Overview
Jade1.7b is a Brazilian Portuguese (PT-BR) conversational finetune of the Qwen3 1.7b model, developed by Madras1. Its primary distinction is a deeply embedded persona, meaning the model consistently speaks like "Jade" with informal phrasing, slang, and a friendly, chat-oriented tone, even without extensive system prompts. This design choice aims to make the persona live in the model's weights rather than relying solely on prompting.
Key Capabilities
- Natural Brazilian Portuguese: Trained to speak authentically in PT-BR, incorporating common abbreviations like
vc, light slang, and colloquial expressions such as tmj, mano, and tlgd. - Consistent Persona: Maintains a strong, informal, and friendly persona throughout conversations, making it suitable for character-style interactions.
- Chat-Oriented Tone: Delivers responses with a human-like, less robotic tone, mimicking casual chat applications like WhatsApp.
- Persona-Driven Training: The finetuning objective focused on internalizing the Jade style directly into the model's weights through synthetic PT-BR prompt generation, persona-driven response distillation, and supervised finetuning on conversational data.
Recommended Use Cases
- PT-BR Chat Assistants: Ideal for creating conversational agents that interact naturally in Brazilian Portuguese.
- Persona Bots: Excellent for applications requiring a consistent and distinct character voice.
- WhatsApp-style Agents: Suited for building conversational interfaces that mimic the informal and quick exchanges found on messaging platforms.
- Lightweight Entertainment/Social AI: Can be used for engaging and casual AI experiences where a friendly, informal tone is desired.
Limitations
Due to its persona-oriented finetune, Jade1.7b may sound informal in contexts requiring a neutral tone and is optimized more for persona consistency than strict formality. It is less ideal for formal writing or high-stakes legal, medical, or financial applications.