Multilingual Pruning Research

Description

A collection of pruned models on different pruning methods, sparsity types and ratios, calibration sets as well as calibration languages. All of the model names are in the format:

pruned-<original model name>-<pruning method>-<sparsity ratio>-<sparsity type>-<calibration dataset>-<calibration languages>-<seed>

Usage

Load the model using the transformers library, which can be installed using pip install transformers:

# main.py
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("multilingual-pruning/<model name>")

Dependencies

• transformers==4.42.3
• numpy==1.26.4