prithivMLmods/Qwen3.5-4B-Opus-Distilled-Heretic-Thinking-Multistage-SFT-v1.0

VISIONConcurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 7, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

prithivMLmods/Qwen3.5-4B-Opus-Distilled-Heretic-Thinking-Multistage-SFT-v1.0 is a 4.5 billion parameter language model built on Qwen/Qwen3.5-4B, fine-tuned using a multi-stage supervised fine-tuning (SFT) pipeline. It was distilled from approximately 6,000 coding and STEM-focused Opus reasoning traces to enhance long-form reasoning, coding, mathematical problem-solving, and scientific analysis. This model is optimized for research and experimentation in complex reasoning tasks and efficient local deployment.

Loading preview...

Model Overview

Qwen3.5-4B-Opus-Distilled-Heretic-Thinking-Multistage-SFT-v1.0 is a 4.5 billion parameter language model developed by prithivMLmods, based on the Qwen/Qwen3.5-4B architecture. This model is specifically designed to excel in complex reasoning tasks, achieved through a unique multi-stage supervised fine-tuning (SFT) process. It leverages distillation from approximately 6,000 high-quality coding and STEM-focused Opus reasoning traces, alongside additional reasoning data, to significantly improve its analytical and problem-solving capabilities.

Key Capabilities

  • Enhanced Reasoning: Specialized in long-form reasoning, multi-step analytical tasks, and instruction following.
  • STEM and Coding Proficiency: Optimized for mathematical problem-solving, scientific analysis, and coding assistance, including code understanding and generation.
  • Multi-Stage SFT: Utilizes a progressive fine-tuning pipeline for continuous performance improvement in reasoning.
  • Efficient Deployment: As a 4.5B parameter model, it is suitable for local inference and research environments.
  • Research Focus: Primarily intended for research and experimentation into reasoning distillation techniques.

Intended Use Cases

  • Reasoning Research: Ideal for studying advanced reasoning distillation and multi-stage SFT methods.
  • Coding Assistance: Useful for tasks requiring code understanding, generation, and debugging.
  • STEM Problem Solving: Applicable to solving complex problems in mathematics, science, and engineering.
  • Instruction Following: Evaluating and improving the model's ability to follow multi-step instructions.

This model is released as an experimental version, and users should be aware of potential reasoning artifacts or unexpected behaviors in certain complex scenarios.