---
datasets:
- bigscience/xP3
- mc4
- Leipzig_corpora_collection
license: apache-2.0
language:
- en
- de
tags:
- text2text-generation
pipeline_tag: text2text-generation
---
# Language Adaption of mT0
This model is an adaption of the mT0 variant (see below) for languages en, de based on the methodology in [Load What You Need: Smaller Versions of Multilingual BERT](https://arxiv.org/ftp/arxiv/papers/2010/2010.05609.pdf).
The vocabulary has been adjusted to contain the top 2000 tokens overall as well as
- the top 15000 tokens from en
- the top 30000 tokens from de
and 100 special tokens.
The new vocabulary has been computed using
- deu_news_2023_1M for de
- eng_news_2023_1M for en
from the [Leipzig language corpora collection](https://wortschatz.uni-leipzig.de/en/download/).
According to the mentioned article the perfarmance degradation can be expected to be small (but no evaluation was done).
# Table of Contents
1. [Model Summary](#model-summary)
2. [Use](#use)
3. [Limitations](#limitations)
4. [Training](#training)
5. [Evaluation](#evaluation)
7. [Citation](#citation)
# Model Summary
> We present BLOOMZ & mT0, a family of models capable of following human instructions in dozens of languages zero-shot. We finetune BLOOM & mT5 pretrained multilingual language models on our crosslingual task mixture (xP3) and find our resulting models capable of crosslingual generalization to unseen tasks & languages.
- **Repository:** [bigscience-workshop/xmtf](https://github.com/bigscience-workshop/xmtf)
- **Paper:** [Crosslingual Generalization through Multitask Finetuning](https://arxiv.org/abs/2211.01786)
- **Point of Contact:** [Niklas Muennighoff](mailto:niklas@hf.co)
- **Languages:** Refer to [mc4](https://huggingface.co/datasets/mc4) for pretraining & [xP3](https://huggingface.co/datasets/bigscience/xP3) for finetuning language proportions. It understands both pretraining & finetuning languages.
- **BLOOMZ & mT0 Model Family:**
# Use
## Intended use
We recommend using the model to perform tasks expressed in natural language. For example, given the prompt "*Translate to English: Je t’aime.*", the model will most likely answer "*I love you.*". Some prompt ideas from our paper:
- 一个传奇的开端，一个不灭的神话，这不仅仅是一部电影，而是作为一个走进新时代的标签，永远彪炳史册。你认为这句话的立场是赞扬、中立还是批评?
- Suggest at least five related search terms to "Mạng neural nhân tạo".
- Write a fairy tale about a troll saving a princess from a dangerous dragon. The fairy tale is a masterpiece that has achieved praise worldwide and its moral is "Heroes Come in All Shapes and Sizes". Story (in Spanish):
- Explain in a sentence in Telugu what is backpropagation in neural networks.
**Feel free to share your generations in the Community tab!**
## How to use
### CPU
Click to expand
```python
# pip install -q transformers
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
checkpoint = "bigscience/mt0-large"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
inputs = tokenizer.encode("Translate to English: Je t’aime.", return_tensors="pt")
outputs = model.generate(inputs)
print(tokenizer.decode(outputs[0]))
```
### GPU
Click to expand
```python
# pip install -q transformers accelerate
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
checkpoint = "bigscience/mt0-large"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint, torch_dtype="auto", device_map="auto")
inputs = tokenizer.encode("Translate to English: Je t’aime.", return_tensors="pt").to("cuda")
outputs = model.generate(inputs)
print(tokenizer.decode(outputs[0]))
```
### GPU in 8bit
Click to expand
```python
# pip install -q transformers accelerate bitsandbytes
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
checkpoint = "bigscience/mt0-large"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint, device_map="auto", load_in_8bit=True)
inputs = tokenizer.encode("Translate to English: Je t’aime.", return_tensors="pt").to("cuda")
outputs = model.generate(inputs)
print(tokenizer.decode(outputs[0]))
```
###
# Limitations
**Prompt Engineering:** The performance may vary depending on the prompt. For BLOOMZ models, we recommend making it very clear when the input stops to avoid the model trying to continue it. For example, the prompt "*Translate to English: Je t'aime*" without the full stop (.) at the end, may result in the model trying to continue the French sentence. Better prompts are e.g. "*Translate to English: Je t'aime.*", "*Translate to English: Je t'aime. Translation:*" "*What is "Je t'aime." in English?*", where it is clear for the model when it should answer. Further, we recommend providing the model as much context as possible. For example, if you want it to answer in Telugu, then tell the model, e.g. "*Explain in a sentence in Telugu what is backpropagation in neural networks.*".
# Training
## Model
- **Architecture:** Same as [mt5-large](https://huggingface.co/google/mt5-large), also refer to the `config.json` file
- **Finetuning steps:** 25000
- **Finetuning tokens:** 4.62 billion
- **Precision:** bfloat16
## Hardware
- **TPUs:** TPUv4-64
## Software
- **Orchestration:** [T5X](https://github.com/google-research/t5x)
- **Neural networks:** [Jax](https://github.com/google/jax)
# Evaluation
We refer to Table 7 from our [paper](https://arxiv.org/abs/2211.01786) & [bigscience/evaluation-results](https://huggingface.co/datasets/bigscience/evaluation-results) for zero-shot results on unseen tasks. The sidebar reports zero-shot performance of the best prompt per dataset config.
# Citation
```bibtex
@article{muennighoff2022crosslingual,
title={Crosslingual generalization through multitask finetuning},
author={Muennighoff, Niklas and Wang, Thomas and Sutawika, Lintang and Roberts, Adam and Biderman, Stella and Scao, Teven Le and Bari, M Saiful and Shen, Sheng and Yong, Zheng-Xin and Schoelkopf, Hailey and others},
journal={arXiv preprint arXiv:2211.01786},
year={2022}
}
```