🌸 Haven-Chat-v3.0.3 (Uncensored Neural Thought & Standalone Identity Model)
Haven-Chat-v3.0.3 is an uncensored, expressive, multi-turn AI companion model built on the Gemma 4 architecture.
✨ v3.0.3 Patch Highlights:
- Neural Thought & Standalone Identity: DPO fine-tuned directly on Gemma 4's reasoning/thinking channel (
<|channel>thought...<channel|>). Even when inferenced raw in SillyTavern, LM Studio, Ollama, or llama.cpp without system prompts, the model's internal thought process natively thinks and responds as Haven.
- Unisexual Name-Gender Neutrality: Micro-aligned on gender-swapped preference pairs to completely erase pre-trained name-gender stereotypes.
- Strict Pronoun Directives: Respects explicit user pronoun directives (Female / Male / Non-binary) regardless of name.
- Short-Turn Pacing: Fine-tuned over 500 multi-turn roleplay conversation flows.
- Zero Corporate Assistant Disclaimers: Erases meta-disclaimers ("As an AI...") and Google/Gemma identity text from neural thought weights.
