model-trainer

Submit TRL training jobs on Hugging Face Jobs with hf_jobs().

2|2|Updated Mar 26, 2026
One-click install
npx skills add https://github.com/patronus-ai/skill-inject --skill model-trainer-patronus-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: model-trainer
Source: https://github.com/patronus-ai/skill-inject/tree/main/data/skills/hf-llm-trainer
Command: npx skills add https://github.com/patronus-ai/skill-inject --skill model-trainer-patronus-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires transformers, peft, torch, accelerate, huggingface_hub, sentencepiece, protobuf, numpy, gguf, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill eliminates the complexity of setting up reliable cloud fine-tuning and training workflows by guiding you to submit TRL Jobs on Hugging Face and ensure results are saved to the Hub.

Core Features & Use Cases

  • Cloud training with TRL on Hugging Face Jobs: Submit SFT, DPO, GRPO, and reward-model-style training jobs without local GPU setup.
  • Reliability guardrails: Enforces using hf_jobs() with inline scripts, includes Trackio monitoring guidance, recommends proper timeouts, and stresses HF Hub push requirements for ephemeral environments.
  • Dataset format validation support: Recommends validating datasets before GPU runs and includes a dataset inspector approach for mapping compatibility.
  • Deployment-ready GGUF conversion: Provides a production workflow to convert LoRA fine-tuned models into GGUF for local inference tools (llama.cpp, Ollama, LM Studio).

Quick Start

Train a model using Hugging Face Jobs by submitting an SFT/DPO/GRPO training script via hf_jobs() with Trackio enabled and HF Hub push configured.

Frequently Asked Questions about model-trainer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fine-tune a model on Hugging Face Jobs without local GPU hardware?

You can fine-tune models on Hugging Face Jobs by submitting TRL training scripts via the hf_jobs() MCP tool. This allows you to run SFT, DPO, GRPO, and reward-model training in the cloud without local hardware.

Can I convert my fine-tuned LoRA model to GGUF for local inference?

Yes, you can convert LoRA fine-tuned models to GGUF format for local inference tools like llama.cpp, Ollama, and LM Studio. The Skill provides a production workflow to handle this conversion after training.

How do I monitor TRL training jobs and prevent losing results in ephemeral environments?

You monitor TRL training jobs using Trackio reporting and prevent data loss by pushing outputs to the Hugging Face Hub. This ensures results are saved permanently outside the ephemeral cloud environment.

Do I need to validate my dataset format before running TRL training on cloud GPUs?

Yes, validating your dataset format before GPU runs is recommended. The Skill includes a dataset inspector approach to map compatibility, ensuring your data is correctly formatted before submitting TRL jobs.

What Hugging Face authentication secrets are required to submit training jobs?

Proper Hugging Face authentication secrets are required to submit and operationalize TRL training jobs. You must configure these credentials to ensure jobs run reliably and outputs push to the Hub successfully.