zycalice/qwen-coder-primvul-attention-0203

TEXT GENERATIONConcurrency Cost:2Model Size:32.8BQuant:FP8Ctx Length:32kPublished:Feb 9, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The zycalice/qwen-coder-primvul-attention-0203 is a 32.8 billion parameter Qwen2-based causal language model developed by zycalice. Finetuned from unsloth/Qwen2.5-Coder-32B-Instruct, it leverages Unsloth and Huggingface's TRL library for accelerated training. This model is optimized for code-related tasks, offering a substantial 131,072 token context window, making it suitable for complex programming challenges and extensive code analysis.

Loading preview...

Overview

The zycalice/qwen-coder-primvul-attention-0203 is a 32.8 billion parameter language model developed by zycalice, finetuned from the unsloth/Qwen2.5-Coder-32B-Instruct base. This model was trained with significant efficiency improvements, utilizing the Unsloth library and Huggingface's TRL library, resulting in a 2x faster training process.

Key Capabilities

  • Code-centric Performance: Inherits and enhances the code generation and understanding capabilities from its Qwen2.5-Coder base.
  • Extended Context Window: Features a large 131,072 token context length, enabling it to process and generate extensive code blocks and complex programming logic.
  • Efficient Training: Benefits from Unsloth's optimization, allowing for quicker iteration and deployment of specialized code models.

Good for

  • Advanced Code Generation: Ideal for generating large, coherent, and contextually relevant code snippets or entire functions.
  • Code Analysis and Refactoring: Its large context window makes it suitable for analyzing extensive codebases, identifying patterns, and assisting with refactoring efforts.
  • Programming Assistance: Can be used for tasks requiring deep understanding of programming languages and structures, such as debugging, documentation generation, and vulnerability detection in code.