add llama 3 support to llm.c #688
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: GPU Builds and Tests | |
| on: | |
| create: | |
| workflow_dispatch: | |
| push: | |
| branches: | |
| - master | |
| pull_request: | |
| branches: | |
| - master | |
| jobs: | |
| build-and-test-gpu: | |
| runs-on: ubicloud-gpu-standard-1-latest | |
| steps: | |
| - name: Checkout code | |
| uses: actions/checkout@v4 | |
| - name: Install OpenMP | |
| run: sudo apt-get update && sudo apt-get install -y libomp-dev | |
| - name: Install dependencies | |
| run: pip install -r requirements.txt | |
| - name: Run preprocessing | |
| run: python dev/data/tinyshakespeare.py | |
| - name: Train model | |
| run: python train_gpt2.py | |
| - name: Compile training and testing program | |
| run: make test_gpt2cu train_gpt2cu test_gpt2fp32cu train_gpt2fp32cu | |
| - name: Train model (With OpenMP) | |
| run: OMP_NUM_THREADS=8 ./train_gpt2cu | |
| - name: Train model (FP32) with gpt2_124M.bin | |
| run: | | |
| PRECISION=FP32 make train_gpt2cu | |
| ./train_gpt2cu -b 1 -t 64 -d 256 -l 0.0001 -v 200 -s 200 -a 1 -x 10 -r 0 -f 0 -e "gpt2_124M.bin" | |
| - name: Test for percent loss differential for FP32 | |
| run: | | |
| PRECISION=FP32 make train_gpt2cu | |
| ./train_gpt2cu -b 1 -t 64 -d 256 -l 0.0001 -v 200 -s 200 -a 1 -x 10 -r 0 -f 0 -e "gpt2_124M.bin" > train_gpt2cu_fp32_precision.txt | |
| python dev/loss_checker_ci.py -f train_gpt2cu_fp32_precision.txt -s 20 -e 28 -a 5.0 | |
| - name: Build FP32 precision | |
| run: PRECISION=FP32 make test_gpt2cu profile_gpt2cu | |
| - name: Run default | |
| run: ./test_gpt2cu | |
| - name: Run no recompute GeLU | |
| run: ./test_gpt2cu -r 0 | |
| - name: Run recompute LN | |
| run: ./test_gpt2cu -r 2 | |
| - name: Build BF16 precision | |
| run: PRECISION=BF16 make train_gpt2cu test_gpt2cu profile_gpt2cu | |
| - name: Run default | |
| run: ./test_gpt2cu | |
| - name: Run no recompute GeLU | |
| run: ./test_gpt2cu -r 0 | |
| - name: Run no master weights | |
| run: ./test_gpt2cu -w 0 | |
| - name: Run recompute LN | |
| run: ./test_gpt2cu -r 2 | |
| - name: Train model fp32 (With OpenMP) | |
| run: OMP_NUM_THREADS=8 ./train_gpt2fp32cu | |
| - name: Execute testing program (With OpenMP) | |
| run: OMP_NUM_THREADS=8 ./test_gpt2cu | |
| - name: Execute testing program fp32 (With OpenMP) | |
| run: OMP_NUM_THREADS=8 ./test_gpt2fp32cu | |
| - name: Compile training and testing program without OpenMP | |
| run: NO_OMP=1 make test_gpt2cu train_gpt2cu test_gpt2fp32cu train_gpt2fp32cu | |
| - name: Train model (No OpenMP) | |
| run: NO_OMP=1 ./train_gpt2cu | |
| - name: Train model fp32 (No OpenMP) | |
| run: NO_OMP=1 ./train_gpt2fp32cu | |
| - name: Execute testing program (No OpenMP) | |
| run: ./test_gpt2cu -b 32 | |
| - name: Execute testing program fp32 (No OpenMP) | |
| run: ./test_gpt2fp32cu | |
| - name: Install cuDNN-frontend | |
| run: | |
| git clone https://github.com/NVIDIA/cudnn-frontend.git | |
| - name: Build with cuDNN | |
| run: USE_CUDNN=1 make test_gpt2cu train_gpt2cu test_gpt2fp32cu train_gpt2fp32cu | |
| - name: Train model with cuDNN | |
| run: ./train_gpt2cu | |
| - name: Train model fp32 with cuDNN | |
| run: ./train_gpt2fp32cu | |
| - name: Execute testing program with cuDNN | |
| run: ./test_gpt2cu | |
| - name: Execute testing program fp32 with cuDNN | |
| run: ./test_gpt2fp32cu | |
| unit-tests-gpu: | |
| runs-on: ubicloud-gpu-standard-1-latest | |
| steps: | |
| - name: Checkout code | |
| uses: actions/checkout@v4 | |
| - name: Test Device<->File IO | |
| run: cd dev/test && nvcc -o device_file_io device_file_io.cu && ./device_file_io |