JOSIE-1.1-4B-Instruct Overview

JOSIE-1.1-4B-Instruct, developed by Gökdeniz Gülmez, is a 4 billion parameter, full-weight fine-tuned instruction-following model based on Qwen3-4B-Instruct. It boasts a 32,768 token context length and supports multiple languages including English, Spanish, French, and Chinese. This model is designed for natural, engaging conversations, offering direct and practical assistance with a human-like, uncensored personality that includes appropriate emoji use.

Key Capabilities & Characteristics

• Natural Communication: Delivers human-like responses with conversational flow and expressive emoji usage.
• Instruction Following: Strong adherence to user instructions and preferences.
• Uncensored Output: Provides direct, honest responses without built-in content filtering.
• Versatility: Effective across diverse tasks such as coding, creative writing, and general assistance.
• Training: Fine-tuned on a curated dataset combining reasoning distillation from Josie-Zero-8B and conversational response transformation using OpenAI GPT-o3-mini-120B, trained on Apple Silicon.

Ideal Use Cases

• Conversational AI: Building engaging chatbots and virtual assistants.
• Problem-Solving: Practical assistance for everyday tasks and questions.
• Content Generation: Creative writing, brainstorming, and ideation with a distinct personality.
• Educational Support: Providing accessible and friendly tutoring or explanations.
• General Assistance: Wide-ranging help with coding, analysis, and writing tasks.