--- title: SLM Instruction Tuning using Unsloth emoji: 🐨 colorFrom: yellow colorTo: indigo sdk: gradio sdk_version: 4.38.1 app_file: app.py pinned: false license: apache-2.0 --- ----- # LoRA Instruction Tuning using Unsloth ### What? This Gradio app is a simple interface to access [unsloth AI's](https://github.com/unslothai) fine-tuning methods hosted on [Huggingface Spaces](https://huggingface.co/docs/hub/en/spaces-overview). This outputs of this fine-tuning will be instruction tuned LoRA weights that will be uploaded into your personal huggingface models repository. This is more for personal use since the A100 GPUs only last for a stated runtime (~60 seconds or more if specified). ### Why? The goal of this demo is to show how you can tune your own language models leveraging industry standard compute and fine tuning methods using a simple point-and-click UI. This is a demo and not a production application and is hosted here simply to . This application is subject a demand queue. ### How? Just start by following the guide below: 1) Navigate through the tabs at the top from left to right. 2) Basic Setup: Populate your username, repository, token, and model details. 3) Upload data: Either from transformers or your local jsonl file. Please view [this guide](https://platform.openai.com/docs/guides/fine-tuning/preparing-your-dataset) for best practices. 4) Fine-tune Model: Eat a snack and wait as you train the model for your use case. For GPU runtimes longer than a minute, remove the imports to huggingface spaces and decorators and run on your local GPU or migrate this work to a workspace like [lightning AI](https://lightning.ai/). ### Coming soon! - More models and added flexibility with guardrails on hyperparameter tuning. - Downloads for a [WandB](https://wandb.ai/home) logger for training monitoring. ### Other resources. - [Unsloth's notebooks](https://colab.research.google.com/drive/1hhdhBa1j_hsymiW9m-WzxQtgqTH_NHqi?usp=sharing) to look at what is going on under the hood. [](https://github.com/unslothai/unsloth)