Madras1/Jade4b
Madras1/Jade4b is a Brazilian Portuguese conversational finetune of the Qwen3 4B model, developed by Madras1. This model is specifically designed to express a strong, persistent persona, optimized for PT-BR chat, chatbot use cases, and character-style interaction. It excels at generating colloquial language, abbreviations, slang, and a WhatsApp-like tone, making it ideal for informal and friendly conversational agents.
Loading preview...
Jade4b: Persona-First Brazilian Portuguese Chat Model
Jade4b, developed by Madras1, is a conversational text-generation finetune of the unsloth/qwen3-4b base model, specifically optimized for Brazilian Portuguese (pt-BR). Its core differentiator is its persona-first design, where the model speaks with a consistent "Jade" personality even without explicit system prompts. This is achieved by embedding the persona directly into the model's weights through supervised finetuning on conversational data.
Key Capabilities & Characteristics
- Strong, Persistent Persona: Maintains a consistent, informal, and friendly "Jade" persona across interactions.
- Natural PT-BR Conversation: Utilizes colloquial language, abbreviations (e.g.,
vc), slang (e.g.,tmj,mano), and a chat-oriented tone. - WhatsApp-like Style: Designed to mimic casual, human-like chat interactions.
- Training Approach: Finetuned on approximately 25,000 examples over 3 epochs using an Unsloth-based SFT pipeline, with persona-driven response distillation.
Recommended Use Cases
- PT-BR Chat Assistants: Ideal for creating engaging and informal chat experiences in Brazilian Portuguese.
- Persona Bots: Excellent for applications requiring a distinct and consistent character voice.
- WhatsApp-style Conversational Agents: Suited for casual, social AI interactions.
- Lightweight Entertainment/Social AI: Provides a friendly and approachable user experience.
Limitations
Due to its persona-oriented nature, Jade4b is less suitable for formal writing, highly neutral assistant behavior, or high-stakes legal, medical, or financial contexts, as it may over-index on its informal chat style.