
-
-  # 8-GPU tests (TP=8) - DeepSeek-R1 all variants (basic, MTP, DP, TC)
-  nightly-test-8-gpu-deepseek-r1:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-8-gpu-deepseek-r1')
-    runs-on: linux-mi325-gpu-8
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: bash scripts/ci/amd_ci_install_dependency.sh
-
-      - name: Nightly Test (8-GPU DeepSeek-R1 all variants)
-        run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e AMD_TEST_MODEL_GROUP=deepseek-r1-all \
-            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
-          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
-          exit ${TEST_EXIT_CODE:-0}
-
-  # 8-GPU Performance Tests (TP=8) - Grok (Grok-1 + Grok-2) performance benchmarks
-  nightly-perf-8-gpu-grok:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-grok')
-    runs-on: linux-mi325-gpu-8
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: bash scripts/ci/amd_ci_install_dependency.sh
-
-      - name: Nightly Perf Test (8-GPU Grok-1 + Grok-2)
+      - name: Accuracy Test (8-GPU Grok1-FP8)
         timeout-minutes: 60
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e RCCL_MSCCL_ENABLE=0 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_grok_perf.py || TEST_EXIT_CODE=$?
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-grok1-fp8 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
-  # 8-GPU Performance Tests (TP=8) - DeepSeek-V3 performance benchmarks
-  nightly-perf-8-gpu-deepseek-v3:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-deepseek-v3')
+  # 8-GPU DeepSeek-R1 Accuracy Test (separate job due to long loading time)
+  nightly-accuracy-8-gpu-deepseek-r1:
+    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-deepseek-r1')
     runs-on: linux-mi325-gpu-8
     steps:
       - name: Checkout code
@@ -242,16 +200,19 @@ jobs:
       - name: Install dependencies
         run: bash scripts/ci/amd_ci_install_dependency.sh
 
-      - name: Nightly Perf Test (8-GPU DeepSeek-V3)
-        timeout-minutes: 300
+      - name: Accuracy Test (8-GPU DeepSeek-R1)
+        timeout-minutes: 240
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e SGLANG_USE_ROCM700A=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_deepseek_v3_perf.py || TEST_EXIT_CODE=$?
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-deepseek-r1 --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
-  # 8-GPU Performance Tests (TP=8) - DeepSeek-V3.1 performance benchmarks
-  nightly-perf-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-perf-8-gpu-deepseek-v31')
+  # ============================================== MI30x Combined Accuracy + Performance Tests ==============================================
+  # 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
     steps:
       - name: Checkout code
@@ -269,18 +230,120 @@ jobs:
       - name: Install dependencies
         run: bash scripts/ci/amd_ci_install_dependency.sh
 
-      - name: Nightly Perf Test (8-GPU DeepSeek-V3.1)
-        timeout-minutes: 300
+      - name: Accuracy Test (8-GPU Grok1-INT4)
+        timeout-minutes: 60
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e SGLANG_USE_ROCM700A=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_deepseek_v31_perf.py || TEST_EXIT_CODE=$?
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-grok1-int4 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+      - name: Performance Test (8-GPU Grok1-INT4)
+        timeout-minutes: 60
+        continue-on-error: true  # Perf test failure doesn't fail the job if accuracy passed
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-perf-8-gpu-grok1-int4 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+  # 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
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: bash scripts/ci/amd_ci_install_dependency.sh
+
+      - name: Accuracy Test (8-GPU Grok2)
+        timeout-minutes: 60
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-grok2 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+      - name: Performance Test (8-GPU Grok2)
+        timeout-minutes: 60
+        continue-on-error: true  # Perf test failure doesn't fail the job if accuracy passed
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-perf-8-gpu-grok2 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+  # 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
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: bash scripts/ci/amd_ci_install_dependency.sh
+
+      - name: Accuracy Test (8-GPU DeepSeek-V3.1)
+        timeout-minutes: 120
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e SGLANG_USE_AITER=1 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-deepseek-v31 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+      - name: Performance Test (8-GPU DeepSeek-V3.1)
+        timeout-minutes: 300
+        continue-on-error: true  # Perf test failure doesn't fail the job if accuracy passed
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e SGLANG_USE_ROCM700A=1 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-perf-8-gpu-deepseek-v31 --nightly --timeout-per-file 18000 || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
   # ============================================== MI35x Tests ==============================================
-  # MI35x 2-GPU tests (TP=2) - Reuses nightly-amd suite
-  nightly-test-2-gpu-mi35x:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-2-gpu-mi35x')
-    runs-on: linux-mi35x-gpu-2
+  # MI35x 1-GPU tests - platform-agnostic tests that may work on CDNA4 (gfx950)
+  nightly-test-1-gpu-mi35x:
+    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-mi35x')
+    runs-on: linux-mi35x-gpu-1
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -300,179 +363,18 @@ jobs:
           # Install tabulate for run_suite.py (missing in MI35x container)
           bash scripts/ci/amd_ci_exec.sh pip install tabulate
 
-      - name: Nightly Test (2-GPU)
-        run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
-          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
-          exit ${TEST_EXIT_CODE:-0}
-
-  # MI35x 2-GPU VLM tests - Reuses nightly-amd-vlm suite
-  nightly-test-2-gpu-vlm-mi35x:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-2-gpu-vlm-mi35x')
-    runs-on: linux-mi35x-gpu-2
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: |
-          bash scripts/ci/amd_ci_install_dependency.sh
-          # Install tabulate for run_suite.py (missing in MI35x container)
-          bash scripts/ci/amd_ci_exec.sh pip install tabulate
-
-      - name: Nightly Test (2-GPU VLM MMMU)
-        timeout-minutes: 180
-        run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-vlm --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
-          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
-          exit ${TEST_EXIT_CODE:-0}
-
-  # MI35x 8-GPU tests (TP=8) - GPT-OSS models (MI35x uses openai/* paths)
-  nightly-test-8-gpu-mi35x-gpt-oss:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-8-gpu-mi35x-gpt-oss')
-    runs-on: linux-mi35x-gpu-8
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: |
-          bash scripts/ci/amd_ci_install_dependency.sh
-          # Install tabulate for run_suite.py (missing in MI35x container)
-          bash scripts/ci/amd_ci_exec.sh pip install tabulate
-
-      - name: Nightly Test MI35x (8-GPU GPT-OSS)
-        timeout-minutes: 180
-        run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e AMD_TEST_MODEL_GROUP=gpt-oss \
-            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
-          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
-          exit ${TEST_EXIT_CODE:-0}
-
-  # MI35x 8-GPU tests (TP=8) - GROK models
-  nightly-test-8-gpu-mi35x-grok:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-8-gpu-mi35x-grok')
-    runs-on: linux-mi35x-gpu-8
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: |
-          bash scripts/ci/amd_ci_install_dependency.sh
-          # Install tabulate for run_suite.py (missing in MI35x container)
-          bash scripts/ci/amd_ci_exec.sh pip install tabulate
-
-      - name: Nightly Test MI35x (8-GPU GROK)
-        run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e AMD_TEST_MODEL_GROUP=grok \
-            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
-          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
-          exit ${TEST_EXIT_CODE:-0}
-
-  # MI35x 8-GPU tests (TP=8) - DeepSeek-R1-0528 basic + MTP only
-  # Same model as MI300X for consistency; MXFP4 only used for perf tests
-  # Note: DP/TC variants disabled for MI35x due to initialization timeouts
-  nightly-test-8-gpu-mi35x-deepseek-r1:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-8-gpu-mi35x-deepseek-r1')
-    runs-on: linux-mi35x-gpu-8
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: |
-          bash scripts/ci/amd_ci_install_dependency.sh
-          # Install tabulate for run_suite.py (missing in MI35x container)
-          bash scripts/ci/amd_ci_exec.sh pip install tabulate
-
-      - name: Nightly Test MI35x (8-GPU DeepSeek-R1-0528 basic + MTP)
-        timeout-minutes: 180
-        run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
-            -e AMD_TEST_MODEL_GROUP=deepseek-r1 \
-            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
-            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
-          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
-          exit ${TEST_EXIT_CODE:-0}
-
-  # MI35x 8-GPU Performance Tests (TP=8) - Grok performance benchmarks
-  nightly-perf-8-gpu-mi35x-grok:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-mi35x-grok')
-    runs-on: linux-mi35x-gpu-8
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Setup docker
-        run: |
-          touch github_summary.md
-          bash scripts/ci/amd_ci_start_container.sh
-        env:
-          GITHUB_WORKSPACE: ${{ github.workspace }}
-
-      - name: Install dependencies
-        run: |
-          bash scripts/ci/amd_ci_install_dependency.sh
-          # Install tabulate for run_suite.py (missing in MI35x container)
-          bash scripts/ci/amd_ci_exec.sh pip install tabulate
-
-      - name: Nightly Perf Test MI35x (8-GPU Grok)
+      - name: Nightly Test MI35x (1-GPU)
         timeout-minutes: 60
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e RCCL_MSCCL_ENABLE=0 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_grok_perf.py || TEST_EXIT_CODE=$?
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-1-gpu-mi35x --nightly --timeout-per-file 600 --continue-on-error || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
-  # MI35x 8-GPU Performance Tests (TP=8) - DeepSeek-R1-MXFP4 performance benchmarks
-  nightly-perf-8-gpu-mi35x-deepseek-r1-mxfp4:
-    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-mi35x-deepseek-r1-mxfp4')
+  # MI35x 8-GPU Accuracy Tests - GPT-OSS (accuracy only)
+  nightly-accuracy-8-gpu-mi35x:
+    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-mi35x')
     runs-on: linux-mi35x-gpu-8
     steps:
       - name: Checkout code
@@ -493,33 +395,204 @@ jobs:
           # Install tabulate for run_suite.py (missing in MI35x container)
           bash scripts/ci/amd_ci_exec.sh pip install tabulate
 
-      - name: Nightly Perf Test MI35x (8-GPU DeepSeek-R1-MXFP4)
-        timeout-minutes: 300
+      - name: Accuracy Test MI35x (8-GPU GPT-OSS)
+        timeout-minutes: 180
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_deepseek_r1_mxfp4_perf.py || TEST_EXIT_CODE=$?
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+  # MI35x 8-GPU DeepSeek-R1-0528 Accuracy Test (separate job due to long loading time)
+  nightly-accuracy-8-gpu-mi35x-deepseek-r1:
+    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-mi35x-deepseek-r1')
+    runs-on: linux-mi35x-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd_ci_install_dependency.sh
+          # Install tabulate for run_suite.py (missing in MI35x container)
+          bash scripts/ci/amd_ci_exec.sh pip install tabulate
+
+      - name: Accuracy Test MI35x (8-GPU DeepSeek-R1-0528)
+        timeout-minutes: 240
+        run: |
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-mi35x-deepseek-r1 --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+  # MI35x 8-GPU Grok1-INT4 (Accuracy + Performance combined)
+  nightly-8-gpu-mi35x-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-mi35x-grok1-int4')
+    runs-on: linux-mi35x-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd_ci_install_dependency.sh
+          # Install tabulate for run_suite.py (missing in MI35x container)
+          bash scripts/ci/amd_ci_exec.sh pip install tabulate
+
+      - name: Accuracy Test MI35x (8-GPU Grok1-INT4)
+        timeout-minutes: 60
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-mi35x-grok1-int4 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+      - name: Performance Test MI35x (8-GPU Grok1-INT4)
+        timeout-minutes: 60
+        continue-on-error: true  # Perf test failure doesn't fail the job if accuracy passed
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-perf-8-gpu-mi35x-grok1-int4 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+  # MI35x 8-GPU Grok2 (Accuracy + Performance combined)
+  nightly-8-gpu-mi35x-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-mi35x-grok2')
+    runs-on: linux-mi35x-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd_ci_install_dependency.sh
+          # Install tabulate for run_suite.py (missing in MI35x container)
+          bash scripts/ci/amd_ci_exec.sh pip install tabulate
+
+      - name: Accuracy Test MI35x (8-GPU Grok2)
+        timeout-minutes: 60
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-mi35x-grok2 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+      - name: Performance Test MI35x (8-GPU Grok2)
+        timeout-minutes: 60
+        continue-on-error: true  # Perf test failure doesn't fail the job if accuracy passed
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e RCCL_MSCCL_ENABLE=0 \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-perf-8-gpu-mi35x-grok2 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+  # MI35x 8-GPU DeepSeek-R1-MXFP4 (Accuracy + Performance combined)
+  nightly-8-gpu-mi35x-deepseek-r1-mxfp4:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-deepseek-r1-mxfp4')
+    runs-on: linux-mi35x-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd_ci_install_dependency.sh
+          # Install tabulate for run_suite.py (missing in MI35x container)
+          bash scripts/ci/amd_ci_exec.sh pip install tabulate
+
+      - name: Accuracy Test MI35x (8-GPU DeepSeek-R1-MXFP4)
+        timeout-minutes: 180
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-r1-mxfp4 --nightly --timeout-per-file 7200 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
+      - name: Performance Test MI35x (8-GPU DeepSeek-R1-MXFP4)
+        timeout-minutes: 300
+        continue-on-error: true  # Perf test failure doesn't fail the job if accuracy passed
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 registered/amd/perf/mi35x/test_deepseek_r1_mxfp4_perf_mi35x.py || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
   check-all-jobs:
     if: always() && (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request' || github.event_name == 'workflow_dispatch')
     needs:
-      # MI325 jobs
-      - nightly-test-2-gpu
-      - nightly-test-2-gpu-vlm
-      - nightly-test-8-gpu-gpt-oss
-      - nightly-test-8-gpu-grok
-      - nightly-test-8-gpu-deepseek-r1
-      - nightly-perf-8-gpu-grok
-      - nightly-perf-8-gpu-deepseek-v3
-      - nightly-perf-8-gpu-deepseek-v31
+      # MI30x Unit Tests
+      - nightly-test-1-gpu-unit
+      # MI30x Accuracy Tests
+      - nightly-accuracy-2-gpu
+      - nightly-accuracy-2-gpu-vlm
+      - nightly-accuracy-8-gpu
+      - nightly-accuracy-8-gpu-deepseek-r1
+      # MI30x Combined Accuracy + Performance Tests
+      - nightly-8-gpu-grok1-int4
+      - nightly-8-gpu-grok2
+      - nightly-8-gpu-deepseek-v31
       # MI35x jobs
-      - nightly-test-2-gpu-mi35x
-      - nightly-test-2-gpu-vlm-mi35x
-      - nightly-test-8-gpu-mi35x-gpt-oss
-      - nightly-test-8-gpu-mi35x-grok
-      - nightly-test-8-gpu-mi35x-deepseek-r1
-      - nightly-perf-8-gpu-mi35x-grok
-      - nightly-perf-8-gpu-mi35x-deepseek-r1-mxfp4
+      - nightly-test-1-gpu-mi35x
+      - nightly-accuracy-8-gpu-mi35x
+      - nightly-accuracy-8-gpu-mi35x-deepseek-r1
+      - nightly-8-gpu-mi35x-grok1-int4
+      - nightly-8-gpu-mi35x-grok2
+      - nightly-8-gpu-mi35x-deepseek-r1-mxfp4
     runs-on: ubuntu-latest
     steps:
       - name: Check if any job failed