Skip to content

NVIDIA PyTorch/CUDA Job #25

NVIDIA PyTorch/CUDA Job

NVIDIA PyTorch/CUDA Job #25

name: NVIDIA PyTorch Job
on:
workflow_dispatch:
inputs:
script_content:
description: 'Content of Python script'
required: true
type: string
filename:
description: 'Name of Python script'
required: true
type: string
jobs:
train:
runs-on: [gpumode-nvidia-arc]
timeout-minutes: 10
container:
image: nvidia/cuda:12.4.0-devel-ubuntu22.04
steps:
- name: Setup Python
uses: actions/setup-python@v4
with:
python-version: '3.10'
- name: Install uv
uses: astral-sh/setup-uv@v3
with:
version: "latest"
- name: Create script
shell: python
run: |
with open('${{ github.event.inputs.filename }}', 'w') as f:
f.write('''${{ github.event.inputs.script_content }}''')
- name: Install dependencies
run: |
# Check if 'import torch' is in any Python file
if grep -rE "(import torch|from torch)" "${{ github.event.inputs.filename }}"; then
echo "PyTorch detected, installing torch"
uv pip install numpy torch
fi
# Check if 'import triton' is in any Python file
if grep -rE "(import triton|from triton)" "${{ github.event.inputs.filename }}"; then
echo "Triton detected, installing triton"
uv pip install triton
fi
- name: Run script with profiler
run: |
# Run the script with NSight Compute profiler and save to CSV
# ncu --csv python "${{ github.event.inputs.filename }}" > profile_results.csv 2>&1
# Also keep regular output in training.log
python "${{ github.event.inputs.filename }}" > training.log 2>&1
- name: Upload training artifacts
uses: actions/upload-artifact@v4
if: always()
with:
name: training-artifacts
path: |
training.log
# profile_results.csv
${{ github.event.inputs.filename }}
env:
CUDA_VISIBLE_DEVICES: 0 # Make sure only one GPU is used for testing