Model Overview

The DeepSeek-R1-Distill-Qwen-Coder-32B-Fusion-9010 is an experimental 32 billion parameter model developed by huihui-ai. It is a mixed model, combining two Qwen-based models: huihui-ai/DeepSeek-R1-Distill-Qwen-32B-abliterated (90%) and huihui-ai/Qwen2.5-Coder-32B-Instruct-abliterated (10%). This fusion aims to improve thinking capabilities specifically in programming and code-related tasks.

Key Characteristics

• Architecture: Based on the Qwen2.5 architecture.
• Parameter Count: 32 billion parameters.
• Composition: A 9:1 blend of a general-purpose Qwen-based model and a code-focused Qwen2.5-Coder-Instruct model.
• Stability: Despite being an experimental mix, the model is reported to be usable without generating gibberish.

Intended Use Cases

This model is particularly suited for:

• Code Generation: Assisting with writing and generating programming code.
• Code Understanding: Tasks requiring comprehension of code logic and structure.
• Programming Assistance: General support for developers in various coding scenarios.
• Experimental Applications: Users interested in exploring the performance of fused models for specialized tasks.