pzal / mim-nlp-project

mim-nlp-project

Installation

• pip install -e .

• Create .env file and provide your NEPTUNE_API_TOKEN and HF_TOKEN token (with write access). Optionally override values from .public_env.
• You need HuggingFace initialized (logged in).

Usage

Baseline training

python3 scripts/train.py --model baseline --embedding-size 64 --version v2 --batch-size-per-gpu 8 --tag <your_tag_for_neptune>

Mandatory arguments:

• --model baseline | matryoshka
• --embedding-size int
• --version string

Optional arguments:

• --batch-size-per-gpu : 8 by default
• --tag "some tag" --tag "some other tag" [] by default
• --load-pretrained False by default

MTEB eval

python3 scripts/evaluate_on_mteb.py