whywhywhywhy/security-qwen2.5-3b-coder-instruct

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:May 6, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The security-qwen2.5-3b-coder-instruct model is a 3.1 billion parameter QWen2.5-Coder-3B variant, fine-tuned by whywhywhywhy for vulnerability detection. It specializes in identifying security issues in C, C++, Java, and Python code. This model leverages the cleaned ReposVul dataset and LoRA fine-tuning to assist in security review processes.

Loading preview...

Model Overview

security-qwen2.5-3b-coder-instruct is a specialized 3.1 billion parameter model, fine-tuned from the QWen2.5-Coder-3B base model. Its primary function is to detect software vulnerabilities across multiple programming languages.

Key Capabilities

  • Vulnerability Detection: Designed to identify potential security flaws in code snippets.
  • Multi-language Support: Proficient in analyzing code written in C, C++, Java, and Python.
  • Fine-tuned for Security: Adapted using the LoRA method on a cleaned version of the ReposVul dataset, which contains 6,134 CVE entries.

Intended Use Cases

This model is intended to assist developers and security professionals in their security review processes. It can be integrated into workflows to help pinpoint security issues in codebases, acting as a valuable tool for proactive security measures.

Limitations

While effective, the model has limitations:

  • Performance may vary with multiple vulnerabilities in a single code snippet.
  • It might not identify all vulnerabilities correctly in complex scenarios.
  • Generalization is limited to the types of vulnerabilities and languages present in the ReposVul dataset, potentially performing less effectively on unrepresented types or languages.

Finetunes
1 models