OddTheGreat/Circuitry_24B_V.3
OddTheGreat/Circuitry_24B_V.3 is a 24 billion parameter merged language model, building upon Circuitry v.2 with improved performance in roleplay, ERP, and assistant tasks. It features enhanced dialogue generation, consistent instruction following, and better long-context memory up to 12K tokens, while addressing self-censorship issues present in its predecessor. This model is optimized for detailed narrative generation and handling multiple characters in conversational scenarios.
Loading preview...
Circuitry_24B_V.3 Overview
Circuitry_24B_V.3 is a 24 billion parameter merged language model developed by OddTheGreat, representing an iteration on the Circuitry v.2 series. This version specifically addresses and improves upon the self-censorship tendencies observed in v.2, offering a more uninhibited response generation while maintaining adequacy.
Key Capabilities
- Enhanced Dialogue and Narration: Produces superior dialogues with a tendency towards longer messages and detailed narration, suitable for roleplay (RP) and erotic roleplay (ERP).
- Instruction Following: Demonstrates strong adherence to instructions and maintains consistent markdown formatting.
- Contextual Memory: Exhibits good memory retention for information located early in its 12K token context window, even with Q8 context quantization.
- Censorship Improvement: Significantly reduces self-censorship compared to v.2, allowing for more explicit language and NSFW content when appropriate, without being gratuitous.
- Versatile Writing Style: Capable of operating effectively across diverse settings, from grimdark to utopian, without scenario bias.
- Multi-Character Handling: Easily manages two characters in a scene and remains stable with up to five, though reply length will increase with more participants.
Good For
- Roleplay and ERP: Excels in generating engaging and detailed roleplay scenarios.
- Assistant Tasks: Performs well as a general assistant, particularly in Russian (RU) language tasks.
- Detailed Narrative Generation: Ideal for applications requiring rich, descriptive text and complex conversational flows.