Tower
Collection
Model weights and SFT data for Tower. • 11 items • Updated • 32
How to use Unbabel/TowerInstruct-WMT24-Chat-7B with Transformers:
# Use a pipeline as a high-level helper
# Warning: Pipeline type "translation" is no longer supported in transformers v5.
# You must load the model directly (see below) or downgrade to v4.x with:
# 'pip install "transformers<5.0.0'
from transformers import pipeline
pipe = pipeline("translation", model="Unbabel/TowerInstruct-WMT24-Chat-7B") # Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("Unbabel/TowerInstruct-WMT24-Chat-7B")
model = AutoModelForCausalLM.from_pretrained("Unbabel/TowerInstruct-WMT24-Chat-7B")TowerInstruct-WMT24-Chat-7B is a language model that results from fine-tuning TowerBase on TowerBlocks and the WMT24 Chat MT Shared task training set.
TowerInstruct-WMT24-Chat-7B was the best submission of the shared task, winning on all 10 language pairs according to human evaluation (see the task's findings paper here).
It is specifically tailoured for context-aware translation of customer support chats.
Check out our paper for more details and information on training and data.
Information on model usage, out-of-scope usages, risks, etc... are the same as the model cards of the TowerInstruct models.
@inproceedings{pombal2024improving,
title={Improving Context Usage for Translating Bilingual Customer Support Chat with Large Language Models},
author={Pombal, Jos{\'e} and Agrawal, Sweta and Martins, Andr{\'e} FT},
booktitle={Proceedings of the Ninth Conference on Machine Translation},
pages={993--1003},
year={2024}
}