app.py

import gradio as gr

#api = gr.Interface.load("huggingface/EleutherAI/gpt-j-6B")
api = gr.Interface.load("models/bigscience/bloom")


def complete_with_gpt(text):
    # Use the last 50 characters of the text as context
    # return text[:-50] + api(text[-50:])
    # Use the last 100 characters of the text as context
    return text[:-100] + api(text[-100:])


with gr.Blocks() as demo:
    with gr.Row():
        textbox = gr.Textbox(placeholder="Type here and press enter...", lines=14)
        with gr.Column():
            btn = gr.Button("Generate")

    btn.click(complete_with_gpt, textbox, textbox)

    with gr.Row():
        gr.Markdown("""
        
# Big Science and Huggingface create 176 Billion Parameter Transformer Large Language Model 

## Bloom Is Setting A New Record for Most Performant and Efficient AI Model for Science Ever!

Bloom stands for:
B: Big Science
L: Large Language Model
O: Open Science
O: Open Access
M: Multi Lingual Language Model

1. [Video Playlist](https://www.youtube.com/playlist?list=PLHgX2IExbFouqnsIqziThlPCX_miiDq14)
2. Summary of Important Models and Sizes:

# Model Sizes to Date

Model Name	|	Model Size (in Parameters)
----------------|---------------------------------
BigScience-tr11-176B|176 billion
GPT-3|175 billion
OpenAI's DALL-E 2.0|500 million
NVIDIA's Megatron|8.3 billion
Google's BERT|340 million
GPT-2|1.5 billion
OpenAI's GPT-1|117 million
ELMo|90 million
ULMFiT|100 million
Transformer-XL|250 million
XLNet|210 million
RoBERTa|125 million
ALBERT|12 million
DistilBERT|66 million

3. Background Information on ChatGPT, Bloom from BigScience on HuggingFace Platform, and RLHF DeepRL and One to Few Shot Learning and Generators:

# ChatGPT Datasets:
1. WebText
2. Common Crawl
3. BooksCorpus
4. English Wikipedia
5. Toronto Books Corpus
6. OpenWebText

# Comparison to BigScience Model - Big Science - How to get started

Big Science is a 176B parameter ML model trained on a set of datasets for Natural Language processing, and many other tasks that are not yet explored..  
Below is the set of the papers, models, links, and datasets around big science which promises to be the best, 
most recent large model of its kind benefitting all science pursuits.

# [Model](https://huggingface.co/bigscience/bloom)

# Papers: 
1. BLOOM: A 176B-Parameter Open-Access Multilingual Language Model [Paper](https://arxiv.org/abs/2211.05100)
2. Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism [Paper](https://arxiv.org/abs/1909.08053)
3. 8-bit Optimizers via Block-wise Quantization [Paper](https://arxiv.org/abs/2110.02861)
4. Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation [Paper](https://arxiv.org/abs/2108.12409)
5. [Paper](https://huggingface.co/models?other=doi:10.57967/hf/0003)
6. 217 Other Models optimizing use of bloom via specialization: [Paper](https://huggingface.co/models?other=bloom)

# Datasets
1. [Universal Dependencies](https://paperswithcode.com/dataset/universal-dependencies)
2. [WMT 2014](https://paperswithcode.com/dataset/wmt-2014)
3. [The Pile](https://paperswithcode.com/dataset/the-pile)
4. [HumanEval](https://paperswithcode.com/dataset/humaneval)
5. [FLORES-101](https://paperswithcode.com/dataset/flores-101)
6. [CrowS-Pairs](https://paperswithcode.com/dataset/crows-pairs)
7. [WikiLingua](https://paperswithcode.com/dataset/wikilingua)
8. [MTEB](https://paperswithcode.com/dataset/mteb)
9. [xP3](https://paperswithcode.com/dataset/xp3)
10. [DiaBLa](https://paperswithcode.com/dataset/diabla)

# Deep RL ML Strategy
1. Language Model Preparation, Human Augmented with Supervised Fine Tuning
2. Reward Model Training with Prompts Dataset Multi-Model Generate Data to Rank
3. Fine Tuning with Reinforcement Reward and Distance Distribution Regret Score
4. Proximal Policy Optimization Fine Tuning

# Variations - Preference Model Pretraining
1. Use Ranking Datasets Sentiment - Thumbs Up/Down, Distribution
2. Online Version Getting Feedback
3. OpenAI - InstructGPT - Humans generate LM Training Text
4. DeepMind - Advantage Actor Critic Sparrow, GopherCite
5. Reward Model Human Prefence Feedback
        
        """)

demo.launch()