datasets	license	language	programming_language	pipeline_tag	widget	model-index
bigscience/xP3	bigscience-bloom-rail-1.0	ak ar as bm bn ca code en es eu fon fr gu hi id ig ki kn lg ln ml mr ne nso ny or pa pt rn rw sn st sw ta te tn ts tum tw ur vi wo xh yo zh zu	C C++ C# Go Java JavaScript Lua PHP Python Ruby Rust Scala TypeScript	text-generation	text example_title 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。Would you rate the previous review as positive, neutral or negative? zh-en sentiment text example_title 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？ zh-zh sentiment text example_title Suggest at least five related search terms to "Mạng neural nhân tạo". vi-en query text example_title Proposez au moins cinq mots clés concernant «Réseau de neurones artificiels». fr-fr query text example_title Explain in a sentence in Telugu what is backpropagation in neural networks. te-en qa text example_title Why is the sky blue? en-en qa text example_title 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): es-en fable text example_title Write a fable about wood elves living in a forest that is suddenly invaded by ogres. The fable is a masterpiece that has achieved praise worldwide and its moral is "Violence is the last refuge of the incompetent". Fable (in Hindi): hi-en fable	name results bloomz-1b1 task dataset metrics type Coreference resolution type name config split revision winogrande Winogrande XL (xl) xl validation a80f460359d1e9a67c006011c94de42a8759430c type value Accuracy 52.33 task dataset metrics type Coreference resolution type name config split revision Muennighoff/xwinograd XWinograd (en) en test 9dd5ea5505fad86b7bedad667955577815300cee type value Accuracy 50.49 task dataset metrics type Coreference resolution type name config split revision Muennighoff/xwinograd XWinograd (fr) fr test 9dd5ea5505fad86b7bedad667955577815300cee type value Accuracy 59.04 task dataset metrics type Coreference resolution type name config split revision Muennighoff/xwinograd XWinograd (jp) jp test 9dd5ea5505fad86b7bedad667955577815300cee type value Accuracy 51.82 task dataset metrics type Coreference resolution type name config split revision Muennighoff/xwinograd XWinograd (pt) pt test 9dd5ea5505fad86b7bedad667955577815300cee type value Accuracy 54.75 task dataset metrics type Coreference resolution type name config split revision Muennighoff/xwinograd XWinograd (ru) ru test 9dd5ea5505fad86b7bedad667955577815300cee type value Accuracy 53.97 task dataset metrics type Coreference resolution type name config split revision Muennighoff/xwinograd XWinograd (zh) zh test 9dd5ea5505fad86b7bedad667955577815300cee type value Accuracy 55.16 task dataset metrics type Natural language inference type name config split revision anli ANLI (r1) r1 validation 9dbd830a06fea8b1c49d6e5ef2004a08d9f45094 type value Accuracy 33.3 task dataset metrics type Natural language inference type name config split revision anli ANLI (r2) r2 validation 9dbd830a06fea8b1c49d6e5ef2004a08d9f45094 type value Accuracy 33.5 task dataset metrics type Natural language inference type name config split revision anli ANLI (r3) r3 validation 9dbd830a06fea8b1c49d6e5ef2004a08d9f45094 type value Accuracy 34.5 task dataset metrics type Natural language inference type name config split revision super_glue SuperGLUE (cb) cb validation 9e12063561e7e6c79099feb6d5a493142584e9e2 type value Accuracy 58.93 task dataset metrics type Natural language inference type name config split revision super_glue SuperGLUE (rte) rte validation 9e12063561e7e6c79099feb6d5a493142584e9e2 type value Accuracy 65.7 task dataset metrics type Natural language inference type name config split revision xnli XNLI (ar) ar validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 46.59 task dataset metrics type Natural language inference type name config split revision xnli XNLI (bg) bg validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 40.4 task dataset metrics type Natural language inference type name config split revision xnli XNLI (de) de validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 40.12 task dataset metrics type Natural language inference type name config split revision xnli XNLI (el) el validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 39.32 task dataset metrics type Natural language inference type name config split revision xnli XNLI (en) en validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 47.11 task dataset metrics type Natural language inference type name config split revision xnli XNLI (es) es validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 47.55 task dataset metrics type Natural language inference type name config split revision xnli XNLI (fr) fr validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 48.51 task dataset metrics type Natural language inference type name config split revision xnli XNLI (hi) hi validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 42.89 task dataset metrics type Natural language inference type name config split revision xnli XNLI (ru) ru validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 42.81 task dataset metrics type Natural language inference type name config split revision xnli XNLI (sw) sw validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 41.29 task dataset metrics type Natural language inference type name config split revision xnli XNLI (th) th validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 42.93 task dataset metrics type Natural language inference type name config split revision xnli XNLI (tr) tr validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 37.51 task dataset metrics type Natural language inference type name config split revision xnli XNLI (ur) ur validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 41.37 task dataset metrics type Natural language inference type name config split revision xnli XNLI (vi) vi validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 47.19 task dataset metrics type Natural language inference type name config split revision xnli XNLI (zh) zh validation a5a45e4ff92d5d3f34de70aaf4b72c3bdf9f7f16 type value Accuracy 47.63 task dataset metrics type Program synthesis type name config split revision openai_humaneval HumanEval None test e8dc562f5de170c54b5481011dd9f4fa04845771 type value Pass@1 2.62 type value Pass@10 6.22 type value Pass@100 11.68 task dataset metrics type Sentence completion type name config split revision story_cloze StoryCloze (2016) 2016 validation e724c6f8cdf7c7a2fb229d862226e15b023ee4db type value Accuracy 62.75 task dataset metrics type Sentence completion type name config split revision super_glue SuperGLUE (copa) copa validation 9e12063561e7e6c79099feb6d5a493142584e9e2 type value Accuracy 63.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (et) et validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 55.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (ht) ht validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 52.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (id) id validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 60.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (it) it validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 56.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (qu) qu validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 56.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (sw) sw validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 64.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (ta) ta validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 57.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (th) th validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 59.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (tr) tr validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 55.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (vi) vi validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 63.0 task dataset metrics type Sentence completion type name config split revision xcopa XCOPA (zh) zh validation 37f73c60fb123111fa5af5f9b705d0b3747fd187 type value Accuracy 61.0 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (ar) ar validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 53.54 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (es) es validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 58.37 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (eu) eu validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 52.35 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (hi) hi validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 55.92 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (id) id validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 57.97 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (my) my validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 47.05 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (ru) ru validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 50.3 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (sw) sw validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 49.97 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (te) te validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 55.86 task dataset metrics type Sentence completion type name config split revision Muennighoff/xstory_cloze XStoryCloze (zh) zh validation 8bb76e594b68147f1a430e86829d07189622b90d type value Accuracy 58.17

Table of Contents

1. Model Summary
2. Use
3. Limitations
4. Training
5. Evaluation
6. 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 the resulting models capable of crosslingual generalization to unseen tasks & languages.

• Repository: bigscience-workshop/xmtf
• Paper: Crosslingual Generalization through Multitask Finetuning
• Point of Contact: Niklas Muennighoff
• Languages: Refer to bloom for pretraining & xP3 for finetuning language proportions. It understands both pretraining & finetuning languages.
• BLOOMZ & mT0 Model Family:

Multitask finetuned on xP3. Recommended for prompting in English.
Parameters	300M	580M	1.2B	3.7B	13B	560M	1.1B	1.7B	3B	7.1B	176B
Finetuned Model	mt0-small	mt0-base	mt0-large	mt0-xl	mt0-xxl	bloomz-560m	bloomz-1b1	bloomz-1b7	bloomz-3b	bloomz-7b1	bloomz
Multitask finetuned on xP3mt. Recommended for prompting in non-English.
Finetuned Model					mt0-xxl-mt					bloomz-7b1-mt	bloomz-mt
Multitask finetuned on P3. Released for research purposes only. Strictly inferior to above models!
Finetuned Model					mt0-xxl-p3					bloomz-7b1-p3	bloomz-p3
Original pretrained checkpoints. Not recommended.
Pretrained Model	mt5-small	mt5-base	mt5-large	mt5-xl	mt5-xxl	bloom-560m	bloom-1b1	bloom-1b7	bloom-3b	bloom-7b1	bloom

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

# pip install -q transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

checkpoint = "bigscience/bloomz-1b1"

tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.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

# pip install -q transformers accelerate
from transformers import AutoModelForCausalLM, AutoTokenizer

checkpoint = "bigscience/bloomz-1b1"

tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.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

# pip install -q transformers accelerate bitsandbytes
from transformers import AutoModelForCausalLM, AutoTokenizer

checkpoint = "bigscience/bloomz-1b1"

tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.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 bloom-1b1, also refer to the config.json file
• Finetuning steps: 250
• Finetuning tokens: 502 million
• Finetuning layout: 1x pipeline parallel, 1x tensor parallel, 1x data parallel
• Precision: float16

Hardware

• CPUs: AMD CPUs with 512GB memory per node
• GPUs: 64 A100 80GB GPUs with 8 GPUs per node (8 nodes) using NVLink 4 inter-gpu connects, 4 OmniPath links
• Communication: NCCL-communications network with a fully dedicated subnet

Software

• Orchestration: Megatron-DeepSpeed
• Optimizer & parallelism: DeepSpeed
• Neural networks: PyTorch (pytorch-1.11 w/ CUDA-11.5)
• FP16 if applicable: apex

Evaluation

We refer to Table 7 from our paper & bigscience/evaluation-results for zero-shot results on unseen tasks. The sidebar reports zero-shot performance of the best prompt per dataset config.

Citation

@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}
}