Dataset, Generation Pipeline, and Model for RolePlay LLM

Talk To Listen Dataset

• The Dataset is hosted on Huggingface Hub and can be accessed here
• The dataset is generated by multiple models such as Gemini Pro (Google), GPT 3.5, Mistral Medium.
• The dataset is can be used to fine-tune the models to generate conversational AI to embody a wide range of original characters, each with a unique persona. It includes fictional characters, complete with their own backgrounds, core traits, relationships, goals, and distinct speaking styles.

Model Fiene-tuning Pipeline

• The model is hosted on Huggingface Hub and can be accessed here
• Script: TTL_RolePlay.ipynb
• Fine-tune from opensource model TheBloke/Mistral-7B-Instruct-v0.2-GPTQ using QLoRA and Supervised Learning.