Model Overview

ChiKoi7/GPT-5-Distill-Qwen3-4B-Instruct-Heretic is a 4 billion parameter instruction-tuned conversational LLM, derived from Jackrong/GPT-5-Distill-Qwen3-4B-Instruct. Its primary distinction is the application of the Heretic tool to significantly reduce censorship and refusal rates in both English and Chinese. The original model was fine-tuned on ShareGPT data and distilled from GPT-5 responses, aiming for high-quality, natural-sounding dialogues with low computational overhead.

Key Differentiators

• Decensored Output: Achieves a refusal rate of 3/100 in English and 10/100 in Chinese, a substantial reduction from the original model's 97/100 (en) and 84/100 (zh).
• Dual-Language Support: Maintains strong performance in both English and Chinese, with the decensoring process affecting both languages.
• Lightweight: At 4 billion parameters, it offers fast inference and low resource usage, suitable for lightweight applications.
• Conversational Fluency: Mimics GPT-5's conversational style and helpfulness due to its distillation process.

Intended Use Cases

This model is recommended for:

• Casual chat in Chinese/English without censorship.
• General knowledge explanations and reasoning guidance.
• Code suggestions and simple debugging tips.
• Writing assistance, including editing, summarizing, and rewriting.
• Role-playing conversations that require less restrictive responses.

It is not suitable for high-risk decision-making or real-time factual tasks.