File size: 911 Bytes
a476eba 1c43b74 a476eba 1c43b74 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 |
---
license: gpl-3.0
datasets:
- yahma/alpaca-cleaned
language:
- en
pipeline_tag: text2text-generation
tags:
- alpaca
- llama
- chat
---
This repository comes with LoRA checkpoint to make LLaMA into a chatbot like language model. The checkpoint is the output of instruction following fine-tuning process with the following settings on 8xA100(40G) DGX system.
- Dataset: [cleaned-up Alpaca dataset](https://github.com/gururise/AlpacaDataCleaned) up to 04/06/23
- Training script: borrowed from the official [Alpaca-LoRA](https://github.com/tloen/alpaca-lora) implementation
- Training script:
```shell
python finetune.py \
--base_model='decapoda-research/llama-13b-hf' \
--num_epochs=10 \
--cutoff_len=512 \
--group_by_length \
--output_dir='./lora-alpaca' \
--lora_target_modules='[q_proj,k_proj,v_proj,o_proj]' \
--lora_r=16 \
--batch_size=... \
--micro_batch_size=...
``` |