[AMD CI] Add 2-GPU sgl-kernel Tests (#17555)

Co-authored-by: YC Tseng <yctseng@amd.com>

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

@@ -152,6 +152,47 @@ jobs:
           docker exec -w /sglang-checkout/sgl-kernel/tests/sgl_diffusion ci_sglang python3 -m pytest test_timestep_embedding.py
           docker exec -w /sglang-checkout/sgl-kernel/tests ci_sglang python3 -m pytest test_moe_topk_sigmoid.py
           docker exec -w /sglang-checkout/sgl-kernel/tests ci_sglang python3 -m pytest test_torch_defaults_reset.py
+
+  sgl-kernel-unit-test-2-gpu-amd:
+    needs: [check-changes]
+    if: |
+      always() &&
+      (
+        (inputs.target_stage == 'sgl-kernel-unit-test-2-gpu-amd') ||
+        (
+          !inputs.target_stage &&
+          needs.check-changes.outputs.sgl_kernel == 'true'
+        )
+      )
+    strategy:
+      fail-fast: false
+      matrix:
+        runner: [linux-mi325-gpu-2]
+    runs-on: ${{matrix.runner}}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.pr_head_sha || inputs.ref || github.sha }}
+
+      - name: Ensure VRAM is clear
+        run: bash scripts/ensure_vram_clear.sh rocm
+
+      - name: Start CI container
+        run: 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: Run test
+        timeout-minutes: 20
+        run: |
+          docker exec -w /sglang-checkout/sgl-kernel/tests ci_sglang python3 -m pytest test_amd_deterministic_custom_allreduce.py
+          docker exec -w /sglang-checkout/sgl-kernel/tests ci_sglang python3 -m pytest test_amd_nccl_allreduce_determinism.py
+
   # =============================================== primary ====================================================
 
   stage-a-test-1-amd:
@@ -842,6 +883,7 @@ jobs:
         check-changes,
 
         sgl-kernel-unit-test-amd,
+        sgl-kernel-unit-test-2-gpu-amd,
         multimodal-gen-test-1-gpu-amd,
         multimodal-gen-test-2-gpu-amd,
 

sgl-kernel/tests/test_amd_deterministic_custom_allreduce.py

@@ -12,12 +12,13 @@ This test compares:
 2. Deterministic kernel (different batch size)
 
 Usage:
-    python test_amd_deterministic_custom_allreduce.py
+    pytest test_amd_deterministic_custom_allreduce.py
 """
 
 import multiprocessing as mp
 import socket
 
+import pytest
 import torch
 import torch.distributed as dist
 
@@ -266,5 +267,14 @@ def main():
         p.join()
 
 
+@pytest.mark.skipif(
+    not torch.cuda.is_available() or torch.cuda.device_count() < 2,
+    reason="Requires at least 2 CUDA GPUs",
+)
+def test_deterministic_custom_allreduce():
+    """Test that deterministic custom all-reduce produces consistent results."""
+    main()
+
+
 if __name__ == "__main__":
     main()

sgl-kernel/tests/test_amd_nccl_allreduce_determinism.py

@@ -10,12 +10,13 @@ This test compares:
 2. Default all-reduce (different batch size) - typically NON-DETERMINISTIC for bfloat16
 
 Usage:
-    python test_ar.py
+    pytest test_amd_nccl_allreduce_determinism.py
 """
 
 import multiprocessing as mp
 import socket
 
+import pytest
 import torch
 import torch.distributed as dist
 
@@ -194,5 +195,14 @@ def main():
         p.join()
 
 
+@pytest.mark.skipif(
+    not torch.cuda.is_available() or torch.cuda.device_count() < 2,
+    reason="Requires at least 2 CUDA GPUs",
+)
+def test_nccl_allreduce_determinism():
+    """Test NCCL all-reduce determinism behavior with varying batch sizes."""
+    main()
+
+
 if __name__ == "__main__":
     main()