Buidling llama Fashion

A end to end demo builds a llama model with fashion data (Routine) is a local branch in Vietnam.

Tasks

1. A script scrapes Routine's data (a fashion data) in Vietnam.
2. Building a llama model with fashion data.
3. Testing model.
4. Run application by streamlit.

Structure of data:

1. 'Link': Link of item.
2. 'Category_Type': Target gender,
3. 'Image Front': Front Image link,
4. 'Image Back': Back Image link,
5. 'Product_ID': Product Id,
6. 'Description': Product description,
7. 'Price': Product Price,
8. 'Brand': Product brand.

A csv format.

Scraping data

python data/database_scrape.py

Training llama

• In the demo, I already use TinyPixel/small-llama2 (https://huggingface.co/TinyPixel/small-llama2)

python train.py

Inference

python inference.py

Running application.

streamlit run app.py