marcosremar2/dumont-job-test

Dumont Cloud Job Test

Quick fine-tuning test for Dumont Cloud Jobs feature.

What this does

1. Loads Qwen2.5-0.5B-Instruct (small, fast model)
2. Fine-tunes for 3 steps (just to test the pipeline)
3. Saves the model to /workspace/output
4. Creates completion marker at /workspace/.job_complete

Expected runtime

~2-5 minutes on any GPU

Usage with Dumont Cloud Jobs

# Via CLI
dumont job create --hf-repo marcosremar2/dumont-job-test --gpu "RTX 4090"

# Via API
POST /api/v1/jobs
{
    "name": "Quick Fine-tune Test",
    "source": "huggingface",
    "hf_repo": "marcosremar2/dumont-job-test",
    "gpu_type": "RTX 4090"
}

Files

• train.py - Main training script
• requirements.txt - Dependencies (torch, transformers, datasets, accelerate)