Raiff1982/codette-llama-3.1-8b-merged

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kTool Calling:SupportedPublished:Mar 21, 2026License:llama3.1Architecture:Transformer Cold

Raiff1982/codette-llama-3.1-8b-merged is an 8 billion parameter Llama 3.1 Instruct model with an integrated 'Orchestrator LoRA' adapter, developed by Jonathan Harrison (Raiff's Bits LLC). This model serves as the inference base for the Codette multi-perspective reasoning system, designed to enhance cognitive capabilities and naturalness. It excels in complex reasoning tasks by enabling multi-agent debate and semantic tension, achieving significant performance gains over baseline models in cognitive benchmarks. With an 8192-token context length, it is optimized for advanced analytical and problem-solving applications when paired with external perspective adapters.

Loading preview...

Codette Llama 3.1 8B Merged Orchestrator Base

This model, developed by Jonathan Harrison (Raiff's Bits LLC), is an 8 billion parameter Llama 3.1 Instruct model with the Codette Orchestrator LoRA adapter permanently merged into its base weights. It functions as the core inference engine for the Codette multi-perspective reasoning system, which aims to improve reasoning and naturalness through a convergent dynamical system approach.

Key Capabilities & Features

  • Enhanced Reasoning: Achieves a +108.8% composite score improvement over baseline in a 17-problem cognitive benchmark, demonstrating superior multi-perspective reasoning.
  • Multi-Agent System: Designed to be paired with external perspective LoRA adapters to enable a full multi-agent debate and semantic tension system.
  • High Coherence & Naturalness: Significantly improves Turing naturalness (+235%) and coherence in generated responses, resolving the depth–naturalness tradeoff.
  • Graduate-Level Science Performance: Shows a 30.8% accuracy on the GPQA graduate-level science benchmark when integrated into the full Codette system, outperforming the base model's 27.8% without orchestration.
  • Robust Architecture: Incorporates an Executive Controller, LoRA Hot-Swap for various perspectives (e.g., newton, empathy), Multi-Agent Debate, AEGIS Ethical Governance, and Cocoon Memory.
  • 8192-token Context: Provides a substantial context window for processing complex queries.

Good For

  • Developers building advanced reasoning systems requiring multi-perspective analysis.
  • Applications demanding high coherence and naturalness in generated text.
  • Research into complex cognitive architectures and AI debate systems.
  • Tasks benefiting from enhanced problem-solving capabilities beyond standard LLM performance.