From 769ac5c99daafb49f63b0ec7cdcf6ab3c166850a Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 11 May 2026 14:23:25 +0800
Subject: [PATCH 01/19] [Fix] user container env

---
 .github/workflows/push_image_harbor.yml | 73 +++++++++++++++++--------
 1 file changed, 51 insertions(+), 22 deletions(-)

diff --git a/.github/workflows/push_image_harbor.yml b/.github/workflows/push_image_harbor.yml
index 118cb4d033..38a4454d13 100644
--- a/.github/workflows/push_image_harbor.yml
+++ b/.github/workflows/push_image_harbor.yml
@@ -25,6 +25,13 @@ jobs:
   prepare:
     name: Detect tar files for ${{ github.sha }}
     runs-on: [self-hosted, Linux, X64, nvidia-0, gpus-8]
+    timeout-minutes: 15
+    container:
+      image: harbor.baai.ac.cn/flagscale/flagscale-train:dev-cu128-py3.12-20260228210721
+      volumes:
+        - /data/image_tar:/data/image_tar
+      options: >-
+        --user root
     outputs:
       needs_promotion: ${{ steps.detect.outputs.needs_promotion }}
       train_tar: ${{ steps.detect.outputs.train_tar }}
@@ -42,7 +49,14 @@ jobs:
         id: config
         run: |
           TAR_DIR=$(grep '^tar_dir:' .github/configs/cuda.yml | awk '{print $2}')
-          echo "tar_dir=${TAR_DIR}" >> $GITHUB_OUTPUT
+          echo "tar_dir=${TAR_DIR}" >> "$GITHUB_OUTPUT"
+
+      - name: Install gh CLI
+        run: |
+          if ! command -v gh &>/dev/null; then
+            apt-get update && apt-get install -y gh
+          fi
+          gh --version
 
       - name: Resolve PR number from merge commit
         id: resolve_pr
@@ -61,10 +75,10 @@ jobs:
 
           if [ -n "$PR_NUMBER" ]; then
             echo "Found PR #${PR_NUMBER} for merge commit ${MERGE_SHA}"
-            echo "suffix=pr${PR_NUMBER}" >> $GITHUB_OUTPUT
+            echo "suffix=pr${PR_NUMBER}" >> "$GITHUB_OUTPUT"
           else
             echo "::warning::Could not find PR for merge commit ${MERGE_SHA}, falling back to short SHA"
-            echo "suffix=${MERGE_SHA:0:7}" >> $GITHUB_OUTPUT
+            echo "suffix=${MERGE_SHA:0:7}" >> "$GITHUB_OUTPUT"
           fi
 
       - name: Find tar files by PR number
@@ -78,29 +92,30 @@ jobs:
           INFERENCE_TAR=$(find "$TAR_DIR" -maxdepth 1 -name "*flagscale-inference*-${SUFFIX}.tar" 2>/dev/null | sort -r | head -1)
           ALL_TAR=$(find "$TAR_DIR" -maxdepth 1 -name "*flagscale-all*-${SUFFIX}.tar" 2>/dev/null | sort -r | head -1)
 
-          echo "train_tar=${TRAIN_TAR}" >> $GITHUB_OUTPUT
-          echo "inference_tar=${INFERENCE_TAR}" >> $GITHUB_OUTPUT
-          echo "all_tar=${ALL_TAR}" >> $GITHUB_OUTPUT
+          echo "train_tar=${TRAIN_TAR}" >> "$GITHUB_OUTPUT"
+          echo "inference_tar=${INFERENCE_TAR}" >> "$GITHUB_OUTPUT"
+          echo "all_tar=${ALL_TAR}" >> "$GITHUB_OUTPUT"
 
           if [ -n "$TRAIN_TAR" ] || [ -n "$INFERENCE_TAR" ] || [ -n "$ALL_TAR" ]; then
-            echo "needs_promotion=true" >> $GITHUB_OUTPUT
+            echo "needs_promotion=true" >> "$GITHUB_OUTPUT"
             echo "Detected tars for ${SUFFIX}:"
             if [ -n "$TRAIN_TAR" ];     then echo "  train:     $TRAIN_TAR"; fi
             if [ -n "$INFERENCE_TAR" ]; then echo "  inference: $INFERENCE_TAR"; fi
             if [ -n "$ALL_TAR" ];       then echo "  all:       $ALL_TAR"; fi
           else
-            echo "needs_promotion=false" >> $GITHUB_OUTPUT
+            echo "needs_promotion=false" >> "$GITHUB_OUTPUT"
             echo "::warning::No tar files found for ${SUFFIX} in $TAR_DIR, skipping promotion"
           fi
 
   # ---------------------------------------------------------------------------
-  # Promote: load tar → retag → push to Harbor → delete tar
+  # Promote: load tar → retag → push to Harbor
   # ---------------------------------------------------------------------------
   promote:
     name: Push validated images to Harbor
     needs: prepare
     if: needs.prepare.outputs.needs_promotion == 'true'
     runs-on: [self-hosted, Linux, X64, nvidia-0, gpus-8]
+    timeout-minutes: 60
     outputs:
       remote_train_tag: ${{ steps.promote_train.outputs.remote_tag }}
       remote_inference_tag: ${{ steps.promote_inference.outputs.remote_tag }}
@@ -121,16 +136,19 @@ jobs:
           TAR_PATH="${{ needs.prepare.outputs.train_tar }}"
 
           echo "Loading tar: $TAR_PATH"
-          LOCAL_TAG=$(sudo docker load -i "$TAR_PATH" | grep 'Loaded image:' | awk '{print $NF}')
+          LOCAL_TAG=$(sudo docker load -i "$TAR_PATH" | grep -oP '(?<=Loaded image: ).+' | tail -1)
+          if [ -z "$LOCAL_TAG" ]; then
+            echo "::error::Failed to parse image tag from docker load output"
+            exit 1
+          fi
           echo "Loaded image tag: $LOCAL_TAG"
 
-          # Strip local registry prefix (e.g. localhost:5000/) to get image:tag
           IMAGE_AND_TAG="${LOCAL_TAG#*/}"
           REMOTE_TAG="${{ env.REMOTE_REGISTRY }}/${{ env.REMOTE_IMAGE_PREFIX }}/${IMAGE_AND_TAG}"
 
           sudo docker tag "${LOCAL_TAG}" "${REMOTE_TAG}"
           sudo docker push "${REMOTE_TAG}"
-          echo "remote_tag=${REMOTE_TAG}" >> $GITHUB_OUTPUT
+          echo "remote_tag=${REMOTE_TAG}" >> "$GITHUB_OUTPUT"
           echo "Pushed: ${REMOTE_TAG}"
 
       - name: Promote inference image to Harbor
@@ -141,7 +159,11 @@ jobs:
           TAR_PATH="${{ needs.prepare.outputs.inference_tar }}"
 
           echo "Loading tar: $TAR_PATH"
-          LOCAL_TAG=$(sudo docker load -i "$TAR_PATH" | grep 'Loaded image:' | awk '{print $NF}')
+          LOCAL_TAG=$(sudo docker load -i "$TAR_PATH" | grep -oP '(?<=Loaded image: ).+' | tail -1)
+          if [ -z "$LOCAL_TAG" ]; then
+            echo "::error::Failed to parse image tag from docker load output"
+            exit 1
+          fi
           echo "Loaded image tag: $LOCAL_TAG"
 
           IMAGE_AND_TAG="${LOCAL_TAG#*/}"
@@ -149,7 +171,7 @@ jobs:
 
           sudo docker tag "${LOCAL_TAG}" "${REMOTE_TAG}"
           sudo docker push "${REMOTE_TAG}"
-          echo "remote_tag=${REMOTE_TAG}" >> $GITHUB_OUTPUT
+          echo "remote_tag=${REMOTE_TAG}" >> "$GITHUB_OUTPUT"
           echo "Pushed: ${REMOTE_TAG}"
 
       - name: Promote all image to Harbor
@@ -160,7 +182,11 @@ jobs:
           TAR_PATH="${{ needs.prepare.outputs.all_tar }}"
 
           echo "Loading tar: $TAR_PATH"
-          LOCAL_TAG=$(sudo docker load -i "$TAR_PATH" | grep 'Loaded image:' | awk '{print $NF}')
+          LOCAL_TAG=$(sudo docker load -i "$TAR_PATH" | grep -oP '(?<=Loaded image: ).+' | tail -1)
+          if [ -z "$LOCAL_TAG" ]; then
+            echo "::error::Failed to parse image tag from docker load output"
+            exit 1
+          fi
           echo "Loaded image tag: $LOCAL_TAG"
 
           IMAGE_AND_TAG="${LOCAL_TAG#*/}"
@@ -168,7 +194,7 @@ jobs:
 
           sudo docker tag "${LOCAL_TAG}" "${REMOTE_TAG}"
           sudo docker push "${REMOTE_TAG}"
-          echo "remote_tag=${REMOTE_TAG}" >> $GITHUB_OUTPUT
+          echo "remote_tag=${REMOTE_TAG}" >> "$GITHUB_OUTPUT"
           echo "Pushed: ${REMOTE_TAG}"
 
   # ---------------------------------------------------------------------------
@@ -178,6 +204,7 @@ jobs:
     name: Update cuda.yml with Harbor image tags
     needs: promote
     runs-on: ubuntu-latest
+    timeout-minutes: 10
     permissions:
       contents: write
     steps:
@@ -185,6 +212,7 @@ jobs:
         uses: actions/checkout@v4
         with:
           token: ${{ secrets.GITHUB_TOKEN }}
+          ref: main
 
       - name: Update image tags in cuda.yml
         run: |
@@ -229,7 +257,8 @@ jobs:
     name: Clean up build cache
     needs: ['prepare', 'promote']
     runs-on: [self-hosted, Linux, X64, nvidia-0, gpus-8]
-    if: always()
+    timeout-minutes: 10
+    if: always() && needs.prepare.result == 'success'
     steps:
       - name: Remove stale tar files for this PR
         run: |
@@ -241,20 +270,20 @@ jobs:
           fi
 
       - name: Remove dangling images
-        run: docker image prune -f 2>/dev/null || true
+        run: sudo docker image prune -f 2>/dev/null || true
 
       - name: Remove build cache older than 7 days
-        run: docker builder prune -f --filter "until=168h" 2>/dev/null || true
+        run: sudo docker builder prune -f --filter "until=168h" 2>/dev/null || true
 
       - name: Remove old localhost registry images
         run: |
-          docker images --format '{{.Repository}}:{{.Tag}} {{.CreatedSince}}' \
+          sudo docker images --format '{{.Repository}}:{{.Tag}} {{.CreatedSince}}' \
             | grep 'localhost:5000' \
             | grep -E '(weeks|months)' \
             | awk '{print $1}' \
-            | xargs -r docker rmi 2>/dev/null || true
+            | xargs -r sudo docker rmi 2>/dev/null || true
 
       - name: Report disk usage
         run: |
           echo "Docker disk usage:"
-          docker system df
+          sudo docker system df

From 5a0deebbb07c5f35d59f307c24c5f70452774c5f Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 11 May 2026 14:34:04 +0800
Subject: [PATCH 02/19] [Fix] user container env

---
 .github/workflows/push_image_harbor.yml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/.github/workflows/push_image_harbor.yml b/.github/workflows/push_image_harbor.yml
index 38a4454d13..e2b2043656 100644
--- a/.github/workflows/push_image_harbor.yml
+++ b/.github/workflows/push_image_harbor.yml
@@ -10,6 +10,7 @@ on:
       - 'tools/install/**'
       - 'requirements/**'
       - '.github/workflows/build_image_cuda.yml'
+  workflow_dispatch:
 
 permissions:
   contents: read

From 5c74ab1fc681a74f8efbed2e149e124d41cbd2b9 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Fri, 15 May 2026 16:39:56 +0800
Subject: [PATCH 03/19] [cicd] test dind

---
 .github/workflows/demo_inference_test.yml | 172 ++++++++++++++++++++++
 1 file changed, 172 insertions(+)
 create mode 100644 .github/workflows/demo_inference_test.yml

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
new file mode 100644
index 0000000000..1b47363746
--- /dev/null
+++ b/.github/workflows/demo_inference_test.yml
@@ -0,0 +1,172 @@
+name: Demo Inference Test - Qwen3
+
+# Demo workflow to test the test-lyz-infer runner with real Qwen3 inference
+# Uses FlagScale inference tests from tests/functional_tests/inference/qwen3
+
+on:
+  workflow_dispatch:
+    inputs:
+      test_case:
+        description: 'Test case to run'
+        required: false
+        default: '4b_tp2'
+        type: choice
+        options:
+          - 4b_tp2
+          - 4b_tp2_ascend
+
+jobs:
+  demo_inference:
+    name: Qwen3 Inference Demo
+    runs-on: test-lyz-infer
+    env:
+      PROJECT_ROOT: ${{ github.workspace }}
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Print test info
+        run: |
+          echo "=========================================="
+          echo "Qwen3 Inference Demo Test"
+          echo "=========================================="
+          echo "Runner: test-lyz-infer"
+          echo "Test Case: ${{ inputs.test_case }}"
+          echo "Workflow: ${{ github.workflow }}"
+          echo "Run ID: ${{ github.run_id }}"
+          echo "Project Root: $PROJECT_ROOT"
+          echo "=========================================="
+
+      - name: Check system info
+        run: |
+          echo "=== System Information ==="
+          echo "Hostname: $(hostname)"
+          echo "OS: $(uname -s)"
+          echo "Kernel: $(uname -r)"
+          echo "Architecture: $(uname -m)"
+          echo ""
+          echo "=== CPU Information ==="
+          lscpu | grep -E "Model name|CPU\(s\)|Thread|Core" || echo "lscpu not available"
+          echo ""
+          echo "=== Memory Information ==="
+          free -h || echo "free command not available"
+          echo ""
+          echo "=== Disk Space ==="
+          df -h / || echo "df command not available"
+
+      - name: Check GPU availability
+        run: |
+          echo "=== GPU Information ==="
+          if command -v nvidia-smi &> /dev/null; then
+            echo "NVIDIA GPU detected:"
+            nvidia-smi
+          else
+            echo "nvidia-smi not found - no NVIDIA GPU or driver not installed"
+          fi
+
+          if command -v npu-smi &> /dev/null; then
+            echo "Ascend NPU detected:"
+            npu-smi info
+          else
+            echo "npu-smi not found - no Ascend NPU"
+          fi
+
+      - name: Check Python environment
+        run: |
+          echo "=== Python Environment ==="
+          if command -v python &> /dev/null; then
+            echo "Python version: $(python --version)"
+            echo "Python location: $(which python)"
+          else
+            echo "Python not found in PATH"
+          fi
+
+      - name: Install FlagScale
+        run: |
+          echo "=== Installing FlagScale ==="
+          cd $PROJECT_ROOT
+          pip install . --no-build-isolation || { echo "❌ FlagScale install failed"; exit 1; }
+
+          # Verify installation
+          command -v flagscale || { echo "❌ FlagScale CLI not found in PATH"; exit 1; }
+          echo "✅ FlagScale CLI installed: $(flagscale --version 2>/dev/null || echo 'version unknown')"
+
+      - name: Run Qwen3 inference test
+        id: inference_test
+        run: |
+          set -euo pipefail
+          cd $PROJECT_ROOT
+
+          TEST_CASE="${{ inputs.test_case }}"
+          TEST_DIR="tests/functional_tests/inference/qwen3"
+          CONFIG_FILE="$TEST_DIR/conf/${TEST_CASE}.yaml"
+          RESULTS_DIR="$TEST_DIR/test_results/${TEST_CASE}"
+          GOLD_FILE="$TEST_DIR/results_gold/${TEST_CASE}"
+
+          echo "=== Running Qwen3 Inference Test ==="
+          echo "Test case: $TEST_CASE"
+          echo "Config: $CONFIG_FILE"
+          echo "Results dir: $RESULTS_DIR"
+          echo ""
+
+          # Check if config exists
+          if [ ! -f "$CONFIG_FILE" ]; then
+            echo "❌ Config file not found: $CONFIG_FILE"
+            exit 1
+          fi
+
+          # Create results directory
+          mkdir -p "$RESULTS_DIR"
+
+          # Run inference using flagscale
+          echo "Starting inference..."
+          flagscale run --config-path "$TEST_DIR/conf" --config-name "${TEST_CASE}" || {
+            echo "❌ Inference failed"
+            exit 1
+          }
+
+          echo "✅ Inference completed"
+
+          # Check results
+          if [ -f "$RESULTS_DIR/output.txt" ]; then
+            echo ""
+            echo "=== Inference Output ==="
+            cat "$RESULTS_DIR/output.txt"
+            echo ""
+
+            # Compare with gold results if available
+            if [ -f "$GOLD_FILE" ]; then
+              echo "=== Comparing with gold results ==="
+              if diff -u "$GOLD_FILE" "$RESULTS_DIR/output.txt"; then
+                echo "✅ Results match gold standard"
+              else
+                echo "⚠️  Results differ from gold standard (this may be expected)"
+              fi
+            fi
+          else
+            echo "⚠️  Output file not found at $RESULTS_DIR/output.txt"
+          fi
+
+          echo ""
+          echo "=========================================="
+          echo "✅ Qwen3 inference test completed!"
+          echo "=========================================="
+
+      - name: Upload test results
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: qwen3-inference-results-${{ inputs.test_case }}-${{ github.run_id }}
+          path: tests/functional_tests/inference/qwen3/test_results/${{ inputs.test_case }}
+          retention-days: 7
+          if-no-files-found: warn
+
+      - name: Test summary
+        if: always()
+        run: |
+          echo "=== Test Summary ==="
+          echo "Status: ${{ job.status }}"
+          echo "Runner: test-lyz-infer"
+          echo "Test Case: ${{ inputs.test_case }}"
+          echo "Timestamp: $(date -u '+%Y-%m-%d %H:%M:%S UTC')"

From d43e7eea919a427180ef85a759f40af26c39ba0f Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Fri, 15 May 2026 16:42:27 +0800
Subject: [PATCH 04/19] [cicd] test dind

---
 .github/workflows/demo_inference_test.yml | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index 1b47363746..7427c124b4 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -4,6 +4,8 @@ name: Demo Inference Test - Qwen3
 # Uses FlagScale inference tests from tests/functional_tests/inference/qwen3
 
 on:
+  pull_request:
+    branches: ["main"]
   workflow_dispatch:
     inputs:
       test_case:

From c12fcf51133960f25046523916af8bc97fd311df Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Fri, 15 May 2026 17:35:39 +0800
Subject: [PATCH 05/19] [cicd] test dind

---
 .github/workflows/demo_inference_test.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index 7427c124b4..e97c373290 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -11,7 +11,7 @@ on:
       test_case:
         description: 'Test case to run'
         required: false
-        default: '4b_tp2'
+        default: '4b_tp2_ascend'
         type: choice
         options:
           - 4b_tp2

From 0d23b4d905898d371934462ed7cc0d9fa885fb24 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 10:48:59 +0800
Subject: [PATCH 06/19] test

---
 .github/workflows/demo_inference_test.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index e97c373290..1642f77b59 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -69,7 +69,7 @@ jobs:
 
           if command -v npu-smi &> /dev/null; then
             echo "Ascend NPU detected:"
-            npu-smi info
+            # npu-smi info
           else
             echo "npu-smi not found - no Ascend NPU"
           fi

From ba372ecc5f2947160778fdec4a2350223962b807 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 15:44:41 +0800
Subject: [PATCH 07/19] test

---
 .github/workflows/demo_inference_test.yml | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index 1642f77b59..e4684672ba 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -34,7 +34,7 @@ jobs:
           echo "Qwen3 Inference Demo Test"
           echo "=========================================="
           echo "Runner: test-lyz-infer"
-          echo "Test Case: ${{ inputs.test_case }}"
+          echo "Test Case: ${{ inputs.test_case || '4b_tp2_ascend' }}"
           echo "Workflow: ${{ github.workflow }}"
           echo "Run ID: ${{ github.run_id }}"
           echo "Project Root: $PROJECT_ROOT"
@@ -69,7 +69,7 @@ jobs:
 
           if command -v npu-smi &> /dev/null; then
             echo "Ascend NPU detected:"
-            # npu-smi info
+            npu-smi info
           else
             echo "npu-smi not found - no Ascend NPU"
           fi
@@ -100,7 +100,7 @@ jobs:
           set -euo pipefail
           cd $PROJECT_ROOT
 
-          TEST_CASE="${{ inputs.test_case }}"
+          TEST_CASE="${{ inputs.test_case || '4b_tp2_ascend' }}"
           TEST_DIR="tests/functional_tests/inference/qwen3"
           CONFIG_FILE="$TEST_DIR/conf/${TEST_CASE}.yaml"
           RESULTS_DIR="$TEST_DIR/test_results/${TEST_CASE}"
@@ -159,8 +159,8 @@ jobs:
         if: always()
         uses: actions/upload-artifact@v4
         with:
-          name: qwen3-inference-results-${{ inputs.test_case }}-${{ github.run_id }}
-          path: tests/functional_tests/inference/qwen3/test_results/${{ inputs.test_case }}
+          name: qwen3-inference-results-${{ inputs.test_case || '4b_tp2_ascend' }}-${{ github.run_id }}
+          path: tests/functional_tests/inference/qwen3/test_results/${{ inputs.test_case || '4b_tp2_ascend' }}
           retention-days: 7
           if-no-files-found: warn
 
@@ -170,5 +170,5 @@ jobs:
           echo "=== Test Summary ==="
           echo "Status: ${{ job.status }}"
           echo "Runner: test-lyz-infer"
-          echo "Test Case: ${{ inputs.test_case }}"
+          echo "Test Case: ${{ inputs.test_case || '4b_tp2_ascend' }}"
           echo "Timestamp: $(date -u '+%Y-%m-%d %H:%M:%S UTC')"

From dc15ff20912fee9a74eccc12beca6e7e86135dc0 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 15:47:47 +0800
Subject: [PATCH 08/19] test

---
 .github/workflows/demo_inference_test.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index e4684672ba..ee68018c9a 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -123,7 +123,7 @@ jobs:
 
           # Run inference using flagscale
           echo "Starting inference..."
-          flagscale run --config-path "$TEST_DIR/conf" --config-name "${TEST_CASE}" || {
+          flagscale run --config-path "$TEST_DIR/conf" --config-name "${TEST_CASE}" --test || {
             echo "❌ Inference failed"
             exit 1
           }

From 9c049bf7ade1e648720c28d13d95940b5ec5f690 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 15:50:26 +0800
Subject: [PATCH 09/19] test

---
 .github/workflows/demo_inference_test.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index ee68018c9a..73519ada78 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -123,7 +123,7 @@ jobs:
 
           # Run inference using flagscale
           echo "Starting inference..."
-          flagscale run --config-path "$TEST_DIR/conf" --config-name "${TEST_CASE}" --test || {
+          flagscale inference qwen3  --config-path "$TEST_DIR/conf" --config-name "${TEST_CASE}" --test || {
             echo "❌ Inference failed"
             exit 1
           }

From c8934345df50355fe06b7c74b7fb991f3aedb12b Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 15:52:01 +0800
Subject: [PATCH 10/19] test

---
 .github/workflows/demo_inference_test.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index 73519ada78..40bd639563 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -123,7 +123,7 @@ jobs:
 
           # Run inference using flagscale
           echo "Starting inference..."
-          flagscale inference qwen3  --config-path "$TEST_DIR/conf" --config-name "${TEST_CASE}" --test || {
+          flagscale inference qwen3  --config "$TEST_DIR/conf/${TEST_CASE}.yaml" --test || {
             echo "❌ Inference failed"
             exit 1
           }

From ac4a4835ea95fb362447d0a3bab42f4304fa1e1d Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 15:56:09 +0800
Subject: [PATCH 11/19] test

---
 .../inference/qwen3/conf/inference/4b_tp2_ascend.yaml         | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/functional_tests/inference/qwen3/conf/inference/4b_tp2_ascend.yaml b/tests/functional_tests/inference/qwen3/conf/inference/4b_tp2_ascend.yaml
index 366e720810..31b6d36fce 100644
--- a/tests/functional_tests/inference/qwen3/conf/inference/4b_tp2_ascend.yaml
+++ b/tests/functional_tests/inference/qwen3/conf/inference/4b_tp2_ascend.yaml
@@ -1,6 +1,6 @@
 llm:
-  model: /home/gitlab-runner/data/Qwen3-4B
-  tokenizer: /home/gitlab-runner/data/Qwen3-4B
+  model: /flagcicd/model/0b0ec7d3-1439-4e5d-9847-f5026942e397/latest
+  tokenizer: /flagcicd/model/0b0ec7d3-1439-4e5d-9847-f5026942e397/latest
   trust_remote_code: true
   tensor_parallel_size: 2
   pipeline_parallel_size: 1

From d1cc929d0b0f544975a868d56e6f7e626bfcfd97 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 16:00:05 +0800
Subject: [PATCH 12/19] test

---
 .github/workflows/demo_inference_test.yml | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/.github/workflows/demo_inference_test.yml b/.github/workflows/demo_inference_test.yml
index 40bd639563..32775c4363 100644
--- a/.github/workflows/demo_inference_test.yml
+++ b/.github/workflows/demo_inference_test.yml
@@ -90,6 +90,12 @@ jobs:
           cd $PROJECT_ROOT
           pip install . --no-build-isolation || { echo "❌ FlagScale install failed"; exit 1; }
 
+          # Install vllm-plugin-FL
+          pip install vllm-plugin-fl==0.1.0+vllm0.13.0 \
+            --extra-index-url https://resource.flagos.net/repository/flagos-pypi-hosted/simple \
+            || { echo "❌ vllm-plugin-FL install failed"; exit 1; }
+          echo "✅ vllm-plugin-FL installed successfully"
+
           # Verify installation
           command -v flagscale || { echo "❌ FlagScale CLI not found in PATH"; exit 1; }
           echo "✅ FlagScale CLI installed: $(flagscale --version 2>/dev/null || echo 'version unknown')"

From a75d26cbf223afc68f0a4cec51f6f89016cd1940 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 16:43:06 +0800
Subject: [PATCH 13/19] test metax

---
 .github/workflows/demo_train_test.yml         | 191 ++++++++++++++++++
 .../train/qwen3/conf/train/data.yaml          |   4 +-
 2 files changed, 193 insertions(+), 2 deletions(-)
 create mode 100644 .github/workflows/demo_train_test.yml

diff --git a/.github/workflows/demo_train_test.yml b/.github/workflows/demo_train_test.yml
new file mode 100644
index 0000000000..9f11977d65
--- /dev/null
+++ b/.github/workflows/demo_train_test.yml
@@ -0,0 +1,191 @@
+name: Demo Train Test - Qwen3
+
+# Demo workflow to test the flagscale-metax-c550 runner with real Qwen3 training
+# Uses FlagScale training tests from tests/functional_tests/train/qwen3
+
+on:
+  pull_request:
+    branches: ["main"]
+  workflow_dispatch:
+    inputs:
+      test_case:
+        description: 'Test case to run'
+        required: false
+        default: '0_6b_metax'
+        type: choice
+        options:
+          - 0_6b_metax
+
+jobs:
+  demo_train:
+    name: Qwen3 Train Demo
+    runs-on: test-lyz-train-metax
+    env:
+      PROJECT_ROOT: ${{ github.workspace }}
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Print test info
+        run: |
+          echo "=========================================="
+          echo "Qwen3 Train Demo Test"
+          echo "=========================================="
+          echo "Runner: ${{ runner.name }}"
+          echo "Test Case: ${{ inputs.test_case || '0_6b_metax' }}"
+          echo "Workflow: ${{ github.workflow }}"
+          echo "Run ID: ${{ github.run_id }}"
+          echo "Project Root: $PROJECT_ROOT"
+          echo "=========================================="
+
+      - name: Check system info
+        run: |
+          echo "=== System Information ==="
+          echo "Hostname: $(hostname)"
+          echo "OS: $(uname -s)"
+          echo "Kernel: $(uname -r)"
+          echo "Architecture: $(uname -m)"
+          echo ""
+          echo "=== CPU Information ==="
+          lscpu | grep -E "Model name|CPU\(s\)|Thread|Core" || echo "lscpu not available"
+          echo ""
+          echo "=== Memory Information ==="
+          free -h || echo "free command not available"
+          echo ""
+          echo "=== Disk Space ==="
+          df -h / || echo "df command not available"
+
+      - name: Check GPU availability
+        run: |
+          echo "=== GPU Information ==="
+          if command -v mx-smi &> /dev/null; then
+            echo "MetaX GPU detected:"
+            mx-smi
+          else
+            echo "mx-smi not found - no MetaX GPU or driver not installed"
+          fi
+
+          if command -v nvidia-smi &> /dev/null; then
+            echo "NVIDIA GPU detected:"
+            nvidia-smi
+          else
+            echo "nvidia-smi not found"
+          fi
+
+      - name: Check Python environment
+        run: |
+          echo "=== Python Environment ==="
+          if command -v python &> /dev/null; then
+            echo "Python version: $(python --version)"
+            echo "Python location: $(which python)"
+          else
+            echo "Python not found in PATH"
+          fi
+
+      - name: Prepare data
+        run: |
+          mkdir -p /opt/data && cd /opt/data
+          wget https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/datasets/enron_emails_demo_text_document_qwen/enron_emails_demo_text_document_qwen.idx
+          wget https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/datasets/enron_emails_demo_text_document_qwen/enron_emails_demo_text_document_qwen.bin
+          mkdir -p /opt/qwentokenizer && cd /opt/qwentokenizer
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/tokenizer_config.json" -O tokenizer_config.json
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/qwen.tiktoken" -O qwen.tiktoken
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/qwen_generation_utils.py" -O qwen_generation_utils.py
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/tokenization_qwen.py" -O tokenization_qwen.py
+
+      - name: Install FlagScale
+        run: |
+          echo "=== Installing FlagScale ==="
+          cd $PROJECT_ROOT
+
+          # Install Megatron-LM-FL
+          pip install megatron_core==0.1.0+megatron0.15.0rc7 \
+            --extra-index-url https://resource.flagos.net/repository/flagos-pypi-hosted/simple \
+            || { echo "❌ Megatron-LM-FL install failed"; exit 1; }
+          echo "✅ Megatron-LM-FL installed successfully"
+
+          # Install TransformerEngine-FL and dependencies
+          git clone --depth 1 https://github.com/flagos-ai/TransformerEngine-FL.git /workspace/TransformerEngine-FL \
+            || { echo "❌ TransformerEngine-FL clone failed"; exit 1; }
+          TE_FL_SKIP_CUDA=1 pip install /workspace/TransformerEngine-FL --no-build-isolation \
+            || { echo "❌ TransformerEngine-FL install failed"; exit 1; }
+          echo "✅ TransformerEngine-FL installed successfully"
+
+          # Install FlagScale
+          pip install . --no-build-isolation || { echo "❌ FlagScale install failed"; exit 1; }
+
+          # Verify installation
+          command -v flagscale || { echo "❌ FlagScale CLI not found in PATH"; exit 1; }
+          echo "✅ FlagScale CLI installed: $(flagscale --version 2>/dev/null || echo 'version unknown')"
+
+      - name: Run Qwen3 train test
+        id: train_test
+        run: |
+          set -euo pipefail
+          cd $PROJECT_ROOT
+
+          TEST_CASE="${{ inputs.test_case || '0_6b_metax' }}"
+          TEST_DIR="tests/functional_tests/train/qwen3"
+          CONFIG_FILE="$TEST_DIR/conf/${TEST_CASE}.yaml"
+          RESULTS_DIR="$TEST_DIR/test_results/${TEST_CASE}"
+
+          echo "=== Running Qwen3 Train Test ==="
+          echo "Test case: $TEST_CASE"
+          echo "Config: $CONFIG_FILE"
+          echo "Results dir: $RESULTS_DIR"
+          echo ""
+
+          # Check if config exists
+          if [ ! -f "$CONFIG_FILE" ]; then
+            echo "❌ Config file not found: $CONFIG_FILE"
+            exit 1
+          fi
+
+          echo "Starting training..."
+          flagscale train qwen3 --config "$CONFIG_FILE" --test || {
+            echo "❌ Training failed"
+            exit 1
+          }
+
+          echo "✅ Training completed"
+
+          # Check results
+          if [ -d "$RESULTS_DIR" ]; then
+            echo ""
+            echo "=== Training Results ==="
+            ls -la "$RESULTS_DIR"
+            echo ""
+
+            # Check for loss log if available
+            if [ -f "$RESULTS_DIR/hydra/logs/flagscale/flagscale.log" ]; then
+              echo "=== Training Log (last 50 lines) ==="
+              tail -50 "$RESULTS_DIR/hydra/logs/flagscale/flagscale.log"
+              echo ""
+            fi
+          else
+            echo "⚠️  Results directory not found at $RESULTS_DIR"
+          fi
+
+          echo ""
+          echo "=========================================="
+          echo "✅ Qwen3 train test completed!"
+          echo "=========================================="
+
+      - name: Upload test results
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: qwen3-train-results-${{ inputs.test_case || '0_6b_metax' }}-${{ github.run_id }}
+          path: tests/functional_tests/train/qwen3/test_results/${{ inputs.test_case || '0_6b_metax' }}
+          retention-days: 7
+          if-no-files-found: warn
+
+      - name: Test summary
+        if: always()
+        run: |
+          echo "=== Test Summary ==="
+          echo "Status: ${{ job.status }}"
+          echo "Runner: ${{ runner.name }}"
+          echo "Test Case: ${{ inputs.test_case || '0_6b_metax' }}"
+          echo "Timestamp: $(date -u '+%Y-%m-%d %H:%M:%S UTC')"
diff --git a/tests/functional_tests/train/qwen3/conf/train/data.yaml b/tests/functional_tests/train/qwen3/conf/train/data.yaml
index 8d9d4bd4ca..bc9db18d8d 100644
--- a/tests/functional_tests/train/qwen3/conf/train/data.yaml
+++ b/tests/functional_tests/train/qwen3/conf/train/data.yaml
@@ -1,9 +1,9 @@
 data:
-  data_path: /home/gitlab-runner/data/pile_wikipedia_demo/pile_wikipedia_demo
+  data_path: /opt/data/pile_wikipedia_demo
   split: 1
   no_mmap_bin_files: true
   tokenizer:
     tokenizer_type: QwenTokenizerFS
-    tokenizer_path: /home/gitlab-runner/tokenizers/qwentokenizer
+    tokenizer_path: /opt/qwentokenizer
     vocab_size: 151936
     make_vocab_size_divisible_by: 64

From 3f0bb5597515d0bf665c5d53f5e3a0502de28a01 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Mon, 18 May 2026 18:02:48 +0800
Subject: [PATCH 14/19] print user

---
 .github/workflows/demo_train_test.yml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/.github/workflows/demo_train_test.yml b/.github/workflows/demo_train_test.yml
index 9f11977d65..7665c191f2 100644
--- a/.github/workflows/demo_train_test.yml
+++ b/.github/workflows/demo_train_test.yml
@@ -37,6 +37,7 @@ jobs:
           echo "Workflow: ${{ github.workflow }}"
           echo "Run ID: ${{ github.run_id }}"
           echo "Project Root: $PROJECT_ROOT"
+          echo "Current User: $(whoami)"
           echo "=========================================="
 
       - name: Check system info

From d08f73e3233028a9ae07e4c402317813896fced3 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Tue, 19 May 2026 10:03:29 +0800
Subject: [PATCH 15/19]  env | sort

---
 .github/workflows/demo_train_test.yml | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/.github/workflows/demo_train_test.yml b/.github/workflows/demo_train_test.yml
index 7665c191f2..74f1ed39af 100644
--- a/.github/workflows/demo_train_test.yml
+++ b/.github/workflows/demo_train_test.yml
@@ -38,6 +38,8 @@ jobs:
           echo "Run ID: ${{ github.run_id }}"
           echo "Project Root: $PROJECT_ROOT"
           echo "Current User: $(whoami)"
+          echo "=== Environment Variables ==="
+          env | sort
           echo "=========================================="
 
       - name: Check system info

From 82a68c8b59c901af122e56cb3db204eae3b033e3 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Tue, 19 May 2026 10:26:44 +0800
Subject: [PATCH 16/19] conda env

---
 .github/workflows/demo_train_test.yml | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/.github/workflows/demo_train_test.yml b/.github/workflows/demo_train_test.yml
index 74f1ed39af..db0ae44fb7 100644
--- a/.github/workflows/demo_train_test.yml
+++ b/.github/workflows/demo_train_test.yml
@@ -99,6 +99,9 @@ jobs:
 
       - name: Install FlagScale
         run: |
+          source /etc/profile.d/conda.sh
+          conda activate base
+
           echo "=== Installing FlagScale ==="
           cd $PROJECT_ROOT
 
@@ -125,6 +128,9 @@ jobs:
       - name: Run Qwen3 train test
         id: train_test
         run: |
+          source /etc/profile.d/conda.sh
+          conda activate base
+          
           set -euo pipefail
           cd $PROJECT_ROOT
 

From bbcf237ac244b689ae4d073bd964dae863ac254b Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Tue, 19 May 2026 10:32:18 +0800
Subject: [PATCH 17/19] conda env

---
 .github/workflows/demo_train_test.yml | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/demo_train_test.yml b/.github/workflows/demo_train_test.yml
index db0ae44fb7..f72da983c7 100644
--- a/.github/workflows/demo_train_test.yml
+++ b/.github/workflows/demo_train_test.yml
@@ -106,9 +106,13 @@ jobs:
           cd $PROJECT_ROOT
 
           # Install Megatron-LM-FL
-          pip install megatron_core==0.1.0+megatron0.15.0rc7 \
-            --extra-index-url https://resource.flagos.net/repository/flagos-pypi-hosted/simple \
-            || { echo "❌ Megatron-LM-FL install failed"; exit 1; }
+          git clone \
+            "https://github.com/flagos-ai/Megatron-LM-FL.git" /tmp/Megatron-LM-FL
+          git -C /tmp/Megatron-LM-FL checkout d092f8df49f7c0b5b4cae42d036b7e4a26b8fc81
+
+          echo "Installing Megatron-LM-FL via pip..."
+          pip install /tmp/Megatron-LM-FL --no-build-isolation --root-user-action=ignore \
+            || { echo "Megatron-LM-FL install failed"; exit 1; }
           echo "✅ Megatron-LM-FL installed successfully"
 
           # Install TransformerEngine-FL and dependencies
@@ -130,7 +134,7 @@ jobs:
         run: |
           source /etc/profile.d/conda.sh
           conda activate base
-          
+
           set -euo pipefail
           cd $PROJECT_ROOT
 

From 2f0c7b78e8ee5c647c63f2645912664a902d40ea Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Tue, 19 May 2026 11:15:51 +0800
Subject: [PATCH 18/19] test musa

---
 .github/workflows/test_demo_musa.yml | 188 +++++++++++++++++++++++++++
 1 file changed, 188 insertions(+)
 create mode 100644 .github/workflows/test_demo_musa.yml

diff --git a/.github/workflows/test_demo_musa.yml b/.github/workflows/test_demo_musa.yml
new file mode 100644
index 0000000000..ccaf91691a
--- /dev/null
+++ b/.github/workflows/test_demo_musa.yml
@@ -0,0 +1,188 @@
+name: Demo Train Test - Qwen3
+
+# Demo workflow to test the flagscale-metax-c550 runner with real Qwen3 training
+# Uses FlagScale training tests from tests/functional_tests/train/qwen3
+
+on:
+  pull_request:
+    branches: ["main"]
+  workflow_dispatch:
+    inputs:
+      test_case:
+        description: 'Test case to run'
+        required: false
+        default: '0_6b_metax'
+        type: choice
+        options:
+          - 0_6b_metax
+
+jobs:
+  demo_train:
+    name: Qwen3 Train Demo
+    runs-on: test-lyz-musa
+    env:
+      PROJECT_ROOT: ${{ github.workspace }}
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Print test info
+        run: |
+          echo "=========================================="
+          echo "Qwen3 Train Demo Test"
+          echo "=========================================="
+          echo "Runner: ${{ runner.name }}"
+          echo "Test Case: ${{ inputs.test_case || '0_6b_metax' }}"
+          echo "Workflow: ${{ github.workflow }}"
+          echo "Run ID: ${{ github.run_id }}"
+          echo "Project Root: $PROJECT_ROOT"
+          echo "Current User: $(whoami)"
+          echo "=== Environment Variables ==="
+          env | sort
+          echo "=========================================="
+
+      - name: Check system info
+        run: |
+          echo "=== System Information ==="
+          echo "Hostname: $(hostname)"
+          echo "OS: $(uname -s)"
+          echo "Kernel: $(uname -r)"
+          echo "Architecture: $(uname -m)"
+          echo ""
+          echo "=== CPU Information ==="
+          lscpu | grep -E "Model name|CPU\(s\)|Thread|Core" || echo "lscpu not available"
+          echo ""
+          echo "=== Memory Information ==="
+          free -h || echo "free command not available"
+          echo ""
+          echo "=== Disk Space ==="
+          df -h / || echo "df command not available"
+
+      - name: Check GPU availability
+        run: |
+          echo "=== GPU Information ==="
+          mthreads-gmi 
+
+      - name: Check Python environment
+        run: |
+          echo "=== Python Environment ==="
+          if command -v python &> /dev/null; then
+            echo "Python version: $(python --version)"
+            echo "Python location: $(which python)"
+          else
+            echo "Python not found in PATH"
+          fi
+
+      - name: Prepare data
+        run: |
+          mkdir -p /opt/data && cd /opt/data
+          wget https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/datasets/enron_emails_demo_text_document_qwen/enron_emails_demo_text_document_qwen.idx
+          wget https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/datasets/enron_emails_demo_text_document_qwen/enron_emails_demo_text_document_qwen.bin
+          mkdir -p /opt/qwentokenizer && cd /opt/qwentokenizer
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/tokenizer_config.json" -O tokenizer_config.json
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/qwen.tiktoken" -O qwen.tiktoken
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/qwen_generation_utils.py" -O qwen_generation_utils.py
+          wget "https://baai-flagscale.ks3-cn-beijing.ksyuncs.com/tokenizers/qwentokenizer/tokenization_qwen.py" -O tokenization_qwen.py
+
+      - name: Install FlagScale
+        run: |
+
+          echo "=== Installing FlagScale ==="
+          cd $PROJECT_ROOT
+
+          # Install Megatron-LM-FL
+          git clone \
+            "https://github.com/flagos-ai/Megatron-LM-FL.git" /tmp/Megatron-LM-FL
+          git -C /tmp/Megatron-LM-FL checkout d092f8df49f7c0b5b4cae42d036b7e4a26b8fc81
+
+          echo "Installing Megatron-LM-FL via pip..."
+          pip install /tmp/Megatron-LM-FL --no-build-isolation --root-user-action=ignore \
+            || { echo "Megatron-LM-FL install failed"; exit 1; }
+          echo "✅ Megatron-LM-FL installed successfully"
+
+          # Install TransformerEngine-FL and dependencies
+          git clone --depth 1 https://github.com/flagos-ai/TransformerEngine-FL.git /workspace/TransformerEngine-FL \
+            || { echo "❌ TransformerEngine-FL clone failed"; exit 1; }
+          TE_FL_SKIP_CUDA=1 pip install /workspace/TransformerEngine-FL --no-build-isolation \
+            || { echo "❌ TransformerEngine-FL install failed"; exit 1; }
+          echo "✅ TransformerEngine-FL installed successfully"
+
+          # Install FlagScale
+          pip install . --no-build-isolation || { echo "❌ FlagScale install failed"; exit 1; }
+
+          # Verify installation
+          command -v flagscale || { echo "❌ FlagScale CLI not found in PATH"; exit 1; }
+          echo "✅ FlagScale CLI installed: $(flagscale --version 2>/dev/null || echo 'version unknown')"
+
+      - name: Run Qwen3 train test
+        id: train_test
+        run: |
+
+          set -euo pipefail
+          cd $PROJECT_ROOT
+
+          TEST_CASE="${{ inputs.test_case || '0_6b_metax' }}"
+          TEST_DIR="tests/functional_tests/train/qwen3"
+          CONFIG_FILE="$TEST_DIR/conf/${TEST_CASE}.yaml"
+          RESULTS_DIR="$TEST_DIR/test_results/${TEST_CASE}"
+
+          echo "=== Running Qwen3 Train Test ==="
+          echo "Test case: $TEST_CASE"
+          echo "Config: $CONFIG_FILE"
+          echo "Results dir: $RESULTS_DIR"
+          echo ""
+
+          # Check if config exists
+          if [ ! -f "$CONFIG_FILE" ]; then
+            echo "❌ Config file not found: $CONFIG_FILE"
+            exit 1
+          fi
+
+          echo "Starting training..."
+          flagscale train qwen3 --config "$CONFIG_FILE" --test || {
+            echo "❌ Training failed"
+            exit 1
+          }
+
+          echo "✅ Training completed"
+
+          # Check results
+          if [ -d "$RESULTS_DIR" ]; then
+            echo ""
+            echo "=== Training Results ==="
+            ls -la "$RESULTS_DIR"
+            echo ""
+
+            # Check for loss log if available
+            if [ -f "$RESULTS_DIR/hydra/logs/flagscale/flagscale.log" ]; then
+              echo "=== Training Log (last 50 lines) ==="
+              tail -50 "$RESULTS_DIR/hydra/logs/flagscale/flagscale.log"
+              echo ""
+            fi
+          else
+            echo "⚠️  Results directory not found at $RESULTS_DIR"
+          fi
+
+          echo ""
+          echo "=========================================="
+          echo "✅ Qwen3 train test completed!"
+          echo "=========================================="
+
+      - name: Upload test results
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: qwen3-train-results-${{ inputs.test_case || '0_6b_metax' }}-${{ github.run_id }}
+          path: tests/functional_tests/train/qwen3/test_results/${{ inputs.test_case || '0_6b_metax' }}
+          retention-days: 7
+          if-no-files-found: warn
+
+      - name: Test summary
+        if: always()
+        run: |
+          echo "=== Test Summary ==="
+          echo "Status: ${{ job.status }}"
+          echo "Runner: ${{ runner.name }}"
+          echo "Test Case: ${{ inputs.test_case || '0_6b_metax' }}"
+          echo "Timestamp: $(date -u '+%Y-%m-%d %H:%M:%S UTC')"

From ac4762b68a767768382077e232087a9e0ddc7c83 Mon Sep 17 00:00:00 2001
From: liyuzhuo <lee.yuzhuo233@gmail.com>
Date: Tue, 19 May 2026 14:29:27 +0800
Subject: [PATCH 19/19] test musa

---
 .github/workflows/test_demo_musa.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/test_demo_musa.yml b/.github/workflows/test_demo_musa.yml
index ccaf91691a..ec4830aea5 100644
--- a/.github/workflows/test_demo_musa.yml
+++ b/.github/workflows/test_demo_musa.yml
@@ -1,4 +1,4 @@
-name: Demo Train Test - Qwen3
+name: Demo Train Test Musa - Qwen3
 
 # Demo workflow to test the flagscale-metax-c550 runner with real Qwen3 training
 # Uses FlagScale training tests from tests/functional_tests/train/qwen3