cherry pick

Author: fzyzcjy

.github/workflows/pr-test.yml

@@ -0,0 +1,57 @@
+name: PR Test
+
+on:
+  # Do not run CI on push to reduce CI time
+  # push:
+  #   branches: [main]
+  pull_request:
+    branches: [main]
+    types: [synchronize, labeled]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  e2e-test:
+    # TODO may use run-ci label etc
+    if: github.event.pull_request.draft == false
+    runs-on: self-hosted
+    container:
+      image: slimerl/slime:latest
+      options: >
+        --gpus all
+        --ipc=host
+        --shm-size=16g
+        --ulimit memlock=-1
+        --ulimit stack=67108864
+        --memory=0
+        --memory-swap=0
+        -v /data/slime_ci:/data/slime_ci
+        -v /data/slime_ci/models:/root/models
+        -v /data/slime_ci/datasets:/root/datasets
+    strategy:
+      fail-fast: false
+      matrix:
+        info:
+          - {test_file: test_quick_start_glm4_9B.py}
+          - {test_file: test_qwen3_30B_A3B.py}
+          # TODO use deterministic kernel
+    defaults:
+      run:
+        working-directory: ${{ github.workspace }}
+    env:
+      GITHUB_COMMIT_NAME: ${{ github.sha }}_${{ github.event.pull_request.number || 'non-pr' }}
+      WANDB_API_KEY: ${{ secrets.WANDB_API_KEY }}
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install
+        shell: bash
+        run: cd $GITHUB_WORKSPACE && pip install -e .
+
+      - name: Execute
+        shell: bash
+        run: python tests/${{ matrix.info.test_file }}

.github/workflows/slime-ci.yml

@@ -0,0 +1,47 @@
+# Doc about CI
+
+## Configure GitHub secrets
+
+https://github.com/slimerl/slime/settings/secrets/actions
+
+* `WANDB_API_KEY`: get from https://wandb.ai/authorize
+
+## Setup new GitHub runners
+
+### Step 1: Env
+
+Write `.env` mimicking `.env.example`.
+The token can be found at https://github.com/slimerl/slime/settings/actions/runners/new?arch=x64&os=linux.
+
+WARN: The `GITHUB_RUNNER_TOKEN` changes after a while.
+
+### Step 2: Prepare `/home/runner/externals`
+
+```shell
+docker run --rm -it --privileged --pid=host -v /:/host_root ubuntu /bin/bash -c 'rm -rf /host_root/home/runner/externals && mkdir -p /host_root/home/runner/externals && chmod -R 777 /host_root/home/runner/externals'
+docker run -d --name temp-runner ghcr.io/actions/actions-runner:2.328.0 tail -f /dev/null
+docker cp temp-runner:/home/runner/externals/. /home/runner/externals
+docker rm -f temp-runner
+ls -alh /home/runner/externals
+```
+
+### Step 3: Run
+
+```shell
+cd /data/tom/primary_synced/slime/tests/ci/github_runner
+docker compose up -d
+```
+
+### Debugging
+
+Logs
+
+```shell
+docker compose logs -f
+```
+
+Exec
+
+```shell
+docker exec -it github_runner-runner-1 /bin/bash
+```

tests/ci/README.md

@@ -0,0 +1,2 @@
+GITHUB_RUNNER_URL=https://github.com/slimerl/slime
+GITHUB_RUNNER_TOKEN=paste-your-token-here

tests/ci/github_runner/.env.example

@@ -0,0 +1 @@
+.env

tests/ci/github_runner/.gitignore

@@ -0,0 +1,29 @@
+# Please refer to `README.md` for how to setup this GitHub action runner
+version: "3.9"
+
+services:
+  runner:
+    image: ghcr.io/actions/actions-runner:2.328.0
+    volumes:
+      - /var/run/docker.sock:/var/run/docker.sock
+      - /data/slime_ci:/data/slime_ci
+      # it requires this folder
+      - /home/runner/externals:/home/runner/externals
+    deploy:
+      # TODO 4 runner + lock gpu when running
+      replicas: 1
+    restart: always
+    environment:
+      RUNNER_ALLOW_RUNASROOT: "1"
+    privileged: true
+    user: root
+    # ref: https://github.com/actions/runner/issues/367#issuecomment-2007558723
+    # ref: https://github.com/actions/runner
+    # args ref: https://github.com/actions/runner/blob/68ff57dbc4c836d50f46602a8a53301fb9513eb4/src/Runner.Listener/CommandSettings.cs#L53
+    # TODO seems we should not run config.sh repeatedly
+    entrypoint: >
+      sh -c "
+      cd /data/slime_ci &&
+      /home/runner/config.sh --url ${GITHUB_RUNNER_URL} --token ${GITHUB_RUNNER_TOKEN} --unattended --work /data/slime_ci/runner_$(hostname) &&
+      /home/runner/run.sh
+      "