skills.rest
punkt2
punkt2COMMUNITY
Explore all skills by this author →
GitHub

llm-fine-tuning

Fine-tune large language models with QLoRA, Spectrum, and full fine-tuning via TRL and PEFT.

Updated Jan 24, 2026
Search Tags:#spectrum#peft#qlora#trl#llm-fine-tuning#deepspeed
One-click install
npx skills add https://github.com/punkt2/llm-fine-tuning-skill --skill llm-fine-tuning
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: llm-fine-tuning
Source: https://github.com/punkt2/llm-fine-tuning-skill/tree/main
Command: npx skills add https://github.com/punkt2/llm-fine-tuning-skill --skill llm-fine-tuning

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires torch, transformers, trl, peft, datasets, huggingface_hub, liger_kernel, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Fine-tune large language models efficiently using modern techniques (QLoRA, Spectrum, full fine-tuning) with TRL and PEFT, enabling domain adaptation and customization for specific tasks.

Core Features & Use Cases

  • Supports QLoRA on consumer GPUs with 4-bit quantization and LoRA adapters.
  • Provides Spectrum and full fine-tuning workflows with distributed training support (DeepSpeed ZeRO3, FSDP) and model merging.
  • Covers dataset preparation, training configuration, evaluation, and post-training steps including merging adapters and pushing to hub.

Quick Start

Run the supervised fine-tuning workflow by executing the training script with a YAML config.

Frequently Asked Questions about llm-fine-tuning

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

FAQPage Schema
How do I fine-tune an LLM with QLoRA on a single GPU?▼

Fine-tune an LLM with QLoRA on a single GPU using 4-bit quantization and LoRA adapters. This approach reduces memory consumption, enabling domain adaptation on consumer hardware without requiring multi-node clusters.

Can I use DeepSpeed ZeRO3 for distributed LLM fine-tuning?▼

Yes, you can use DeepSpeed ZeRO3 for distributed LLM fine-tuning. It supports Spectrum and full fine-tuning workflows across multi-node clusters, enabling scalable training and model merging for large models.

What is the best way to configure TRL and PEFT for supervised fine-tuning?▼

The best way to configure TRL and PEFT for supervised fine-tuning is through a YAML-driven configuration. This method streamlines dataset preparation, training setup, and evaluation workflows into a single executable script.

How does Spectrum fine-tuning compare to full parameter training?▼

Spectrum fine-tuning offers a targeted alternative to full parameter training by selectively updating weights. Both methods are supported, with Spectrum providing efficient domain adaptation while full training maximizes performance improvements.

Do I need to merge LoRA adapters after training to evaluate the model?▼

Yes, you need to merge LoRA adapters after training to evaluate the model. The workflow includes post-training steps to merge adapters and optionally push the consolidated model to the Hugging Face Hub.

What datasets are required for large language model domain adaptation?▼

Large language model domain adaptation requires prepared datasets for supervised fine-tuning. The workflow covers dataset preparation, allowing you to customize training data for specific tasks and performance improvements.

Related Skills

Explore More Software Engineering Skills→
huggingfacehuggingface
10.9k

hugging-face-model-trainer

Train and fine-tune language models on Hugging Face Jobs using TRL methods.

Official
Advanced
ovachieverovachiever
52

llama-cpp

Run LLM inference on CPU and non-NVIDIA hardware with GGUF quantization.

Community
Advanced
pmcdowallpmcdowall

fine-tuning-with-trl

Align large language models via reinforcement learning and preference-based fine-tuning workflows.

Community
Advanced
JKhyroJKhyro

peft-fine-tuning

Fine-tune large language models with parameter-efficient adapters like LoRA and QLoRA.

Community
Advanced