Clean up AMD CI (#6365)

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

@@ -27,32 +27,15 @@ jobs:
 
       - name: Setup docker
         run: |
-          # Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
-          if [ -f "/etc/podinfo/gha-render-devices" ]; then
-            DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
-          else
-            DEVICE_FLAG="--device /dev/dri"
-          fi
           touch github_summary.md
-          docker pull lmsysorg/sglang:v0.4.6.post3-rocm630
-          docker run -dt --user root --device=/dev/kfd $DEVICE_FLAG \
-            -v ${{ github.workspace }}:/sglang-checkout --ipc=host --group-add video \
-            --cap-add=SYS_PTRACE -e HF_TOKEN=${HF_TOKEN} --security-opt seccomp=unconfined \
-            -w /sglang-checkout --name ci_sglang \
-            lmsysorg/sglang:v0.4.6.post3-rocm630
+          bash scripts/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
 
       - name: Install dependencies
-        run: |
-          docker exec ci_sglang pip install --upgrade pip
-          docker exec ci_sglang pip uninstall sgl-kernel -y || true
-          docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
-          docker exec ci_sglang pip install -e "python[dev_hip]"
-
-          docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
-          docker exec -w /human-eval ci_sglang pip install -e .
-          docker exec ci_sglang pip install huggingface_hub[hf_xet]
+        run: bash scripts/amd_ci_install_dependency.sh
 
       - name: Nightly Test
         run: |
-          docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" ci_sglang python3 run_suite.py --suite nightly-amd --timeout-per-file 7200
+          bash scripts/amd_ci_exec.sh -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 run_suite.py --suite nightly-amd --timeout-per-file 7200
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY

scripts/amd_ci_exec.sh

+#!/bin/bash
+set -euo pipefail
+
+# Default working directory
+WORKDIR="/sglang-checkout/test/srt"
+ENV_ARGS=(
+  -e SGLANG_AMD_CI=1
+  -e SGLANG_IS_IN_CI=1
+  -e SGLANG_AITER_MOE=1
+)
+
+# Parse optional -w/--workdir and -e ENV=VAL flags
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    -w|--workdir)
+      WORKDIR="$2"
+      shift 2
+      ;;
+    -e)
+      ENV_ARGS+=("-e" "$2")
+      shift 2
+      ;;
+    --)
+      shift
+      break
+      ;;
+    *)
+      break
+      ;;
+  esac
+done
+
+# Run docker exec
+docker exec \
+  -w "$WORKDIR" \
+  "${ENV_ARGS[@]}" \
+  ci_sglang "$@"

scripts/amd_ci_install_dependency.sh

+#!/bin/bash
+set -euo pipefail
+
+# Install the required dependencies in CI.
+docker exec ci_sglang pip install --upgrade pip
+docker exec ci_sglang pip uninstall sgl-kernel -y || true
+docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
+docker exec ci_sglang pip install -e "python[dev_hip]"
+
+docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
+docker exec -w /human-eval ci_sglang pip install -e .
+
+docker exec -w / ci_sglang mkdir -p /dummy-grok
+mkdir -p dummy-grok && wget https://sharkpublic.blob.core.windows.net/sharkpublic/sglang/dummy_grok.json -O dummy-grok/config.json
+docker cp ./dummy-grok ci_sglang:/

scripts/amd_ci_start_container.sh

+#!/bin/bash
+set -euo pipefail
+
+# Set up DEVICE_FLAG based on Kubernetes pod info
+if [ -f "/etc/podinfo/gha-render-devices" ]; then
+  DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
+else
+  DEVICE_FLAG="--device /dev/dri"
+fi
+
+# Pull the image
+IMAGE="lmsysorg/sglang:v0.4.6.post3-rocm630"
+echo "Pulling Docker image: $IMAGE"
+docker pull "$IMAGE"
+
+# Run the container
+echo "Starting container: ci_sglang"
+docker run -dt --user root --device=/dev/kfd $DEVICE_FLAG \
+  -v "${GITHUB_WORKSPACE:-$PWD}:/sglang-checkout" \
+  --ipc=host --group-add video \
+  --cap-add=SYS_PTRACE \
+  -e HF_TOKEN="${HF_TOKEN:-}" \
+  --security-opt seccomp=unconfined \
+  -w /sglang-checkout \
+  --name ci_sglang \
+  "$IMAGE"