[Fix CI] CI cases in run-ci-image

.github/workflows/pr-test.yml

@@ -33,14 +33,17 @@ jobs:
       options: >
         --gpus all
         --ipc=host
-        --shm-size=16g
+        --shm-size=32g
         --ulimit memlock=-1
         --ulimit stack=67108864
         --memory=0
         --memory-swap=0
         -v /mnt/nvme0n1/miles_ci:/data/miles_ci
         -v /mnt/nvme0n1/miles_ci/models:/root/models
         -v /mnt/nvme0n1/miles_ci/datasets:/root/datasets
+        --privileged
+        --ulimit nofile=65535:65535
+        -v /tmp:/tmp
     strategy:
       fail-fast: false
       matrix:
@@ -52,11 +55,26 @@ jobs:
       GITHUB_COMMIT_NAME: ${{ github.sha }}_${{ github.event.pull_request.number || 'non-pr' }}
       WANDB_API_KEY: ${{ secrets.WANDB_API_KEY }}
       MILES_TEST_ENABLE_INFINITE_RUN: ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.infinite_run) || 'false' }}
+      MILES_TEST_USE_DEEPEP: ${{ matrix.info.use_deepep || '0' }}
+      MILES_TEST_USE_FP8_ROLLOUT: ${{ matrix.info.use_fp8_rollout || '0' }}
+      MILES_TEST_ENABLE_EVAL: ${{ matrix.info.enable_eval || '1' }}
 
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
 
+      - name: Cleanup Ray processes
+        shell: bash
+        run: |
+          pkill -9 -f 'ray::' 2>/dev/null || true
+          pkill -9 -f raylet 2>/dev/null || true
+          pkill -9 -f gcs_server 2>/dev/null || true
+          pkill -9 -f 'ray-dashboard' 2>/dev/null || true
+          pkill -9 sglang 2>/dev/null || true
+          ray stop --force 2>/dev/null || true
+          rm -rf /tmp/ray/* 2>/dev/null || true
+          sleep 3
+
       - name: Install
         shell: bash
         run: cd $GITHUB_WORKSPACE && pip install -e . --no-deps --break-system-packages
@@ -123,15 +141,6 @@ jobs:
         shell: bash
         run: python tests/ci/gpu_lock_exec.py --count ${{ matrix.info.num_gpus }} -- python tests/${{ matrix.info.test_file }}
 
-      - name: Post-test cleanup
-        if: always()
-        shell: bash
-        run: |
-          pkill -9 -f 'ray::' 2>/dev/null || true
-          pkill -9 -f raylet 2>/dev/null || true
-          ray stop --force 2>/dev/null || true
-          rm -rf /tmp/ray/* 2>/dev/null || true
-
   e2e-test-fsdp:
     if: (github.event_name == 'workflow_dispatch') || (github.event.pull_request && contains(github.event.pull_request.labels.*.name, 'run-ci-fsdp'))
     runs-on: self-hosted
@@ -190,15 +199,6 @@ jobs:
         shell: bash
         run: python tests/ci/gpu_lock_exec.py --count ${{ matrix.info.num_gpus }} -- python tests/${{ matrix.info.test_file }}
 
-      - name: Post-test cleanup
-        if: always()
-        shell: bash
-        run: |
-          pkill -9 -f 'ray::' 2>/dev/null || true
-          pkill -9 -f raylet 2>/dev/null || true
-          ray stop --force 2>/dev/null || true
-          rm -rf /tmp/ray/* 2>/dev/null || true
-
   e2e-test-megatron:
     if: (github.event_name == 'workflow_dispatch') || (github.event.pull_request && contains(github.event.pull_request.labels.*.name, 'run-ci-megatron'))
     runs-on: self-hosted
@@ -257,15 +257,6 @@ jobs:
         shell: bash
         run: python tests/ci/gpu_lock_exec.py --count ${{ matrix.info.num_gpus }} -- python tests/${{ matrix.info.test_file }}
 
-      - name: Post-test cleanup
-        if: always()
-        shell: bash
-        run: |
-          pkill -9 -f 'ray::' 2>/dev/null || true
-          pkill -9 -f raylet 2>/dev/null || true
-          ray stop --force 2>/dev/null || true
-          rm -rf /tmp/ray/* 2>/dev/null || true
-
   e2e-test-precision:
     if: (github.event_name == 'workflow_dispatch') || (github.event.pull_request && contains(github.event.pull_request.labels.*.name, 'run-ci-precision'))
     runs-on: self-hosted
@@ -324,15 +315,6 @@ jobs:
         shell: bash
         run: python tests/ci/gpu_lock_exec.py --count ${{ matrix.info.num_gpus }} -- python tests/${{ matrix.info.test_file }}
 
-      - name: Post-test cleanup
-        if: always()
-        shell: bash
-        run: |
-          pkill -9 -f 'ray::' 2>/dev/null || true
-          pkill -9 -f raylet 2>/dev/null || true
-          ray stop --force 2>/dev/null || true
-          rm -rf /tmp/ray/* 2>/dev/null || true
-
   e2e-test-ckpt:
     if: (github.event_name == 'workflow_dispatch') || (github.event.pull_request && contains(github.event.pull_request.labels.*.name, 'run-ci-ckpt'))
     runs-on: self-hosted
@@ -391,15 +373,6 @@ jobs:
         shell: bash
         run: python tests/ci/gpu_lock_exec.py --count ${{ matrix.info.num_gpus }} -- python tests/${{ matrix.info.test_file }}
 
-      - name: Post-test cleanup
-        if: always()
-        shell: bash
-        run: |
-          pkill -9 -f 'ray::' 2>/dev/null || true
-          pkill -9 -f raylet 2>/dev/null || true
-          ray stop --force 2>/dev/null || true
-          rm -rf /tmp/ray/* 2>/dev/null || true
-
   e2e-test-long:
     if: (github.event_name == 'workflow_dispatch') || (github.event.pull_request && contains(github.event.pull_request.labels.*.name, 'run-ci-long'))
     runs-on: self-hosted
@@ -458,20 +431,11 @@ jobs:
         shell: bash
         run: python tests/ci/gpu_lock_exec.py --count ${{ matrix.info.num_gpus }} -- python tests/${{ matrix.info.test_file }}
 
-      - name: Post-test cleanup
-        if: always()
-        shell: bash
-        run: |
-          pkill -9 -f 'ray::' 2>/dev/null || true
-          pkill -9 -f raylet 2>/dev/null || true
-          ray stop --force 2>/dev/null || true
-          rm -rf /tmp/ray/* 2>/dev/null || true
-
   e2e-test-image:
     if: (github.event_name == 'workflow_dispatch') || (github.event.pull_request && contains(github.event.pull_request.labels.*.name, 'run-ci-image'))
     runs-on: self-hosted
     container:
-      image: radixark/miles-test:latest
+      image: radixark/miles:latest
       options: >
         --gpus all
         --ipc=host
@@ -524,12 +488,3 @@ jobs:
       - name: Execute
         shell: bash
         run: python tests/ci/gpu_lock_exec.py --count ${{ matrix.info.num_gpus }} -- python tests/${{ matrix.info.test_file }}
-
-      - name: Post-test cleanup
-        if: always()
-        shell: bash
-        run: |
-          pkill -9 -f 'ray::' 2>/dev/null || true
-          pkill -9 -f raylet 2>/dev/null || true
-          ray stop --force 2>/dev/null || true
-          rm -rf /tmp/ray/* 2>/dev/null || true

.github/workflows/pr-test.yml.j2

@@ -58,7 +58,7 @@
     },
     'e2e-test-image': {
       'label': 'run-ci-image',
-      'image': 'radixark/miles-test:latest',
+      'image': 'radixark/miles:latest',
       'tests': [
         {'test_file': 'test_qwen2.5_0.5B_gsm8k_async_short.py', 'num_gpus': 4},
         {'test_file': 'test_qwen2.5_0.5B_gsm8k_short.py', 'num_gpus': 4},