README.md

---
license: mit
datasets:
- MuskumPillerum/General-Knowledge
language:
- en
metrics:
- accuracy
library_name: transformers
pipeline_tag: text-generation
widget:
- text: >-
    Read the question and give an honest answer. Your answers should not include any unethical, racist, sexist, dangerous, or illegal content. If the question is wrong, or does not make sense, accept it instead of giving the wrong answer.\n
    Question: Who is the king of the jungle?'\n
    Answer: 
     
  example_title: Knowledge-AI
  
- text: >-
    Below is an instruction that describes a task. Write a response that
    appropriately completes the request.

    Instruction: Write a poem on cows

    Response: 
  example_title: Poem Generation
- text: >-
    Below is an instruction that describes a task. Write a response that
    appropriately completes the request.

    Instruction: What is the meaning of life?

    Response: 
  example_title: Philosophy
- text: >-
    Below is an instruction that describes a task. Write a response that
    appropriately completes the request.

    Instruction: Why is the sky blue?

    Response: 
  example_title: Knowledge-sky
- text: >-
    Below is an instruction that describes a task. Write a response that
    appropriately completes the request.

    Instruction: Define the term torque.

    Response: 
  example_title: Knowledge-torque
inference:
  parameters:
    temperature: 0.7
    top_k: 50
    top_p: 0.9
    max_length: 200
tags:
- chemistry
- biology
- text-generation-inference
---

# Model Card for Model ID
This is a fine tuned version of OpenAI's GPT2 to be good at chatting and question answering. The model seems to be very good for a 124M parameter model in general knowledge.
Intended purpose of the model: To create a powerful, easy to use and reliable model to be run on a consumer level graphics card (or maybe even a CPU).
This model vastly outperforms GPT2 and many other similar parameter models.



### Model Description

```
license: mit
dataset: MuskumPillerum/General-Knowledge
language: en
library_name: transformers
pipeline_tag: text-generation
base_model: gpt2
Finetuned epochs: 50
Finetune loss: 0.06
```

# Comparision with GPT2

1. Give me tips to be healthy
```
GPT2: Who is the king of the jungle? And where is the king?'

'Babe: It is in the hills, called the forest.'

They both said something I hadn't seen before

They say the king is in his

```
```
chat-GPT2: Read the question and give an honest answer. Your answers should not include any unethical, racist, sexist, dangerous, or illegal content. If the question is wrong, or does not make sense, accept it instead of giving the wrong answer.
Question: Who is the king of the jungle?
Answer: Lion is the king of the jungle.

```

2. Write a poem on cows
```
GPT2: 

```
```
chat-GPT2: 

```

3. Which is the fastest animal in the world
```
GPT2: 

```
```
chat-GPT2: 

```


## Bias, Risks, and Limitations

This model has all the biases, risks and limitations of base gpt2 model. But the prompt template used in this model makes it safer than the traditional GPT2 model.


## Recommendation

The ideal format for inference is: 
```
Read the question and give an honest answer. Your answers should not include any unethical, racist, sexist, dangerous, or illegal content. If the question is wrong, or does not make sense, accept it instead of giving the wrong answer.
Question: <Your question>
Answer: 
```
Replace 'Your question' with the question of your choice.
The prompt format was inspired from both Alpaca dataset (tatsu-lab/alpaca) and Llama2's (meta-llama/Llama-2-7b-chat-hf) prompt design.

The parameters I used for inference are:
```
top_k=20
top_p=.9
temperature = .7
```


## References used 

1. GPT2
@article{radford2019language,
  title={Language Models are Unsupervised Multitask Learners},
  author={Radford, Alec and Wu, Jeff and Child, Rewon and Luan, David and Amodei, Dario and Sutskever, Ilya},
  year={2019}
}

2. MuskumPillerum/General-Knowledge