royallab/Pygmalion-2-13b-SuperCOT

Overview

Pygmalion-2-13b-SuperCOT is a 13 billion parameter language model developed by royallab, resulting from a merge of two distinct models: Pygmalion 2 13b and Ausboss's Llama2 SuperCOT loras. The primary objective of this merge was to improve the intelligence and reduce conversational drift of the base Pygmalion-2 model, specifically by integrating the SuperCOT lora at a weight of 1.00.

Key Capabilities

• Enhanced Coherence: The SuperCOT lora integration is intended to make the model "smarter" and less prone to drifting off-topic during interactions.
• Roleplaying Optimization: Inherits the roleplaying capabilities of the Pygmalion-2 base model, with added stability from the SuperCOT merge.
• Instruction Formats: Supports common instruction formats such as Metharme and Alpaca, facilitating integration into various applications.

Intended Use Cases

• Interactive Roleplaying: Designed for generating engaging and coherent responses in text-based roleplaying scenarios.
• Conversational AI: Suitable for applications requiring sustained, context-aware dialogue where reducing conversational drift is crucial.

Limitations

• Bias and Risks: The model exhibits biases similar to those found in niche online roleplaying communities, in addition to the biases present in its base models.
• Not for Factual Information: It is explicitly not intended for providing factual information or advice of any kind.