[AMD] Fix CI and add retry logic for git clone timeout (#15663)

Co-authored-by: Bingxu Chen <bingxche@amd.com>

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

@@ -182,7 +182,7 @@ jobs:
       - name: Run test
         timeout-minutes: 10
         run: |
-          docker exec -w /sglang-checkout/test ci_sglang python3 run_suite.py --hw amd --suite stage-a-test-1
+          bash scripts/ci/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-a-test-1
 
   stage-b-test-small-1-gpu-amd:
     needs: [check-changes, stage-a-test-1-amd]
@@ -222,7 +222,7 @@ jobs:
       - name: Run test
         timeout-minutes: 30
         run: |
-          docker exec -w /sglang-checkout/test ci_sglang python3 run_suite.py --hw amd --suite stage-b-test-small-1-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 4
+          bash scripts/ci/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-small-1-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 4
 
   multimodal-gen-test-1-gpu-amd:
     needs: [check-changes]

scripts/ci/amd_ci_install_dependency.sh

@@ -60,6 +60,41 @@ install_with_retry() {
   return 1
 }
 
+# Helper function to git clone with retries
+git_clone_with_retry() {
+  local repo_url="$1"
+  local dest_dir="${2:-}"
+  local branch_args="${3:-}"
+  local max_attempts=3
+
+  for attempt in $(seq 1 $max_attempts); do
+    echo "Git clone attempt $attempt/$max_attempts: $repo_url"
+
+    # prevent from partial clone
+    if [ -n "$dest_dir" ] && [ -d "$dest_dir" ]; then
+      rm -rf "$dest_dir"
+    fi
+
+    if git \
+      -c http.lowSpeedLimit=1000 \
+      -c http.lowSpeedTime=30 \
+      clone --depth 1 ${branch_args:+$branch_args} "$repo_url" "$dest_dir"; then
+      echo "Git clone succeeded."
+      return 0
+    fi
+
+    if [ $attempt -lt $max_attempts ]; then
+      echo "Git clone failed, retrying in 5 seconds..."
+      sleep 5
+    fi
+  done
+
+  echo "Git clone failed after $max_attempts attempts: $repo_url"
+  return 1
+}
+
+
+
 case "${GPU_ARCH}" in
   mi35x)
     echo "Runner uses ${GPU_ARCH}; will fetch mi35x image."
@@ -82,7 +117,9 @@ case "${GPU_ARCH}" in
     ;;
 esac
 
-docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
+#docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
+git_clone_with_retry https://github.com/merrymercy/human-eval.git human-eval
+docker cp human-eval ci_sglang:/
 install_with_retry docker exec -w /human-eval ci_sglang pip install --cache-dir=/sgl-data/pip-cache -e .
 
 docker exec -w / ci_sglang mkdir -p /dummy-grok