hugging-face-model-trainer

You can fine-tune LLMs on Hugging Face cloud GPUs by using this Skill to set up training jobs on Hugging Face Jobs infrastructure. It eliminates local GPU setup by leveraging TRL library support for SFT, DPO, GRPO, and reward modeling.

Yes, TRL is fully supported for direct preference optimization on Hugging Face Jobs. The Skill facilitates DPO, group relative policy optimization (GRPO), supervised fine-tuning (SFT), and reward modeling using the TRL library.

Yes, it includes automated GGUF conversion for local deployment after training. This allows you to easily run your fine-tuned models locally using tools like Ollama or LM Studio without complex manual conversion steps.

The best way to monitor LLM training progress is through integrated Trackio support. This Skill uses Trackio to provide real-time monitoring of your training jobs running on Hugging Face cloud infrastructure.

Yes, accelerate and peft are required dependencies for training models with this Skill. The environment requires trl, transformers, accelerate, and peft to properly facilitate LLM training and fine-tuning on Hugging Face Jobs.

You can save your fine-tuned model to the Hugging Face Hub automatically through the training job setup. The Skill streamlines this process by managing the infrastructure, monitoring via Trackio, and automatically saving the resulting model to the Hub.