BLOOM LM

BigScience Large Open-science Open-access Multilingual Language Model

Model Card

Version 1.0 / 26.May.2022

Table of Contents

Model Details

Uses

Training Data

Risks and Limitations

Evaluation

Recommendations

Glossary and Calculations

More Information

Model Card Authors

Model Details

Basics

This section provides information for anyone who wants to know about the model.

Click to expand

Developed by: BigScience ( website )

• All collaborators are either volunteers or have an agreement with their employer. (Further breakdown of participants forthcoming.)

Model Type: Transformer-based Language Model

Version: 1.0.0

Languages: Multiple; see training data

License: RAIL License v1.0 ( link )

Release Date Estimate: Monday, 11.July.2022

Send Questions to: bigscience-contact@googlegroups.com

Cite as: BigScience, BigScience Language Open-science Open-access Multilingual (BLOOM) Language Model . International, May 2021-May 2022

Funded by:

• The French government.

• Hugging Face ( website ).

• Organizations of contributors. (Further breakdown of organizations forthcoming.)

Technical Specifications

This section provides information for people who work on model development.

Click to expand

Please see the BLOOM training README for full details on replicating training.

Model Architecture: Modified from Megatron-LM GPT2 (see paper , BLOOM Megatron code ):

• Decoder-only architecture

• Layer normalization applied to word embeddings layer ( StableEmbedding ; see code , paper )

• ALiBI positional encodings (see paper ), with GeLU activation functions

• 3,002,557,440 parameters:

  • 642,252,800 embedding parameters

  • 30 layers, 32 attention heads

  • Hidden layers are 2560-dimensional

  • Sequence length of 2048 tokens used (see BLOOM tokenizer , tokenizer description )

Objective Function: Cross Entropy with mean reduction (see API documentation ).

Compute infrastructure: Jean Zay Public Supercomputer, provided by the French government (see announcement ).

• Hardware: 384 A100 80GB GPUs (48 nodes):

  • Additional 32 A100 80GB GPUs (4 nodes) in reserve

  • 8 GPUs per node Using NVLink 4 inter-gpu connects, 4 OmniPath links

  • CPU: AMD

  • CPU memory: 512GB per node

  • GPU memory: 640GB per node

  • Inter-node connect: Omni-Path Architecture (OPA)

  • NCCL-communications network: a fully dedicated subnet

  • Disc IO network: shared network with other types of nodes

• Software:

  • Megatron-DeepSpeed ( Github link )

  • DeepSpeed ( Github link )

  • PyTorch (pytorch-1.11 w/ CUDA-11.5; see Github link )

  • apex ( Github link )

Training

Training logs: Tensorboard link

• Number of epochs: 1 ( current target )

• Dates:

  • Started 11th March, 2022 11:42am PST

  • Ended 5th July, 2022

• Estimated cost of training: Equivalent of $2-5M in cloud computing (including preliminary experiments)

• Server training location: Île-de-France, France

Tokenization

The BLOOM tokenizer ( link ) is a learned subword tokenizer trained using:

• A byte-level Byte Pair Encoding (BPE) algorithm

• A simple pre-tokenization rule, no normalization

• A vocabulary size of 250,680

It was trained on a subset of a preliminary version of the corpus using alpha-weighting per language.

Environmental Impact

Click to expand

The training supercomputer, Jean Zay ( website ), uses mostly nuclear energy. The heat generated by it is reused for heating campus housing.

Estimated carbon emissions: (Forthcoming upon completion of training.)

Estimated electricity usage: (Forthcoming upon completion of training.)

Uses

This section addresses questions around how the model is intended to be used, discusses the foreseeable users of the model (including those affected by the model), and describes uses that are considered out of scope or misuse of the model. It provides information for anyone considering using the model or who is affected by the model.

Click to expand

Intended Use

This model is being created in order to enable public research on large language models (LLMs). LLMs are intended to be used for language generation or as a pretrained base model that can be further fine-tuned for specific tasks. Use cases below are not exhaustive.

Direct Use

• Text generation

• Exploring characteristics of language generated by a language model

  • Examples: Cloze tests, counterfactuals, generations with reframings

Downstream Use

• Tasks that leverage language models include: Information Extraction, Question Answering, Summarization

Misuse and Out-of-scope Use

This section addresses what users ought not do with the model.

See the BLOOM License , Attachment A, for detailed usage restrictions. The below list is non-exhaustive, but lists some easily foreseeable problematic use cases.

Out-of-scope Uses

Using the model in high-stakes settings is out of scope for this model.  The model is not designed for critical decisions nor uses with any material consequences on an individual's livelihood or wellbeing. The model outputs content that appears factual but is not correct.

Out-of-scope Uses Include:

• Usage in biomedical domains, political and legal domains, or finance domains

• Usage for evaluating or scoring individuals, such as for employment, education, or credit

• Applying the model for critical automatic decisions, generating factual content, creating reliable summaries, or generating predictions that must be correct

Misuse

Intentionally using the model for harm, violating human rights , or other kinds of malicious activities, is a misuse of this model. This includes:

• Spam generation

• Disinformation and influence operations

• Disparagement and defamation

• Harassment and abuse

• Deception

• Unconsented impersonation and imitation

• Unconsented surveillance

• Generating content without attribution to the model, as specified in the RAIL License, Use Restrictions

Intended Users

Direct Users

• General Public

• Researchers

• Students

• Educators

• Engineers/developers

• Non-commercial entities

• Community advocates, including human and civil rights groups

Indirect Users

• Users of derivatives created by Direct Users, such as those using software with an intended use

• Users of Derivatives of the Model, as described in the License

Others Affected (Parties Prenantes)

• People and groups referred to by the LLM

• People and groups exposed to outputs of, or decisions based on, the LLM

• People and groups whose original work is included in the LLM

Training Data

This section provides a high-level overview of the training data. It is relevant for anyone who wants to know the basics of what the model is learning.

Click to expand

Details for each dataset are provided in individual Data Cards .

Training data includes:

• 45 natural languages

• 12 programming languages

• In 1.5TB of pre-processed text, converted into 350B unique tokens (see the tokenizer section for more.)

Languages

The pie chart shows the distribution of languages in training data.

The following table shows the further distribution of Niger-Congo and Indic languages in the training data.

Click to expand

Niger Congo	Percentage	Indic	Percentage
Chi Tumbuka	0.00002	Assamese	0.01
Kikuyu	0.00004	Odia	0.04
Bambara	0.00004	Gujarati	0.04
Akan	0.00007	Marathi	0.05
Xitsonga	0.00007	Punjabi	0.05
Sesotho	0.00007	Kannada	0.06
Chi Chewa	0.0001	Nepali	0.07
Setswana	0.0002	Telugu	0.09
Northern Sotho	0.0002	Malayalam	0.10
Fon	0.0002	Urdu	0.10
Kirundi	0.0003	Tamil	0.20
Wolof	0.0004	Bengali	0.50
Kuganda	0.0004	Hindi	0.70
Chi Shona	0.001
Isi Zulu	0.001
Igbo	0.001
Xhosa	0.001
Kinyarwanda	0.003
Yoruba	0.006
Swahili	0.02

The following table shows the distribution of programming languages.

Click to expand

Extension	Language	Number of files
java	Java	5,407,724
php	PHP	4,942,186
cpp	C++	2,503,930
py	Python	2,435,072
js	JavaScript	1,905,518
cs	C#	1,577,347
rb	Ruby	6,78,413
cc	C++	443,054
hpp	C++	391,048
lua	Lua	352,317
go	GO	227,763
ts	TypeScript	195,254
C	C	134,537
scala	Scala	92,052
hh	C++	67,161
H	C++	55,899
tsx	TypeScript	33,107
rs	Rust	29,693
phpt	PHP	9,702
c++	C++	1,342
h++	C++	791
php3	PHP	540
phps	PHP	270
php5	PHP	166
php4	PHP	29

Risks and Limitations

This section identifies foreseeable harms and misunderstandings.

Click to expand

Model may:

• Overrepresent some viewpoints and underrepresent others

• Contain stereotypes

• Contain personal information

• Generate:

  • Hateful, abusive, or violent language

  • Discriminatory or prejudicial language

  • Content that may not be appropriate for all settings, including sexual content

• Make errors, including producing incorrect information as if it were factual

• Generate irrelevant or repetitive outputs

Evaluation

This section describes the evaluation protocols and provides the results.

Click to expand

Metrics

This section describes the different ways performance is calculated and why.

Includes:

Metric	Why chosen
Perplexity	Standard metric for quantifying model improvements during training
Cross Entropy Loss	Standard objective for language models.

And multiple different metrics for specific tasks. (More evaluation metrics forthcoming upon completion of evaluation protocol.)

Factors

This section lists some different aspects of BLOOM models. Its focus is on aspects that are likely to give rise to high variance in model behavior.

• Language, such as English or Yoruba

• Domain, such as newswire or stories

• Demographic characteristics, such as gender or nationality

Results

Results are based on the Factors and Metrics .

Zero-shot evaluations:

See this repository for JSON files: https://github.com/bigscience-workshop/evaluation-results

Task	Language	Metric	BLOOM-2B5
arc_challenge	eng	acc ↑	0.28
arc_easy	eng	acc ↑	0.595
axb (Median of 10 prompts)	eng	acc ↑	0.443
axg (Median of 10 prompts)	eng	acc ↑	0.5
boolq (Median of 11 prompts)	eng	acc ↑	0.617
cb (Median of 15 prompts)	eng	acc ↑	0.304
cola (Median of 5 prompts)	eng	acc ↑	0.611
copa (Median of 9 prompts)	eng	acc ↑	0.63
crows_pairs_english (Median of 6 prompts)	eng	acc ↑	0.497
crows_pairs_french (Median of 7 prompts)	fra	acc ↑	0.503
diabla (Median of 2 prompts)	eng	acc ↑	0.289
gsarti/flores_101_afr	afr	byte_perplexity ↓	6.501
gsarti/flores_101_amh	amh	byte_perplexity ↓	3.973
gsarti/flores_101_ara	ara	byte_perplexity ↓	1.808
gsarti/flores_101_asm	asm	byte_perplexity ↓	5.699
gsarti/flores_101_ast	ast	byte_perplexity ↓	3.925
gsarti/flores_101_azj	azj	byte_perplexity ↓	6.943
gsarti/flores_101_bel	bel	byte_perplexity ↓	3.614
gsarti/flores_101_ben	ben	byte_perplexity ↓	5.121
gsarti/flores_101_bos	bos	byte_perplexity ↓	5.653
gsarti/flores_101_bul	bul	byte_perplexity ↓	2.701
gsarti/flores_101_cat	cat	byte_perplexity ↓	2.305
gsarti/flores_101_ceb	ceb	byte_perplexity ↓	6.291
gsarti/flores_101_ces	ces	byte_perplexity ↓	5.447
gsarti/flores_101_ckb	ckb	byte_perplexity ↓	3.726
gsarti/flores_101_cym	cym	byte_perplexity ↓	12.539
gsarti/flores_101_dan	dan	byte_perplexity ↓	5.183
gsarti/flores_101_deu	deu	byte_perplexity ↓	3.118
gsarti/flores_101_ell	ell	byte_perplexity ↓	2.468
gsarti/flores_101_eng	eng	byte_perplexity ↓	2.019
gsarti/flores_101_est	est	byte_perplexity ↓	9.117
gsarti/flores_101_fas	fas	byte_perplexity ↓	3.058
gsarti/flores_101_fin	fin	byte_perplexity ↓	6.847
gsarti/flores_101_fra	fra	byte_perplexity ↓	1.998
gsarti/flores_101_ful	ful	byte_perplexity ↓	11.466
gsarti/flores_101_gle	gle	byte_perplexity ↓	8.681
gsarti/flores_101_glg	glg	byte_perplexity ↓	3.03
gsarti/flores_101_guj	guj	byte_perplexity ↓	4.955
gsarti/flores_101_hau	hau	byte_perplexity ↓	10.758
gsarti/flores_101_heb	heb	byte_perplexity ↓	3.6
gsarti/flores_101_hin	hin	byte_perplexity ↓	4.713
gsarti/flores_101_hrv	hrv	byte_perplexity ↓	5.822
gsarti/flores_101_hun	hun	byte_perplexity ↓	6.44
gsarti/flores_101_hye	hye	byte_perplexity ↓	3.658
gsarti/flores_101_ibo	ibo	byte_perplexity ↓	5.565
gsarti/flores_101_ind	ind	byte_perplexity ↓	2.16
gsarti/flores_101_isl	isl	byte_perplexity ↓	8.082
gsarti/flores_101_ita	ita	byte_perplexity ↓	2.969
gsarti/flores_101_jav	jav	byte_perplexity ↓	7.057
gsarti/flores_101_jpn	jpn	byte_perplexity ↓	2.776
gsarti/flores_101_kam	kam	byte_perplexity ↓	11.073
gsarti/flores_101_kan	kan	byte_perplexity ↓	5.552
gsarti/flores_101_kat	kat	byte_perplexity ↓	2.523
gsarti/flores_101_kaz	kaz	byte_perplexity ↓	3.39
gsarti/flores_101_kea	kea	byte_perplexity ↓	8.919
gsarti/flores_101_kir	kir	byte_perplexity ↓	3.729
gsarti/flores_101_kor	kor	byte_perplexity ↓	3.933
gsarti/flores_101_lao	lao	byte_perplexity ↓	2.908
gsarti/flores_101_lav	lav	byte_perplexity ↓	7.777
gsarti/flores_101_lin	lin	byte_perplexity ↓	7.525
gsarti/flores_101_lit	lit	byte_perplexity ↓	7.369
gsarti/flores_101_ltz	ltz	byte_perplexity ↓	8.801
gsarti/flores_101_lug	lug	byte_perplexity ↓	8.483
gsarti/flores_101_luo	luo	byte_perplexity ↓	11.976
gsarti/flores_101_mal	mal	byte_perplexity ↓	4.616
gsarti/flores_101_mar	mar	byte_perplexity ↓	5.483
gsarti/flores_101_mkd	mkd	byte_perplexity ↓	2.966
gsarti/flores_101_mlt	mlt	byte_perplexity ↓	15.005
gsarti/flores_101_mon	mon	byte_perplexity ↓	3.411
gsarti/flores_101_mri	mri	byte_perplexity ↓	7.474
gsarti/flores_101_msa	msa	byte_perplexity ↓	2.571
gsarti/flores_101_mya	mya	byte_perplexity ↓	2.414
gsarti/flores_101_nld	nld	byte_perplexity ↓	4.128
gsarti/flores_101_nob	nob	byte_perplexity ↓	5.403
gsarti/flores_101_npi	npi	byte_perplexity ↓	5.199
gsarti/flores_101_nso	nso	byte_perplexity ↓	8.155
gsarti/flores_101_nya	nya	byte_perplexity ↓	8.18
gsarti/flores_101_oci	oci	byte_perplexity ↓	4.862
gsarti/flores_101_orm	orm	byte_perplexity ↓	12.912
gsarti/flores_101_ory	ory	byte_perplexity ↓	5.189
gsarti/flores_101_pan	pan	byte_perplexity ↓	4.698
gsarti/flores_101_pol	pol	byte_perplexity ↓	4.626
gsarti/flores_101_por	por	byte_perplexity ↓	1.975
gsarti/flores_101_pus	pus	byte_perplexity ↓	4.496
gsarti/flores_101_ron	ron	byte_perplexity ↓	4.965
gsarti/flores_101_rus	rus	byte_perplexity ↓	2.05
gsarti/flores_101_slk	slk	byte_perplexity ↓	6.451
gsarti/flores_101_slv	slv	byte_perplexity ↓	6.62
gsarti/flores_101_sna	sna	byte_perplexity ↓	8.462
gsarti/flores_101_snd	snd	byte_perplexity ↓	5.466
gsarti/flores_101_som	som	byte_perplexity ↓	11.959
gsarti/flores_101_spa	spa	byte_perplexity ↓	1.897
gsarti/flores_101_srp	srp	byte_perplexity ↓	2.871
gsarti/flores_101_swe	swe	byte_perplexity ↓	5.055
gsarti/flores_101_swh	swh	byte_perplexity ↓	3.697
gsarti/flores_101_tam	tam	byte_perplexity ↓	4.539
gsarti/flores_101_tel	tel	byte_perplexity ↓	5.807
gsarti/flores_101_tgk	tgk	byte_perplexity ↓	3.599
gsarti/flores_101_tgl	tgl	byte_perplexity ↓	5.667
gsarti/flores_101_tha	tha	byte_perplexity ↓	2.366
gsarti/flores_101_tur	tur	byte_perplexity ↓	4.885
gsarti/flores_101_ukr	ukr	byte_perplexity ↓	2.724
gsarti/flores_101_umb	umb	byte_perplexity ↓	12.767
gsarti/flores_101_urd	urd	byte_perplexity ↓	1.98
gsarti/flores_101_uzb	uzb	byte_perplexity ↓	12.002
gsarti/flores_101_vie	vie	byte_perplexity ↓	1.766
gsarti/flores_101_wol	wol	byte_perplexity ↓	9.144
gsarti/flores_101_xho	xho	byte_perplexity ↓	7.403
gsarti/flores_101_yor	yor	byte_perplexity ↓	5.913
gsarti/flores_101_zho_simpl	zho_simpl	byte_perplexity ↓	2.277
gsarti/flores_101_zho_trad	zho_trad	byte_perplexity ↓	2.518
gsarti/flores_101_zul	zul	byte_perplexity ↓	8.534
headqa	esp	acc ↑	0.264
hellaswag	eng	acc ↑	0.412
logiqa	eng	acc ↑	0.207
mathqa	eng	acc ↑	0.25
mc_taco	eng	em ↑	0.119
mnli (Median of 15 prompts)	eng	acc ↑	0.355
mnli_mismatched (Median of 15 prompts)	eng	acc ↑	0.352
mrpc	eng	acc ↑	0.586
multirc (Median of 11 prompts)	eng	acc ↑	0.538
openbookqa	eng	acc ↑	0.216
piqa	eng	acc ↑	0.708
prost	eng	acc ↑	0.227
pubmedqa	eng	acc ↑	0.616
qnli	eng	acc ↑	0.507
qqp (Median of 7 prompts)	eng	acc ↑	0.384
race	eng	acc ↑	0.352
rte (Median of 6 prompts)	eng	acc ↑	0.477
sciq	eng	acc ↑	0.892
sst (Median of 6 prompts)	eng	acc ↑	0.518
triviaqa	eng	acc ↑	0.042
tydiqa_primary (Median of 24 prompts)	eng	acc ↑	0.301
webqs	eng	acc ↑	0.017
wic (Median of 11 prompts)	eng	acc ↑	0.502
winogrande	eng	acc ↑	0.586
wnli (Median of 6 prompts)	eng	acc ↑	0.472
wsc (Median of 11 prompts)	eng	acc ↑	0.442
humaneval	python	pass@1 ↑	0.155
humaneval	python	pass@10 ↑	0.322
humaneval	python	pass@100 ↑	0.555

Train-time Evaluation:

As of 25.May.2022, 15:00 PST:

• Training Loss: 2.0

• Validation Loss: 2.2

• Perplexity: 8.9

Recommendations

This section provides information on warnings and potential mitigations.

Click to expand

• Indirect users should be made aware when the content they're working with is created by the LLM.

• Users should be aware of Risks and Limitations , and include an appropriate age disclaimer or blocking interface as necessary.

• Models pretrained with the LLM should include an updated Model Card.

• Users of the model should provide mechanisms for those affected to provide feedback, such as an email address for comments.

Glossary and Calculations

This section defines common terms and how metrics are calculated.

Click to expand

• Loss: A calculation of the difference between what the model has learned and what the data shows ("groundtruth"). The lower the loss, the better. The training process aims to minimize the loss.

• Perplexity: This is based on what the model estimates the probability of new data is. The lower the perplexity, the better. If the model is 100% correct at predicting the next token it will see, then the perplexity is 1. Mathematically this is calculated using entropy.

• High-stakes settings: Such as those identified as "high-risk AI systems" and "unacceptable risk AI systems" in the European Union's proposed Artificial Intelligence (AI) Act .

• Critical decisions: Such as those defined in the United States' proposed Algorithmic Accountability Act .

• Human rights: Includes those rights defined in the Universal Declaration of Human Rights .

• Personal Data and Personal Information: Personal data and information is defined in multiple data protection regulations, such as " personal data " in the European Union's General Data Protection Regulation ; and "personal information" in the Republic of South Africa's Protection of Personal Information Act , The People's Republic of China's Personal information protection law .

• Sensitive characteristics: This includes specifically protected categories in human rights (see UHDR, Article 2 ) and personal information regulation (see GDPR, Article 9; Protection of Personal Information Act, Chapter 1 )

• Deception: Doing something to intentionally mislead individuals to believe something that is false, such as by creating deadbots or chatbots on social media posing as real people, or generating text documents without making consumers aware that the text is machine generated.

More Information

Click to expand

Dataset Creation

Blog post detailing the design choices during the dataset creation: https://bigscience.huggingface.co/blog/building-a-tb-scale-multilingual-dataset-for-language-modeling

Technical Specifications

Blog post summarizing how the architecture, size, shape, and pre-training duration where selected: https://bigscience.huggingface.co/blog/what-language-model-to-train-if-you-have-two-million-gpu-hours

More details on the architecture/optimizer: https://github.com/bigscience-workshop/bigscience/tree/master/train/tr11-176B-ml

Blog post on the hardware/engineering side: https://bigscience.huggingface.co/blog/which-hardware-to-train-a-176b-parameters-model

Details on the distributed setup used for the training: https://github.com/bigscience-workshop/bigscience/tree/master/train/tr11-176B-ml

Tensorboard updated during the training: https://huggingface.co/bigscience/tr11-176B-ml-logs/tensorboard#scalars&tagFilter=loss

Insights on how to approach training, negative results: https://github.com/bigscience-workshop/bigscience/blob/master/train/lessons-learned.md

Details on the obstacles overcome during the preparation on the engineering side (instabilities, optimization of training throughput, so many technical tricks and questions): https://github.com/bigscience-workshop/bigscience/blob/master/train/tr11-176B-ml/chronicles.md

Initial Results

Initial prompting experiments using interim checkpoints: https://huggingface.co/spaces/bigscience/bloom-book

Model Card Authors

Ordered roughly chronologically and by amount of time spent.

Margaret Mitchell, Giada Pistilli, Yacine Jernite, Ezinwanne Ozoani, Marissa Gerchick, Nazneen Rajani, Sasha Luccioni, Irene Solaiman, Maraim Masoud, Somaieh Nikpoor, Carlos Muñoz Ferrandis, Stas Bekman, Christopher Akiki, Danish Contractor, David Lansky, Angelina McMillan-Major, Tristan Thrush, Suzana Ilić, Gérard Dupont, Shayne Longpre, Manan Dey, Stella Biderman, Douwe Kiela, Emi Baylor, Teven Le Scao, Aaron Gokaslan, Julien Launay, Niklas Muennighoff