TowerBase-7B-v0.1/README.md

license, language, metrics, pipeline_tag, model-index

license	language	metrics	pipeline_tag	model-index
cc-by-nc-4.0	en de fr zh pt nl ru ko it es	comet	translation	name results TowerBase-7B-v0.1 task dataset metrics source type name text-generation Text Generation name type config split args AI2 Reasoning Challenge (25-Shot) ai2_arc ARC-Challenge test num_few_shot 25 type value name acc_norm 51.02 normalized accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Unbabel/TowerBase-7B-v0.1 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type split args HellaSwag (10-Shot) hellaswag validation num_few_shot 10 type value name acc_norm 77.68 normalized accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Unbabel/TowerBase-7B-v0.1 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args MMLU (5-Shot) cais/mmlu all test num_few_shot 5 type value name acc 43.48 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Unbabel/TowerBase-7B-v0.1 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args TruthfulQA (0-shot) truthful_qa multiple_choice validation num_few_shot 0 type value mc2 37.29 url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Unbabel/TowerBase-7B-v0.1 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args Winogrande (5-shot) winogrande winogrande_xl validation num_few_shot 5 type value name acc 72.06 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Unbabel/TowerBase-7B-v0.1 Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args GSM8k (5-shot) gsm8k main test num_few_shot 5 type value name acc 13.12 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Unbabel/TowerBase-7B-v0.1 Open LLM Leaderboard

Model Card for TowerBase-7B-v0.1

Model Details

Model Description

TowerBase-7B is a language model that results from continuing the pretraining of Llama 2 on a mix of 20 billion tokens of monolingual data in ten different languages — English, Portuguese, Spanish, French, German, Dutch, Italian, Korean, Chinese, Russian — and bilingual data. TowerBase-7B-v0.1 is the first model in the series. The resulting model shows improved performance on the supported languages, while maintaining Llama 2's capabilities on English. It is particularly well-suited for fine-tuning on translation and related tasks: check out TowerInstruct.

We will release more details in the upcoming technical report.

• Developed by: Unbabel, Instituto Superior Técnico, CentraleSupélec University of Paris-Saclay
• Model type: A 7B parameter model built on top of Llama 2 by continuing pretraining on multilingual data.
• Language(s) (NLP): English, Portuguese, Spanish, French, German, Dutch, Italian, Korean, Chinese, Russian
• License: CC-BY-NC-4.0, Llama 2 is licensed under the LLAMA 2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved.

Intended uses & limitations

The model is intended for research purposes in the 10 languages it supports. The model is able to perform well on translation and related tasks (e.g., APE, GEC) on a few-shot regime. It can also be fine-tuned to perform these tasks in a zero-shot fashion (see TowerInstruct, as well as other multilingual tasks.

Out-of-Scope Use

The model is not guaranteed to perform well for languages other than the 10 languages it supports.

Bias, Risks, and Limitations

TowerBase-v0.1 has not been aligned to human preferences, so the model may generate problematic outputs (e.g., hallucinations, harmful content, or false statements).

Run the model

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Unbabel/TowerBase-7B-v0.1"
tokenizer = AutoTokenizer.from_pretrained(model_id)

model = AutoModelForCausalLM.from_pretrained(model_id)

text = "English: My name is TowerBase.\nPortuguese:"
inputs = tokenizer(text, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=20)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Data

Filtered versions of mc4 and bilingual data from various sources (e.g., OPUS).

Citation

@misc{tower_llm_2024,
      title={Tower: An Open Multilingual Large Language Model for Translation-Related Tasks}, 
      author={Duarte M. Alves and José Pombal and Nuno M. Guerreiro and Pedro H. Martins and João Alves and Amin Farajian and Ben Peters and Ricardo Rei and Patrick Fernandes and Sweta Agrawal and Pierre Colombo and José G. C. de Souza and André F. T. Martins},
      year={2024},
      eprint={2402.17733},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}