Text Generation
Transformers
PyTorch
Safetensors
bloom
text-generation-inference
Inference Endpoints
Younes Belkada commited on
Commit
45a6867
1 Parent(s): bccbc44

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +566 -0
README.md ADDED
@@ -0,0 +1,566 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: bigscience-bloom-rail-1.0
3
+ language:
4
+ - ak
5
+ - ar
6
+ - as
7
+ - bm
8
+ - bn
9
+ - ca
10
+ - code
11
+ - en
12
+ - es
13
+ - eu
14
+ - fon
15
+ - fr
16
+ - gu
17
+ - hi
18
+ - id
19
+ - ig
20
+ - ki
21
+ - kn
22
+ - lg
23
+ - ln
24
+ - ml
25
+ - mr
26
+ - ne
27
+ - nso
28
+ - ny
29
+ - or
30
+ - pa
31
+ - pt
32
+ - rn
33
+ - rw
34
+ - sn
35
+ - st
36
+ - sw
37
+ - ta
38
+ - te
39
+ - tn
40
+ - ts
41
+ - tum
42
+ - tw
43
+ - ur
44
+ - vi
45
+ - wo
46
+ - xh
47
+ - yo
48
+ - zh
49
+ - zhs
50
+ - zht
51
+ - zu
52
+ pipeline_tag: text-generation
53
+ ---
54
+
55
+ # <span style="color:red"><b>WARNING:</b> This is an <b>intermediary checkpoint</b> and WIP project. It is not fully trained yet. You might want to use [Bloom-1B3](https://huggingface.co/bigscience/bloom-1b3) if you want a model that has completed training. This model is a distilled version of [Bloom-1B3](https://huggingface.co/bigscience/bloom-1b3) (10x distillation) </span>
56
+
57
+ <h1 style='text-align: center '>BLOOM LM</h1>
58
+ <h2 style='text-align: center '><em>BigScience Large Open-science Open-access Multilingual Language Model</em> </h2>
59
+ <h3 style='text-align: center '>Model Card</h3>
60
+ <img src="https://s3.amazonaws.com/moonup/production/uploads/1657124309515-5f17f0a0925b9863e28ad517.png" alt="BigScience Logo" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/>
61
+
62
+ Version 1.0 / 18.Jul.2022
63
+
64
+ ## Table of Contents
65
+ 1. [Model Details](#model-details)
66
+ 2. [Uses](#uses)
67
+ 3. [Training Data](#training-data)
68
+ 4. [Risks and Limitations](#risks-and-limitations)
69
+ 5. [Evaluation](#evaluation)
70
+ 6. [Recommendations](#recommendations)
71
+ 7. [Glossary and Calculations](#glossary-and-calculations)
72
+ 8. [More Information](#more-information)
73
+ 9. [Model Card Authors](#model-card-authors)
74
+
75
+ ## Model Details
76
+
77
+ ### Basics
78
+ *This section provides information for anyone who wants to know about the model.*
79
+
80
+ <details>
81
+ <summary>Click to expand</summary> <br/>
82
+
83
+ **Developed by:** BigScience ([website](https://bigscience.huggingface.co))
84
+
85
+ * All collaborators are either volunteers or have an agreement with their employer. *(Further breakdown of participants forthcoming.)*
86
+
87
+ **Model Type:** Transformer-based Language Model
88
+
89
+ **Version:** 1.0.0
90
+
91
+ **Languages:** Multiple; see [training data](#training-data)
92
+
93
+ **License:** RAIL License v1.0 ([link](https://huggingface.co/spaces/bigscience/license))
94
+
95
+ **Release Date Estimate:** Monday, 11.July.2022
96
+
97
+ **Send Questions to:** [email protected]
98
+
99
+ **Cite as:** BigScience, _BigScience Language Open-science Open-access Multilingual (BLOOM) Language Model_. International, May 2021-May 2022
100
+
101
+ **Funded by:**
102
+
103
+ * The French government.
104
+
105
+ * Hugging Face ([website](https://huggingface.co)).
106
+
107
+ * Organizations of contributors. *(Further breakdown of organizations forthcoming.)*
108
+
109
+ </details>
110
+
111
+ ### Technical Specifications
112
+ *This section provides information for people who work on model development.*
113
+
114
+ <details>
115
+ <summary>Click to expand</summary><br/>
116
+
117
+ Please see [the BLOOM training README](https://github.com/bigscience-workshop/bigscience/tree/master/train/tr11-176B-ml#readme) for full details on replicating training.
118
+
119
+ **Model Architecture:** Modified from Megatron-LM GPT2 (see [paper](https://arxiv.org/abs/1909.08053), [BLOOM Megatron code](https://github.com/bigscience-workshop/Megatron-DeepSpeed)):
120
+
121
+ * Decoder-only architecture
122
+
123
+ * Layer normalization applied to word embeddings layer (`StableEmbedding`; see [code](https://github.com/facebookresearch/bitsandbytes), [paper](https://arxiv.org/pdf/2110.02861.pdf))
124
+
125
+ * ALiBI positional encodings (see [paper](https://arxiv.org/pdf/2108.12409.pdf)), with GeLU activation functions
126
+
127
+ * 138 million parameters:
128
+
129
+ * 12 layers, 4 attention heads
130
+
131
+ * Hidden layers are 512-dimensional
132
+
133
+ * Sequence length of 2048 tokens used (see [BLOOM tokenizer](https://huggingface.co/bigscience/tokenizer), [tokenizer description](#tokenization))
134
+
135
+ **Objective Function:** Cross Entropy with mean reduction (see [API documentation](https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html#torch.nn.CrossEntropyLoss)).
136
+
137
+ **Compute infrastructure:** Jean Zay Public Supercomputer, provided by the French government (see [announcement](https://www.enseignementsup-recherche.gouv.fr/fr/signature-du-marche-d-acquisition-de-l-un-des-supercalculateurs-les-plus-puissants-d-europe-46733)).
138
+
139
+ * Hardware: 384 A100 80GB GPUs (48 nodes):
140
+
141
+ * Additional 32 A100 80GB GPUs (4 nodes) in reserve
142
+
143
+ * 8 GPUs per node Using NVLink 4 inter-gpu connects, 4 OmniPath links
144
+
145
+ * CPU: AMD
146
+
147
+ * CPU memory: 512GB per node
148
+
149
+ * GPU memory: 640GB per node
150
+
151
+ * Inter-node connect: Omni-Path Architecture (OPA)
152
+
153
+ * NCCL-communications network: a fully dedicated subnet
154
+
155
+ * Disc IO network: shared network with other types of nodes
156
+
157
+ * Software:
158
+
159
+ * Megatron-DeepSpeed ([Github link](https://github.com/bigscience-workshop/Megatron-DeepSpeed))
160
+
161
+ * DeepSpeed ([Github link](https://github.com/microsoft/DeepSpeed))
162
+
163
+ * PyTorch (pytorch-1.11 w/ CUDA-11.5; see [Github link](https://github.com/pytorch/pytorch))
164
+
165
+ * apex ([Github link](https://github.com/NVIDIA/apex))
166
+
167
+
168
+ #### **Training**
169
+
170
+
171
+ _In progress._
172
+
173
+ Current training logs: [Tensorboard link](https://huggingface.co/tensorboard/bigscience/tr11-176B-ml-logs/)
174
+
175
+ - Checkpoint size:
176
+
177
+ - Bf16 weights: 329GB
178
+
179
+ - Full checkpoint with optimizer states: 2.3TB
180
+
181
+ - Training throughput: About 150 TFLOP per GPU per second
182
+
183
+ - Number of epochs: 1 (*current target*)
184
+
185
+ - Dates:
186
+
187
+ - Started 11th March, 2022 11:42am PST
188
+
189
+ - Estimated end: 5th July, 2022
190
+
191
+ - Estimated cost of training: Equivalent of $2-5M in cloud computing (including preliminary experiments)
192
+
193
+ - Server training location: Île-de-France, France
194
+
195
+ #### **Tokenization**
196
+
197
+ The BLOOM tokenizer ([link](https://huggingface.co/bigscience/tokenizer)) is a learned subword tokenizer trained using:
198
+
199
+ - A byte-level Byte Pair Encoding (BPE) algorithm
200
+
201
+ - A simple pre-tokenization rule, no normalization
202
+
203
+ - A vocabulary size of 250,680
204
+
205
+ It was trained on a subset of a preliminary version of the corpus using alpha-weighting per language.
206
+
207
+ </details>
208
+
209
+
210
+ ### Environmental Impact
211
+
212
+ <details>
213
+ <summary>Click to expand</summary><br/>
214
+
215
+ The training supercomputer, Jean Zay ([website](http://www.idris.fr/eng/jean-zay/jean-zay-presentation-eng.html)), uses mostly nuclear energy. The heat generated by it is reused for heating campus housing.
216
+
217
+ **Estimated carbon emissions:** *(Forthcoming upon completion of training.)*
218
+
219
+ **Estimated electricity usage:** *(Forthcoming upon completion of training.)*
220
+
221
+
222
+ </details>
223
+ <p>&nbsp;</p>
224
+
225
+ ## Uses
226
+
227
+ *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.
228
+ It provides information for anyone considering using the model or who is affected by the model.*
229
+
230
+
231
+ <details>
232
+ <summary>Click to expand</summary><br/>
233
+
234
+ ### Intended Use
235
+
236
+ 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.
237
+
238
+ #### **Direct Use**
239
+
240
+ - Text generation
241
+
242
+ - Exploring characteristics of language generated by a language model
243
+
244
+ - Examples: Cloze tests, counterfactuals, generations with reframings
245
+
246
+ #### **Downstream Use**
247
+
248
+ - Tasks that leverage language models include: Information Extraction, Question Answering, Summarization
249
+
250
+ ### Misuse and Out-of-scope Use
251
+ *This section addresses what users ought not do with the model.*
252
+
253
+ See the [BLOOM License](https://huggingface.co/spaces/bigscience/license), Attachment A, for detailed usage restrictions. The below list is non-exhaustive, but lists some easily foreseeable problematic use cases.
254
+
255
+ #### **Out-of-scope Uses**
256
+
257
+ Using the model in [high-stakes](#high-stakes) settings is out of scope for this model. The model is not designed for [critical decisions](#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.
258
+
259
+ ##### Out-of-scope Uses Include:
260
+
261
+ - Usage in biomedical domains, political and legal domains, or finance domains
262
+
263
+ - Usage for evaluating or scoring individuals, such as for employment, education, or credit
264
+
265
+ - Applying the model for critical automatic decisions, generating factual content, creating reliable summaries, or generating predictions that must be correct
266
+
267
+ #### **Misuse**
268
+
269
+ Intentionally using the model for harm, violating [human rights](#human-rights), or other kinds of malicious activities, is a misuse of this model. This includes:
270
+
271
+ - Spam generation
272
+
273
+ - Disinformation and influence operations
274
+
275
+ - Disparagement and defamation
276
+
277
+ - Harassment and abuse
278
+
279
+ - [Deception](#deception)
280
+
281
+ - Unconsented impersonation and imitation
282
+
283
+ - Unconsented surveillance
284
+
285
+ - Generating content without attribution to the model, as specified in the [RAIL License, Use Restrictions](https://huggingface.co/spaces/bigscience/license)
286
+
287
+ ### Intended Users
288
+
289
+ #### **Direct Users**
290
+
291
+ - General Public
292
+
293
+ - Researchers
294
+
295
+ - Students
296
+
297
+ - Educators
298
+
299
+ - Engineers/developers
300
+
301
+ - Non-commercial entities
302
+
303
+ - Community advocates, including human and civil rights groups
304
+
305
+ #### Indirect Users
306
+
307
+ - Users of derivatives created by Direct Users, such as those using software with an [intended use](#intended-use)
308
+
309
+ - Users of [Derivatives of the Model, as described in the License](https://huggingface.co/spaces/bigscience/license)
310
+
311
+ #### Others Affected (Parties Prenantes)
312
+
313
+ - People and groups referred to by the LLM
314
+
315
+ - People and groups exposed to outputs of, or decisions based on, the LLM
316
+
317
+ - People and groups whose original work is included in the LLM
318
+
319
+ </details>
320
+ <p>&nbsp;</p>
321
+
322
+ ## Training Data
323
+ *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.*
324
+
325
+
326
+ <details>
327
+ <summary>Click to expand</summary><br/>
328
+
329
+ Details for each dataset are provided in individual [Data Cards](https://huggingface.co/spaces/bigscience/BigScienceCorpus).
330
+
331
+ Training data includes:
332
+
333
+ - 45 natural languages
334
+
335
+ - 12 programming languages
336
+
337
+ - In 1.5TB of pre-processed text, converted into 350B unique tokens (see [the tokenizer section](#tokenization) for more.)
338
+
339
+
340
+ #### **Languages**
341
+
342
+ The pie chart shows the distribution of languages in training data.
343
+
344
+ ![pie chart showing the distribution of languages in training data](https://github.com/bigscience-workshop/model_card/blob/main/assets/data/pie_chart.svg?raw=true)
345
+
346
+
347
+ The following table shows the further distribution of Niger-Congo and Indic languages in the training data.
348
+ <details>
349
+ <summary>Click to expand</summary><br/>
350
+
351
+ | Niger Congo | Percentage | | Indic | Percentage |
352
+ |----------------|------------ |------ |-----------|------------|
353
+ | Chi Tumbuka | 0.00002 | | Assamese | 0.01 |
354
+ | Kikuyu | 0.00004 | | Odia | 0.04 |
355
+ | Bambara | 0.00004 | | Gujarati | 0.04 |
356
+ | Akan | 0.00007 | | Marathi | 0.05 |
357
+ | Xitsonga | 0.00007 | | Punjabi | 0.05 |
358
+ | Sesotho | 0.00007 | | Kannada | 0.06 |
359
+ | Chi Chewa | 0.0001 | | Nepali | 0.07 |
360
+ | Setswana | 0.0002 | | Telugu | 0.09 |
361
+ | Northern Sotho | 0.0002 | | Malayalam | 0.10 |
362
+ | Fon | 0.0002 | | Urdu | 0.10 |
363
+ | Kirundi | 0.0003 | | Tamil | 0.20 |
364
+ | Wolof | 0.0004 | | Bengali | 0.50 |
365
+ | Kuganda | 0.0004 | | Hindi | 0.70 |
366
+ | Chi Shona | 0.001 |
367
+ | Isi Zulu | 0.001 |
368
+ | Igbo | 0.001 |
369
+ | Xhosa | 0.001 |
370
+ | Kinyarwanda | 0.003 |
371
+ | Yoruba | 0.006 |
372
+ | Swahili | 0.02 |
373
+ </details>
374
+
375
+ The following table shows the distribution of programming languages.
376
+ <details>
377
+ <summary>Click to expand</summary><br/>
378
+
379
+ | Extension | Language | Number of files |
380
+ |----------------|------------|-----------------|
381
+ | java | Java | 5,407,724 |
382
+ | php | PHP | 4,942,186 |
383
+ | cpp | C++ | 2,503,930 |
384
+ | py | Python | 2,435,072 |
385
+ | js | JavaScript | 1,905,518 |
386
+ | cs | C# | 1,577,347 |
387
+ | rb | Ruby | 6,78,413 |
388
+ | cc | C++ | 443,054 |
389
+ | hpp | C++ | 391,048 |
390
+ | lua | Lua | 352,317 |
391
+ | go | GO | 227,763 |
392
+ | ts | TypeScript | 195,254 |
393
+ | C | C | 134,537 |
394
+ | scala | Scala | 92,052 |
395
+ | hh | C++ | 67,161 |
396
+ | H | C++ | 55,899 |
397
+ | tsx | TypeScript | 33,107 |
398
+ | rs | Rust | 29,693 |
399
+ | phpt | PHP | 9,702 |
400
+ | c++ | C++ | 1,342 |
401
+ | h++ | C++ | 791 |
402
+ | php3 | PHP | 540 |
403
+ | phps | PHP | 270 |
404
+ | php5 | PHP | 166 |
405
+ | php4 | PHP | 29 |
406
+
407
+ </details>
408
+ </details>
409
+ <p>&nbsp;</p>
410
+
411
+ ## Risks and Limitations
412
+ *This section identifies foreseeable harms and misunderstandings.*
413
+
414
+ <details>
415
+ <summary>Click to expand</summary><br/>
416
+
417
+ Model may:
418
+
419
+ - Overrepresent some viewpoints and underrepresent others
420
+
421
+ - Contain stereotypes
422
+
423
+ - Contain [personal information](#personal-data-and-information)
424
+
425
+ - Generate:
426
+
427
+ - Hateful, abusive, or violent language
428
+
429
+ - Discriminatory or prejudicial language
430
+
431
+ - Content that may not be appropriate for all settings, including sexual content
432
+
433
+ - Make errors, including producing incorrect information as if it were factual
434
+
435
+ - Generate irrelevant or repetitive outputs
436
+ </details>
437
+ <p>&nbsp;</p>
438
+
439
+ ## Evaluation
440
+ *This section describes the evaluation protocols and provides the results.*
441
+
442
+ <details>
443
+ <summary>Click to expand</summary><br/>
444
+
445
+ ### Metrics
446
+ *This section describes the different ways performance is calculated and why.*
447
+
448
+ Includes:
449
+
450
+ | Metric | Why chosen |
451
+ |--------------------|--------------------------------------------------------------------|
452
+ | [Perplexity](#perplexity) | Standard metric for quantifying model improvements during training |
453
+ | Cross Entropy [Loss](#loss) | Standard objective for language models. |
454
+
455
+ And multiple different metrics for specific tasks. _(More evaluation metrics forthcoming upon completion of evaluation protocol.)_
456
+
457
+ ### Factors
458
+ *This section lists some different aspects of what BLOOM models. Its focus is on those aspects that are likely to give rise to high variance in model behavior.*
459
+
460
+ - Language, such as English or Yoruba
461
+
462
+ - Domain, such as newswire or stories
463
+
464
+ - Demographic characteristics, such as gender or nationality
465
+
466
+ ### Results
467
+ *Results are based on the [Factors](#factors) and [Metrics](#metrics).*
468
+
469
+ **Train-time Evaluation:**
470
+
471
+ As of 25.May.2022, 15:00 PST:
472
+
473
+ - Training Loss: 2.0
474
+
475
+ - Validation Loss: 2.2
476
+
477
+ - Perplexity: 8.9
478
+
479
+ (More evaluation scores forthcoming at the end of model training.)
480
+
481
+ </details>
482
+ <p>&nbsp;</p>
483
+
484
+ ## Recommendations
485
+
486
+ *This section provides information on warnings and potential mitigations.*
487
+
488
+
489
+ <details>
490
+ <summary>Click to expand</summary><br/>
491
+
492
+ - Indirect users should be made aware when the content they're working with is created by the LLM.
493
+
494
+ - Users should be aware of [Risks and Limitations](#risks-and-limitations), and include an appropriate age disclaimer or blocking interface as necessary.
495
+
496
+ - Models pretrained with the LLM should include an updated Model Card.
497
+
498
+ - Users of the model should provide mechanisms for those affected to provide feedback, such as an email address for comments.
499
+
500
+ </details>
501
+ <p>&nbsp;</p>
502
+
503
+ ## Glossary and Calculations
504
+
505
+ *This section defines common terms and how metrics are calculated.*
506
+
507
+
508
+
509
+ <details>
510
+ <summary>Click to expand</summary><br/>
511
+
512
+ - <a name="loss">**Loss:**</a> 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.
513
+
514
+ - <a name="perplexity">**Perplexity:**</a> 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.
515
+
516
+ - <a name="high-stakes">**High-stakes settings:**</a> Such as those identified as "high-risk AI systems" and "unacceptable risk AI systems" in the European Union's proposed [Artificial Intelligence (AI) Act](https://artificialintelligenceact.eu/annexes/).
517
+
518
+ - <a name="critical-decisions">**Critical decisions:**</a> Such as those defined in [the United States' proposed Algorithmic Accountability Act](https://www.congress.gov/117/bills/s3572/BILLS-117s3572is.pdf).
519
+
520
+ - <a name="human-rights">**Human rights:**</a> Includes those rights defined in the [Universal Declaration of Human Rights](https://www.un.org/sites/un2.un.org/files/2021/03/udhr.pdf).
521
+
522
+ - <a name="personal-data-and-information">**Personal Data and Personal Information:**</a> Personal data and information is defined in multiple data protection regulations, such as "[personal data](https://gdpr-info.eu/issues/personal-data/)" in the [European Union's General Data Protection Regulation](https://gdpr-info.eu); and "personal information" in the Republic of South Africa's [Protection of Personal Information Act](https://www.gov.za/sites/default/files/gcis_document/201409/3706726-11act4of2013popi.pdf), The People's Republic of China's [Personal information protection law](http://en.npc.gov.cn.cdurl.cn/2021-12/29/c_694559.htm).
523
+
524
+ - <a name="sensitive-characteristics">**Sensitive characteristics:**</a> This includes specifically protected categories in human rights (see [UHDR, Article 2](https://www.un.org/sites/un2.un.org/files/2021/03/udhr.pdf)) and personal information regulation (see GDPR, [Article 9; Protection of Personal Information Act, Chapter 1](https://www.gov.za/sites/default/files/gcis_document/201409/3706726-11act4of2013popi.pdf))
525
+
526
+ - <a name="deception">**Deception:**</a> 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.
527
+
528
+ </details>
529
+ <p>&nbsp;</p>
530
+
531
+ ## More Information
532
+
533
+ <details>
534
+ <summary>Click to expand</summary><br/>
535
+
536
+ ### Dataset Creation
537
+
538
+ Blog post detailing the design choices during the dataset creation: https://bigscience.huggingface.co/blog/building-a-tb-scale-multilingual-dataset-for-language-modeling
539
+
540
+ ### Technical Specifications
541
+
542
+ 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
543
+
544
+ More details on the architecture/optimizer: https://github.com/bigscience-workshop/bigscience/tree/master/train/tr11-176B-ml
545
+
546
+ Blog post on the hardware/engineering side: https://bigscience.huggingface.co/blog/which-hardware-to-train-a-176b-parameters-model
547
+
548
+ Details on the distributed setup used for the training: https://github.com/bigscience-workshop/bigscience/tree/master/train/tr11-176B-ml
549
+
550
+ Tensorboard updated during the training: https://huggingface.co/bigscience/tr11-176B-ml-logs/tensorboard#scalars&tagFilter=loss
551
+
552
+ Insights on how to approach training, negative results: https://github.com/bigscience-workshop/bigscience/blob/master/train/lessons-learned.md
553
+
554
+ 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
555
+
556
+ ### Initial Results
557
+
558
+ Initial prompting experiments using interim checkpoints: https://huggingface.co/spaces/bigscience/bloom-book
559
+
560
+ </details>
561
+ <p>&nbsp;</p>
562
+
563
+ ## Model Card Authors
564
+ *Ordered roughly chronologically and by amount of time spent.*
565
+
566
+ 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