skills.rest
ScientiaCapital
ScientiaCapitalCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

unsloth-training

Train LLMs with Unsloth using GRPO RL or SFT with FP8 optimization.

28|3|Updated Dec 10, 2025
Search Tags:#training#fp8#rl#packing#sft#grpo#unsloth
One-click install
npx skills add https://github.com/ScientiaCapital/skills --skill unsloth-training
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: unsloth-training
Source: https://github.com/ScientiaCapital/skills/tree/main/active/unsloth-training-skill
Command: npx skills add https://github.com/ScientiaCapital/skills --skill unsloth-training

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Fine-tune large language models efficiently using Unsloth via GRPO reinforcement learning or supervised fine-tuning, reducing memory and time costs while enabling advanced features like FP8, vision fine-tuning, and mobile deployment.

Core Features & Use Cases

  • GRPO RL training with reward design and LoRA adapters
  • SFT training with packing and long-context options
  • FP8 training for significant VRAM savings
  • Vision fine-tuning for VLM tasks
  • Docker-based training and reproducible environments
  • Mobile deployment support through QAT/ExecuTorch export
  • GGUF export options for various serving backends
  • End-to-end pipeline from data prep to export and deployment

Quick Start

Run the GRPO training script with a small dataset to start RL fine-tuning.

Frequently Asked Questions about unsloth-training

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

FAQPage Schema
How do I fine-tune an LLM with GRPO reinforcement learning using Unsloth?▼

Fine-tune LLMs with GRPO RL by configuring reward design and LoRA adapters to achieve targeted behavior and faster convergence. You can start RL training quickly by running the provided GRPO training script with a small dataset.

What's the best way to reduce VRAM usage during supervised fine-tuning?▼

Reduce VRAM usage during supervised fine-tuning by applying FP8 memory optimization and sequence packing. These features allow you to train large language models efficiently while handling long-context options.

Can I export fine-tuned models to GGUF for deployment on Ollama and vLLM?▼

Yes, you can export fine-tuned models to GGUF formats to enable deployment across various serving backends. This supports seamless integration with vLLM, Ollama, and LM Studio for production inference.

Does Unsloth support vision fine-tuning for VLM tasks?▼

Yes, Unsloth supports vision fine-tuning for VLM tasks. You can train vision-language models alongside standard LLM training pipelines to handle multimodal inputs efficiently.

How do I deploy fine-tuned LLMs to mobile runtimes?▼

Deploy fine-tuned LLMs to mobile runtimes by exporting models through QAT and ExecuTorch formats. This enables efficient inference on mobile devices after completing your GRPO or SFT training.

Do I need Docker to set up a reproducible LLM training environment?▼

Docker is supported to configure reproducible training environments for fine-tuning LLMs with Unsloth. This ensures consistent execution from research experiments to production deployment across different machines.

Related Skills

Explore More Software Engineering Skills→
duyetduyet

unsloth-training

Fine-tune LLMs with Unsloth using GRPO and SFT methods.

Community
Advanced
yankeeinlondonyankeeinlondon
3

model-citizen

Inventory and manage local LLM models across Ollama, LM Studio, and Llama.cpp.

Community
Advanced
ScientiaCapitalScientiaCapital
1

model-deployment

Export fine-tuned models to GGUF and deploy with vLLM or Ollama.

Community
Advanced
ScientiaCapitalScientiaCapital
1

unsloth-finetuning

Fine-tune Llama, Mistral, Phi, or Gemma models with LoRA and 4-bit quantization.

Community
Intermediate