Model Card for Spivavtor-xxl

This model was obtained by instruction tuning CohereForAI/aya-101 model on the Spivavtor dataset. All details of the dataset and fine tuning process can be found in our paper.

Paper: Spivavtor: An Instruction Tuned Ukrainian Text Editing Model

Authors: Aman Saini, Artem Chernodub, Vipul Raheja, Vivek Kulkarni

Model Details

Model Description

• Language: Ukrainian
• Finetuned from model: CohereForAI/aya-101

How to use

We make the following models available from our paper.

Model	Number of parameters	Reference name in Paper
Spivavtor-large	1.2B	SPIVAVTOR-MT0-LARGE
Spivavtor-xxl	13B	SPIVAVTOR-AYA-101

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("grammarly/spivavtor-xxl")
model = AutoModelForSeq2SeqLM.from_pretrained("grammarly/spivavtor-xxl")

# Paraphrase the sentence: What is the greatest compliment that you ever received from anyone?
input_text = 'Перефразуйте речення: Який найкращий комплімент, який ти отримував від будь-кого?'

inputs = tokenizer.encode(input_text, return_tensors="pt")
output = model.generate(inputs, max_length=256)
output_text = tokenizer.decode(outputs[0], skip_special_tokens=True)