search

OpenAdaptAI / OpenAdapt

Open Source Generative Process Automation (i.e. Generative RPA). AI-First Process Automation with Large ([Language (LLMs) / Action (LAMs) / Multimodal (LMMs)] / Visual Language (VLMs)) Models
https://www.OpenAdapt.AI
MIT License
997 stars 138 forks source link

Investigate QLoRA #208

Open abrichr opened 1 year ago

abrichr commented 1 year ago

https://github.com/artidoro/qlora

https://arxiv.org/abs/2305.14314

We present QLoRA, an efficient finetuning approach that reduces memory usage enough to finetune a 65B parameter model on a single 48GB GPU while preserving full 16-bit finetuning task performance. … Our best model family, which we name Guanaco, outperforms all previous openly released models on the Vicuna benchmark, reaching 99.3% of the performance level of ChatGPT while only requiring 24 hours of finetuning on a single GPU. … Our results show that QLoRA finetuning on a small high-quality dataset leads to state-of-the-art results, even when using smaller models than the previous SoTA

Ultimately we want to get to a place where we can fine tune models offline. Maybe that involves QLoRA, maybe something else.

This task involves exploring what it would take to get QLoRA (or some alternative) up and running locally.

FFFiend commented 1 year ago

See https://github.com/kuleshov-group/llmtune/blob/main/llmtune/engine/lora/lora.py for an implementation.