Resources

This is a hyper link to BLUE and ROGUE scores README.md
Model designed for interactive chat Text-to-text model
Model designed for interactive script generation Text-to-code model

Repository Overview

This repository contains all the essential components and final results of our project. The primary contents are organized in the Final Folder.

Directory Structure

.github
├── ISSUE_TEMPLATE
│   ├─── bug_report.md
│   └─── feature_request.md
└── workflows
    └── python-package-conda.yml

Final Folder
├── Chatbot Interface
│   ├── Gradio_Chatbot.ipynb
│   └── Merged_Model_Push_to_HuggingFace.ipynb
├── Conclusion
│   ├── Presentation_Advanced_Generative_Chatbot_Design.pdf
│   ├── REFERENCE.md
│   └── Report_Advanced_Generative_Chatbot_Design.pdf
├── Data Analysis and Cleansing
│   ├── Data_Analysis_and_Trivial_Modeling.ipynb
│   └── Data_Cleanup.ipynb
├── Model Selection
│   ├── assets
│   │   ├── Photos.jpeg
│   │   └── QLora _1.jpeg
│   ├── experiments
│   │   ├── flan_t5_fine_tune_experiment.ipynb
│   │   └── opt_fine_tune_experiment.ipynb
│   ├── Final_Linux_CodeLlama_2__Fine_tuning_v2.ipynb
│   ├── LLama_2_Chat_with_Report_Writeup.ipynb
│   ├── Llama_2_Chat__Fine_Tuning_V2.ipynb
│   └── Llama_2__Fine_Tuning_V2.ipynb
├── Performance and Evaluation
│   ├── Llama 2 Models BLEU and ROUGE Metrics.xlsx
│   ├── Llama_2_Chat__Fine_Tuning_with_Cleaned_Data_Dropoff_0_5.ipynb
│   ├── Llama_2_Chat__Fine_Tuning_with_Cleaned_Data_Early_Stop.ipynb
│   ├── Llama_2__Fine_Tuning_with_Cleaned_Data.ipynb
│   ├── Llama_2__Fine_Tuning_with_Uncleaned_Data.ipynb
│   ├── Performance and Evaluation.docx
│   └── README.md
└── .DS_store

.DS_Store
CODE_OF_CONDUCT.md
CONTRIBUTING.md
LICENCE.md
README.md

Folder Descriptions

• Chatbot Interface Include interface for chatbot
• Conclusion: Contains the final report summarizing the project's outcomes.
• Dataset: Includes Jupyter notebooks for initial data analysis and data cleaning.
• Model Selection: Houses various Jupyter notebooks detailing the fine-tuning processes for different models.
• Performance and Evaluation: Reserved for performance metrics and evaluation results. Before using this reposatory, for your personal convinience, run python-package-conda.yml that creates conda (anaconda) environment and set all dependencies for both Python and Java.

Question-Answering Chatbot with Arcpolar/Ubuntu_Llama_Chat_7B and SweatyCrayfish/Linux-CodeLlama2

This README outlines the steps to set up a chatbot using the Arcpolar/Ubuntu_Llama_Chat_7B model from Hugging Face for a single-turn question-answer dialogue.

Prerequisites

• Python 3.x
• pip package manager

Installation

Install the required Python package:

pip install transformers

Code Example

Below is a Python script that demonstrates a single interaction between a user and the chatbot. This example for the model's token limit of 1024.
Command downloads Ubuntu_llama_Chat_7B from huggingface

from transformers import AutoModelForCausalLM, AutoTokenizer

# Initialize the model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("Arcpolar/Ubuntu_Llama_Chat_7B") #optionally we can use SweatyCrayfish/Linux-CodeLlama2 tokenizer
model = AutoModelForCausalLM.from_pretrained("Arcpolar/Ubuntu_Llama_Chat_7B") #for scripts we recommend to use SweatyCrayfish/Linux-CodeLlama2 for script generating tasks

Generate a dialogue history where user_input is a question we want to ask our model

# Sample dialogue history
dialogue_history = ""

# Sample user input
user_input = "What is Ubuntu?" #we can ask any other questions

# Append user input to dialogue history
dialogue_history += user_input + tokenizer.eos_token

# Tokenize the dialogue history and ensure it does not exceed 1024 tokens
input_ids = tokenizer.encode(dialogue_history, return_tensors='pt')
if len(input_ids[0]) > 1024:
    print("Token limit exceeded. Please shorten the dialogue history.")

# Generate model response
output_ids = model.generate(input_ids, max_length=1024, pad_token_id=tokenizer.eos_token_id)

Display model's response

# Decode and display model response
generated_output = tokenizer.decode(output_ids[:, input_ids.shape[-1]:][0], skip_special_tokens=True)
print("Bot Response:", generated_output)

Note

The model has a token limit of 1024. Ensure your dialogue history does not exceed this limit for optimal performance.

Disclaimer

Ubuntu_Llama_Chat_7B
Ubuntu_Llama_Chat_7B is a fine-tuned model based on Llama 2 Chat 7b base model and fine-tuned on the data set Ubuntu Dialogue Corpus

Acknowledgments

Base Model: Llama-2-7b-chat-hf

• We utilized the Llama2 Chat 7b model as the base model for our project. The model was obtained from meta-llama/Llama-2-2b-chat-hf.
• Special thanks to AI at Meta for providing the model and the community around it for the support.
• License: A custom commercial license is available at: https://ai.meta.com/resources/models-and-libraries/llama-downloads/.

Fine-Tune Dataset

• The fine-tuning was performed on Ubuntu Dialogue Corpus dataset, which was crucial for achieving the results.
• The dataset is provided under Apache License, 2.0 license. We thank Ryan Lowe, Nissan Pow , Iulian V. Serban, and Joelle Pineau for making the dataset publicly available.
• Ryan Lowe, Nissan Pow, Iulian V. Serban and Joelle Pineau, "The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems", SIGDial 2015. URL: http://www.sigdial.org/workshops/conference16/proceedings/pdf/SIGDIAL40.pdf

References

Bin Lu, Isaack Karanja, & Viktor Veselov. 10/23/2023. Ubuntu Lllama 2. GitHub. https://github.com/SweatyCrayfish/Ubuntu-Lllama-2