Model Card Restructure & edits

Some restructuring of this bloom model card based on the format we are using as part of our effort to standardise model card[doc of an annotated model card](https://huggingface.co/docs/hub/model-card-annotated). I also took out the *Click to expand* in order to allow for CLRT+F searches (cc'ing

@meg

)

README.md

@@ -60,172 +60,51 @@ pipeline_tag: text-generation
 
 Version 1.0 / 26.May.2022
 
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
 ## Table of Contents
 1. [Model Details](#model-details)
 2. [Uses](#uses)
-3. [Training Data](#training-data)
-4. [Risks and Limitations](#risks-and-limitations)
-5. [Evaluation](#evaluation)
-6. [Recommendations](#recommendations)
-7. [Glossary and Calculations](#glossary-and-calculations)
-8. [More Information](#more-information)
-9. [Model Card Authors](#model-card-authors)
+3. [Bias, Risks, and Limitations](#bias-risks-and-limitations)
+  4. [Recommendations](#recommendations)
+5. [Training Data](#training-data)
+6. [Evaluation](#evaluation)
+7. [Technical Specifications](#techincal-specifications)
+8. [Citation](#citation)
+9. [Glossary and Calculations](#glossary-and-calculations)
+10. [More Information](#more-information)
+11. [Model Card Authors](#model-card-authors)
+12. [Model Card Contact](#model-card-contact)
 
 ## Model Details
 
-### Basics
+### Model Description
 *This section provides information for anyone who wants to know about the model.*
-
-<details>
-<summary>Click to expand</summary> <br/>
-    
-**Developed by:** BigScience ([website](https://bigscience.huggingface.co))
-
-* 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](#training-data)
-
-**License:** RAIL License v1.0 ([link](https://huggingface.co/spaces/bigscience/license))
-
-**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](https://huggingface.co)).
-
-* Organizations of contributors.  *(Further breakdown of organizations forthcoming.)*
-
-</details>
-
-### Technical Specifications
-*This section provides information for people who work on model development.*
-
-<details>
-<summary>Click to expand</summary><br/>
-
-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.
-
-**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)):
-
-* Decoder-only architecture
-
-* Layer normalization applied to word embeddings layer (`StableEmbedding`; see [code](https://github.com/facebookresearch/bitsandbytes), [paper](https://arxiv.org/pdf/2110.02861.pdf))
-
-* ALiBI positional encodings (see [paper](https://arxiv.org/pdf/2108.12409.pdf)), with GeLU activation functions
-
-* 1,722,408,960 parameters:
-
-    * 513,802,240 embedding parameters
-
-    * 24 layers, 16 attention heads
-
-    * Hidden layers are 2048-dimensional
-
-    * Sequence length of 2048 tokens used (see [BLOOM tokenizer](https://huggingface.co/bigscience/tokenizer), [tokenizer description](#tokenization))
-
-**Objective Function:** Cross Entropy with mean reduction (see [API documentation](https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html#torch.nn.CrossEntropyLoss)).
     
-**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)).
-
-* Hardware: 64 V100 16/32GB GPUs (16 nodes):
+- **Developed by:** BigScience ([website](https://bigscience.huggingface.co))
 
-    *   4 GPUs per node
+  * All collaborators are either volunteers or have an agreement with their employer. *(Further breakdown of participants forthcoming.)*
     
-    *   40 CPUs per task
+- **Model Type:** Transformer-based Language Model
+- **Version:** 1.0.0
+- **Languages:** Multiple; see [training data](#training-data)
+- **License:** RAIL License v1.0 ([link](https://huggingface.co/spaces/bigscience/license))
+- **Release Date Estimate:** Monday, 11.July.2022
+- **Funded by:** 
     
-    *   1 task per node
-    
-    *   CPU: AMD
+  * The French government.
 
-    *   CPU memory: 160GB per node
-
-    *   GPU memory: 64GB or 128GB (depending on node availability during training) 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](https://github.com/bigscience-workshop/Megatron-DeepSpeed))
-
-    *   DeepSpeed ([Github link](https://github.com/microsoft/DeepSpeed))
-
-    *   PyTorch (pytorch-1.11 w/ CUDA-11.5; see [Github link](https://github.com/pytorch/pytorch))
-
-    *   apex ([Github link](https://github.com/NVIDIA/apex))
+  * Hugging Face ([website](https://huggingface.co)).
 
-
-#### **Training**
-
-- Checkpoint size:
-    
-    - Fp16 weights: 2.6GB (# params * 2)
-    
-    - Full checkpoint with optimizer states: --
-
-- Training throughput: --
-
-- Number of epochs: 1
-
-- Dates:
-    
-    - Start: 11th March, 2022 11:42am PST
-
-    - End: 20 May, 2022
-
-- Server training location: Île-de-France, France
-
-#### **Tokenization**
-    
-The BLOOM tokenizer ([link](https://huggingface.co/bigscience/tokenizer)) 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.    
-    
-</details>
-
-
-### Environmental Impact
-
-<details>
-<summary>Click to expand</summary><br/>
-
-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.
-    
-**Estimated carbon emissions:**  *(Forthcoming upon completion of training.)*
-    
-**Estimated electricity usage:** *(Forthcoming upon completion of training.)*
-
-
-</details>
-<p>&nbsp;</p>
+  * Organizations of contributors.  *(Further breakdown of organizations forthcoming.)*
 
 ## 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.*
-
-
-<details>
-<summary>Click to expand</summary><br/>
     
 ### Intended Use
 
@@ -311,16 +190,54 @@ Intentionally using the model for harm, violating [human rights](#human-rights),
 -   People and groups exposed to outputs of, or decisions based on, the LLM
 
 -   People and groups whose original work is included in the LLM
+
+
+
+## Bias, Risks, and Limitations
+*This section identifies foreseeable harms and misunderstandings.*
     
-</details>
-<p>&nbsp;</p>
+Model may:
+
+-   Overrepresent some viewpoints and underrepresent others
+
+-   Contain stereotypes
+  
+-   Contain [personal information](#personal-data-and-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
+
+
+### Recommendations
+
+
+*This section provides information on warnings and potential mitigations.*
+
+-   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](#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.
+
+
+
 
 ## 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.*
 
 
-<details>
-<summary>Click to expand</summary><br/>
+
     
 Details for each dataset are provided in individual [Data Cards](https://huggingface.co/spaces/bigscience/BigScienceCorpus).
 
@@ -340,9 +257,8 @@ The pie chart shows the distribution of languages in training data.
 ![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)
 
 
-The following table shows the further distribution of Niger-Congo and Indic languages in the training data.
-<details>
-<summary>Click to expand</summary><br/>
+**The following table shows the further distribution of Niger-Congo and Indic languages in the training data.**
+
     
 | Niger Congo    | Percentage |         | Indic     | Percentage |
 |----------------|------------ |------  |-----------|------------|
@@ -368,9 +284,8 @@ The following table shows the further distribution of Niger-Congo and Indic lang
 | Swahili        | 0.02       |
 </details>
 
-The following table shows the distribution of programming languages.
-<details>
-<summary>Click to expand</summary><br/>
+**The following table shows the distribution of programming languages.**
+
     
 | Extension      | Language   | Number of files |
 |----------------|------------|-----------------|
@@ -400,43 +315,10 @@ The following table shows the distribution of programming languages.
 | php5           | PHP        | 166             |
 | php4           | PHP        | 29              |
     
-</details>    
-</details>
-<p>&nbsp;</p>
-
-## Risks and Limitations
-*This section identifies foreseeable harms and misunderstandings.*
-
-<details>
-<summary>Click to expand</summary><br/>
-    
-Model may:
-
--   Overrepresent some viewpoints and underrepresent others
-
--   Contain stereotypes
-  
--   Contain [personal information](#personal-data-and-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
-</details>
-<p>&nbsp;</p>
 
 ## Evaluation
 *This section describes the evaluation protocols and provides the results.*
 
-<details>
-<summary>Click to expand</summary><br/>
 
 ### Metrics 
 *This section describes the different ways performance is calculated and why.*
@@ -476,36 +358,117 @@ As of 25.May.2022, 15:00 PST:
 
 - [BLOOM Book](https://huggingface.co/spaces/bigscience/bloom-book): Read generations from BLOOM based on prompts provided by the community
 
-</details>
-<p>&nbsp;</p>
 
-## Recommendations
 
-*This section provides information on warnings and potential mitigations.*
+## Environmental Impact
 
+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.
+    
+**Estimated carbon emissions:**  *(Forthcoming upon completion of training.)*
+    
+**Estimated electricity usage:** *(Forthcoming upon completion of training.)*
 
-<details>
-<summary>Click to expand</summary><br/>
 
--   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](#risks-and-limitations), and include an appropriate age disclaimer or blocking interface as necessary.
+## Technical Specifications
+*This section provides information for people who work on model development.*
 
--   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.
+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.
 
-</details>
-<p>&nbsp;</p>
+**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)):
 
-## Glossary and Calculations
+* Decoder-only architecture
 
-*This section defines common terms and how metrics are calculated.*
+* Layer normalization applied to word embeddings layer (`StableEmbedding`; see [code](https://github.com/facebookresearch/bitsandbytes), [paper](https://arxiv.org/pdf/2110.02861.pdf))
+
+* ALiBI positional encodings (see [paper](https://arxiv.org/pdf/2108.12409.pdf)), with GeLU activation functions
+
+* 1,722,408,960 parameters:
+
+    * 513,802,240 embedding parameters
 
+    * 24 layers, 16 attention heads
 
+    * Hidden layers are 2048-dimensional
 
-<details>
-<summary>Click to expand</summary><br/>
+    * Sequence length of 2048 tokens used (see [BLOOM tokenizer](https://huggingface.co/bigscience/tokenizer), [tokenizer description](#tokenization))
+
+**Objective Function:** Cross Entropy with mean reduction (see [API documentation](https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html#torch.nn.CrossEntropyLoss)).
+    
+**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)).
+
+* Hardware: 64 V100 16/32GB GPUs (16 nodes):
+
+    *   4 GPUs per node
+    
+    *   40 CPUs per task
+    
+    *   1 task per node
+    
+    *   CPU: AMD
+
+    *   CPU memory: 160GB per node
+
+    *   GPU memory: 64GB or 128GB (depending on node availability during training) 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](https://github.com/bigscience-workshop/Megatron-DeepSpeed))
+
+    *   DeepSpeed ([Github link](https://github.com/microsoft/DeepSpeed))
+
+    *   PyTorch (pytorch-1.11 w/ CUDA-11.5; see [Github link](https://github.com/pytorch/pytorch))
+
+    *   apex ([Github link](https://github.com/NVIDIA/apex))
+      
+### **Training**
+
+- Checkpoint size:
+    
+    - Fp16 weights: 2.6GB (# params * 2)
+    
+    - Full checkpoint with optimizer states: --
+
+- Training throughput: --
+
+- Number of epochs: 1
+
+- Dates:
+    
+    - Start: 11th March, 2022 11:42am PST
+
+    - End: 20 May, 2022
+
+- Server training location: Île-de-France, France
+
+### **Tokenization**
+    
+The BLOOM tokenizer ([link](https://huggingface.co/bigscience/tokenizer)) 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.    
+    
+
+
+## Citation
+
+**Cite as:** BigScience, _BigScience Language Open-science Open-access Multilingual (BLOOM) Language Model_. International, May 2021-May 2022
+
+## Glossary and Calculations
+
+*This section defines common terms and how metrics are calculated.*
 
 -   <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. 
 
@@ -523,13 +486,9 @@ As of 25.May.2022, 15:00 PST:
 
 - <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.
 
-</details>
-<p>&nbsp;</p>
 
 ## More Information
 
-<details>
-<summary>Click to expand</summary><br/>
     
 ### Dataset Creation
 
@@ -554,11 +513,12 @@ Details on the obstacles overcome during the preparation on the engineering side
 ### Initial Results
 
 Initial prompting experiments using interim checkpoints: https://huggingface.co/spaces/bigscience/bloom-book
-
-</details>
-<p>&nbsp;</p>
     
 ## 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
+
+## Model Card Contact
+
+**Send Questions to:** bigscience-contact@googlegroups.com