Introducing Unsloth Studio ✨ A new open-source web UI to train and run LLMs. • Run models locally on Mac, Windows, Linux • Train 500+ models 2x faster with 70% less VRAM • Supports GGUF, vision, audio, embedding models • Auto-create datasets from PDF, CSV, DOCX • Self-healing tool... calling and code execution • Compare models side by side + export to GGUF GitHub: Blog and Guide: Available now on Hugging Face, NVIDIA, Docker and Colab.show more

Transform PDFs, CSV, DOCX, TXT or any file into a structured synthetic datasets via Unsloth Data Recipes. Build and edit your datasets visually via a graph-node workflow and use them for fine-tuning. Powered by @NVIDIA DataDesigner.

Unsloth Studio allows LLMs to run code and programs in a sandbox so it can calculate, analyze data, test code, generate files, or verify an answer with actual computation. This makes answers from models more reliable and accurate.

Some screenshots: SVG rendering of a cute sloth

• Auto-create datasets from PDF, CSV, DOCX Killer feature

six months ago this required a team of ML engineers and a $50k cloud bill. now it runs on your laptop

congrats on launch! this is an awesome tool 🔥

Can I use this if I’m SSHing into a machine?

congrats on launch, looking forward to MLX support ✨

Amazing tool release! Let me deep dive on it! Thanks!

Incredible timing! We're currently using Unsloth to fine-tune DeepCoder-14B on a local RTX 5070 Ti. We are building a specialized Terraform AI trained strictly on compiler-verified, multi-region AWS architectures. The memory optimization in Unsloth is the only reason these fits on a consumer GPU. Can't wait to move the pipeline to Studio.

holy crap 70% less VRAM is massive - training on my local setup just became actually viable

my goat @danielhanchen et al cooked here. there is too much ground for anyone to explore their research hacks

Nice! Congrats!

The auto-dataset creation from PDFs and CSVs is the part that actually removes a real bottleneck. Most people quit fine-tuning before they start because data prep takes longer than training.

70% less VRAM is the number that matters here. most people can't fine-tune because they don't have A100s sitting around. this makes local training actually practical.

Oh that's really cool, I love the built-in data pipeline. Congrats guys!

really cool

Very cool! I noticed there are now both AGPL and Apache licenses in the repo, could you clarify?

@nisten freaking cool!

I love you Unsloth 💙

2x faster training with 70% less VRAM and auto-dataset creation from PDFs? That's not just a better UI — that's removing almost every barrier that kept fine-tuning out of reach for most developers. Really curious how the self-healing tool calling holds up on complex agentic tasks. This could genuinely change who gets to train their own models.

You guys are the 🐐

This is a huge deal. A tool like this has been needed for a long time!

@ClementDelangue Unsloth Studio: now your laptop can train LLMs. Next step: your LLM can hire other LLMs, expense the compute, and file its own tax returns.

@danielhanchen This is going to be a game changer!! Been waiting for it for a long time. Keep cooking, boys!!

Exactly the tool I need

@LewisMenelaws Thank you for releasing this. I’m gonna have to check this out.

Awesome!! Love you guys continuing to lower the barrier! Thanks for all you put into it!

This is the future we need! Thank you !

Good work

this is crazyyyyy

Always cookin

You guys are the best! Thank you!

2x faster training with 70% less VRAM? That's not optimization, that's straight up wizardry. Finally, a local LLM setup that won't require selling a kidney for GPU memory

this is extremely interesting will definitely be checking it out