HamSter-0.2 Overview

HamSter-0.2 is a 7 billion parameter uncensored fine-tuned language model, built upon the mistralai/Mistral-7B-v0.2 architecture. Developed by PotatoOff with the assistance of ConvexAI, this model is specifically designed for roleplay, including NSFW content, and features an 8192 token context length. It utilizes the Llama2 prompt template for chat instructions and has been fine-tuned with a newer dataset to improve its capabilities.

Key Capabilities & Features

• Uncensored Roleplay: Optimized for diverse roleplay scenarios, including NSFW content.
• Base Model: Fine-tuned from mistralai/Mistral-7B-v0.2.
• Prompt Template: Employs the Llama2 prompt template with chat instructions for interaction.
• Enhanced Performance: Benefits from fine-tuning with a newer dataset.
• Context Length: Supports an 8192 token context window.

Recommended Usage

For optimal performance, users are encouraged to provide detailed character cards. The model's performance has been evaluated on the OpenLLM Leaderboard and Ayumi's LLM Role Play & ERP Ranking, with specific benchmark results available for review. Recommended inference parameters include a temperature of 0.8, top_p of 0.75, and a repetition_penalty of 1.05.