Benchmarking v2 GH workflows

.github/workflows/benchmark_v2.yml

@@ -0,0 +1,82 @@
+name: Benchmark v2 Framework
+
+on:
+  workflow_call:
+    inputs:
+      runner:
+        description: 'GH Actions runner group to use'
+        required: true
+        type: string
+      commit_sha:
+        description: 'Commit SHA to benchmark'
+        required: false
+        type: string
+        default: ''
+      upload_to_hub:
+        description: 'Enable/disable uploading results to a HuggingFace Dataset'
+        required: false
+        type: string
+        default: 'false'
+      run_id:
+        description: 'Custom run ID for organizing results (auto-generated if not provided)'
+        required: false
+        type: string
+        default: ''
+      benchmark_repo_id:
+        description: 'HuggingFace Dataset to upload results to (e.g., "org/benchmark-results")'
+        required: false
+        type: string
+        default: ''
+
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access.
+  # This token is created under the bot `hf-transformers-bot`.
+  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
+
+jobs:
+  benchmark-v2:
+    name: Benchmark v2
+    runs-on: ${{ inputs.runner }}
+    if: |
+      (github.event_name == 'pull_request' && contains( github.event.pull_request.labels.*.name, 'run-benchmark')) ||
+      (github.event_name == 'schedule')
+    container:
+      image: huggingface/transformers-pytorch-gpu
+      options: --gpus all --privileged --ipc host --shm-size "16gb"
+    steps:
+      - name: Get repo
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.commit_sha || github.sha }}
+
+      - name: Install benchmark dependencies
+        run: |
+          python3 -m pip install -r benchmark_v2/requirements.txt
+
+      - name: Reinstall transformers in edit mode
+        run: |
+          python3 -m pip uninstall -y transformers
+          python3 -m pip install -e ".[torch]"
+
+      - name: Show installed libraries and their versions
+        run: |
+          python3 -m pip list
+          python3 -c "import torch; print(f'PyTorch version: {torch.__version__}')"
+          python3 -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')"
+          python3 -c "import torch; print(f'CUDA device count: {torch.cuda.device_count()}')" || true
+          nvidia-smi || true
+
+      - name: Run benchmark v2
+        working-directory: benchmark_v2
+        run: |
+          echo "Running benchmarks"
+          python3 run_benchmarks.py \
+          --commit-id '${{ inputs.commit_sha || github.sha }}' \
+          --upload-to-hub '${{ inputs.upload_to_hub || false}}' \
+          --run-id '${{ inputs.run_id }}' \
+          --benchmark-repo-id '${{ inputs.benchmark_repo_id}}' \
+          --log-level INFO
+        env:
+          HF_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}

.github/workflows/benchmark_v2_a10_caller.yml

@@ -0,0 +1,20 @@
+name: Benchmark v2 Scheduled Runner - A10 Single-GPU
+
+on:
+  schedule:
+    # Run daily at 16:30 UTC
+    - cron: "30 16 * * *"
+  pull_request:
+    types: [ opened, labeled, reopened, synchronize ]
+
+jobs:
+  benchmark-v2-default:
+    name: Benchmark v2 - Default Models
+    uses: ./.github/workflows/benchmark_v2.yml
+    with:
+      runner: aws-g5-4xlarge-cache-use1-public-80
+      commit_sha: ${{ github.sha }}
+      upload_to_hub: true
+      run_id: ${{ github.run_id }}
+      benchmark_repo_id: hf-internal-testing/transformers-daily-benchmarks
+    secrets: inherit

.github/workflows/benchmark_v2_mi325_caller.yml

@@ -0,0 +1,20 @@
+name: Benchmark v2 Scheduled Runner - MI325 Single-GPU
+
+on:
+  schedule:
+    # Run daily at 16:30 UTC
+    - cron: "30 16 * * *"
+  pull_request:
+    types: [ opened, labeled, reopened, synchronize ]
+
+jobs:
+  benchmark-v2-default:
+    name: Benchmark v2 - Default Models
+    uses: ./.github/workflows/benchmark_v2.yml
+    with:
+      runner: amd-mi325-ci-1gpu
+      commit_sha: ${{ github.sha }}
+      upload_to_hub: true
+      run_id: ${{ github.run_id }}
+      benchmark_repo_id: hf-internal-testing/transformers-daily-benchmarks
+    secrets: inherit

benchmark_v2/README.md

@@ -21,6 +21,36 @@ python run_benchmarks.py \
     --num-tokens-to-generate 200
 ```
 
+### Uploading Results to HuggingFace Dataset
+
+You can automatically upload benchmark results to a HuggingFace Dataset for tracking and analysis:
+
+```bash
+# Upload to a public dataset with auto-generated run ID
+python run_benchmarks.py --upload-to-hf username/benchmark-results
+
+# Upload with a custom run ID for easy identification
+python run_benchmarks.py --upload-to-hf username/benchmark-results --run-id experiment_v1
+```
+
+**Dataset Directory Structure:**
+```
+dataset_name/
+├── 2025-01-15/
+│   ├── runs/                       # Non-scheduled runs (manual, PR, etc.)
+│   │   └── 123-1245151651/         # GitHub run number and ID
+│   │       └── benchmark_results/
+│   │           ├── benchmark_summary_20250115_143022.json
+│   │           └── model-name/
+│   │               └── model-name_benchmark_20250115_143022.json
+│   └── benchmark_results_abc123de/ # Scheduled runs (daily CI)
+│       ├── benchmark_summary_20250115_143022.json
+│       └── model-name/
+│           └── model-name_benchmark_20250115_143022.json
+└── 2025-01-16/
+    └── ...
+```
+
 ### Running Specific Benchmarks
 
 ```bash

benchmark_v2/benches/llama.py

@@ -20,7 +20,6 @@
 from benchmark_framework import ModelBenchmark
 
 
-os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
 os.environ["TOKENIZERS_PARALLELISM"] = "1"
 torch.set_float32_matmul_precision("high")
 

benchmark_v2/requirements.txt

@@ -3,4 +3,5 @@ psutil>=5.8.0
 gpustat>=1.0.0
 torch>=2.0.0
 transformers>=4.30.0
-datasets>=2.10.0 
+datasets>=2.10.0
+huggingface_hub>=0.16.0