DIGITAL TWIN MODEL

This repository is dedicated to exploring, experimenting, and finetuning various pre-trained models, primarily from Huggingface, to develop a customized chatbot. Our aim is to delve into language models, enhancing their capabilities for personalized user interactions.

Chat Model Section

In the chatmodel directory, you'll find finetuning notebooks along with the associated model weights. Each model has its dedicated folder.

Included Models:

• Blenderbot (variants: 400M, 1B)
• DialoGPT (sizes: small, medium, large)
• GPT-2
• Llama2 (7B version)
• Mistral (7B version)
• QA (Question Answering) Model

Text-to-Speech (TTS) Model Section

The text-to-speech directory houses finetuning notebooks, sample voices, and a preprocessing notebook for TTS models. Detailed finetuning guidelines and deployment instructions for the XTTS model are also available.

Available TTS Models:

• MetaVoice (1B version)
• Speech T5
• XTTS
• OpenVoice

XTTS Finetuning and Deployment:

• Finetuning the XTTS model is user-friendly through a GUI, detailed in text-to-speech\finetune\XTTS_FT.ipynb.
• Post-finetuning, model weights are available for creating a new Docker image, simplifying deployment. For detailed deployment instructions, refer to the text-to-speech\deploy directory.