|
--- |
|
license: apache-2.0 |
|
--- |
|
# Finetuning llama 2 on our celebrity news dataset located [here](https://huggingface.co/datasets/2nji/makebelieve-480) |
|
|
|
Disclaimer: This is still work in progress as we need to preprocess our celebrity news dataset to match Llama 2's prompt format as described [here](https://huggingface.co/blog/llama2#how-to-prompt-llama-2) |
|
|
|
## Reserve GPU on g5k |
|
|
|
Log into your Grid5000 account using ssh and run the following code in the terminal |
|
|
|
```script |
|
oarsub -p "cluster='graffiti'" -l gpu=1 -I -q production |
|
``` |
|
|
|
Wait till GPUs are available and assigned to you, if you need more information about g5k, you can refer to [here](https://www.grid5000.fr/w/Getting_Started) |
|
|
|
### Create a virtual environment |
|
|
|
- Installing PIP |
|
|
|
```script |
|
pip install virtualenv |
|
``` |
|
|
|
- Creating environment |
|
|
|
```script |
|
virtualenv venv |
|
``` |
|
|
|
- Activating environment |
|
|
|
```script |
|
source venv/bin/activate |
|
``` |
|
|
|
## Install requirements file |
|
|
|
```script |
|
pip install -r requirements.txt |
|
``` |
|
|
|
## Running the script to finetune Llama-2-7b-chat-hf and push to huggingface model repository |
|
|
|
```script |
|
python makebelieve.py |
|
``` |
|
|