cgato/Thespis-Krangled-7b-v2

Overview

cgato/Thespis-Krangled-7b-v2 is a 7 billion parameter language model, distinguished by its extensive fine-tuning on a wide variety of datasets. These include general conversational datasets like Dolphin, Ultrachat, and Capybara, as well as specialized datasets such as Magiccoder-Evol-Instruct-110k for code-related instructions, and ToxicQA for handling nuanced content. The model supports a context length of 8192 tokens, making it suitable for maintaining longer conversations and processing more extensive inputs.

Key Capabilities

• Broad Instruction Following: Trained on diverse datasets like OpenOrca and Airoboros 3.1, enabling it to respond to a wide range of prompts and instructions.
• Conversational AI: Optimized for interactive chat applications, with a recommended prompt format for Username: {Input} and BotName: {Response}.
• Code-Related Tasks: Inclusion of Magiccoder-Evol-Instruct-110k in its training suggests capabilities in understanding and generating code-related text.
• Robustness: Training on datasets like ToxicQA and Yahoo Answers may contribute to its ability to handle varied and sometimes challenging conversational contexts.

Good For

• General-purpose chatbots and virtual assistants requiring versatile instruction following.
• Interactive storytelling and role-playing applications due to its conversational fine-tuning.
• Developers experimenting with models that have been trained on a highly eclectic mix of public datasets to achieve broad utility.