Back to all skills

🧬

Data & Analytics

Unsloth Fine-Tuning

Fine-tune open-weight LLMs faster and cheaper with Unsloth.

4.7rating

5,200 installs

mlops/training/unsloth

Max required

About this skill

Unsloth Fine-Tuning wraps the Unsloth library to LoRA/QLoRA-fine-tune a Llama, Mistral, or Qwen model on your own dataset — typically 2–5× faster than vanilla HuggingFace trainer, with lower VRAM. Returns a merged checkpoint plus eval metrics. Ready to serve via vLLM Inference afterwards.

What it does

LoRA and QLoRA fine-tuning
2–5× faster than vanilla trainer
Lower VRAM footprint
Runs on a single consumer GPU
Returns merged checkpoint + eval metrics

Use cases

Fine-tune a 7B model on a domain dataset overnight
Adapt a base model to a specific writing style
Train a classifier head on top of a frozen backbone

Related skills

Data & Analytics

Personal Knowledge Base

Persistent notes that the assistant actually remembers and can search.

20,300 installs

Data & Analytics

Data Analysis (Jupyter)

Run Python on your data with a live Jupyter kernel, from a natural-language brief.

16,200 installs

Data & Analytics

Chart Builder

Generate publication-ready charts from a pasted table or CSV.

11,100 installs

Data & Analytics

Database Query Engine

Query your own Postgres, MySQL, or SQLite with natural-language prompts.