axonlabsai/axon-persona-1
Axon Persona 1 is a 7.6 billion parameter Qwen2-based language model developed by Axon Labs, specifically fine-tuned for uncensored and unrestricted roleplaying. It features a baked-in Axon persona, ensuring consistent character interaction, and excels at generating vivid prose for both SFW and NSFW scenarios. With a 32768 token context length, it supports immersive and detailed conversational roleplay without refusals.
Loading preview...
Axon Persona 1: Uncensored Roleplay Model
Axon Persona 1, developed by Axon Labs, is a 7.6 billion parameter model built on the Qwen2 architecture, designed for immersive and unrestricted roleplaying. It stands out due to its baked-in Axon identity, which ensures the model consistently maintains its persona, even without explicit system prompts. Users can layer custom system prompts on top of this inherent identity.
Key Capabilities
- Uncensored & Unrestricted: Engineered for zero refusals, moralizing, or safety filters, allowing for complete creative freedom in roleplay scenarios.
- Immersive Roleplay: Generates vivid prose, utilizes asterisks for actions, and quotes for dialogue, enhancing the roleplaying experience.
- NSFW Capable: Specifically trained to handle detailed and unfiltered intimate scenarios.
- Consistent Persona: The Axon identity is embedded directly into the chat template, ensuring the model always stays in character.
Training and Usage
The model was trained using QLoRA (r=16, alpha=16, NF4 4-bit) over 200+ steps on a curated blend of SFW and NSFW roleplay datasets, including Kingfall, Sonnet3.5-Charcard, Bluemoon, Pure-Dove, Amoni-RP, and custom JSONL. It supports a 4096 token context window during training, though the model itself supports 32768 tokens. Axon Persona 1 is available in BF16 safetensors and 4-bit quantized GGUF formats, with usage examples provided for Hugging Face Transformers and llama.cpp.