# FastLoRAChat Instruct-tune LLaMA on consumer hardware with shareGPT data

**URL:** <https://discuss.huggingface.co/t/fastlorachat-instruct-tune-llama-on-consumer-hardware-with-sharegpt-data/37077>\
**Category:** Show and Tell\
**Created:** [April 19, 2023, 7:23am UTC](https://discuss.huggingface.co/t/fastlorachat-instruct-tune-llama-on-consumer-hardware-with-sharegpt-data/37077 "2023-04-19T07:23:34Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![icybee](https://avatars.discourse-cdn.com/v4/letter/i/db5fbb/32.png) [@icybee](https://discuss.huggingface.co/u/icybee)\
**Post date:** [April 19, 2023, 7:23am UTC](https://discuss.huggingface.co/t/fastlorachat-instruct-tune-llama-on-consumer-hardware-with-sharegpt-data/37077/1 "2023-04-19T07:23:34Z")

</div>

Announcing [FastLoRAChat](https://github.com/bupticybee/FastLoRAChat) , training chatGPT without A100.

Releasing model: [icybee/fast\_lora\_chat\_v1\_sunlight · Hugging Face](https://huggingface.co/icybee/fast_lora_chat_v1_sunlight)

The purpose of this project is to produce similar result to the Fastchat model, but in much cheaper hardware (especially in non-Ampere GPUs).

This repository combined features of alpaca-lora and Fastchat:

1. Like Fastchat, support multilanguage and multi round chat.
2. Like alpaca-lora, support training and inference on low-end graphic cards (using LORA).
3. Opensource everything, include dataset, training code, export model code, and more.