Haiintel/haijava-surgeon-qwen2.5-coder-7b-sft-v2

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Jan 15, 2026Architecture:Transformer0.0K Cold

Haiintel/haijava-surgeon-qwen2.5-coder-7b-sft-v2 is a 7.6 billion parameter code language model, fine-tuned by HaiIntel Research on the Qwen2.5-Coder-7B-Instruct base, specifically for Java bug fixing. This model excels at automatically identifying and correcting method-level bugs in Java code, including null checks, bounds checking, and exception handling. It achieves a BLEU score of 0.5876 on the CodeXGLUE Code Refinement benchmark, making it a specialized tool for Java code repair. The model supports a context length of up to 32,768 tokens and is optimized for modern GPUs with BF16 precision.

Loading preview...

HaiJava-Surgeon-v2: Specialized Java Bug Fixing Model

HaiJava-Surgeon-v2 is a 7.6 billion parameter code language model developed by HaiIntel Research, built upon the Qwen2.5-Coder-7B-Instruct base. It has been meticulously fine-tuned using LoRA adapters on the CodeXGLUE Code Refinement dataset, making it highly specialized for automatically identifying and fixing bugs in Java code.

Key Capabilities

  • Java-Specific Bug Fixing: Optimized for common Java bug patterns, including null pointer prevention, bounds checking, and exception handling.
  • Method-Level Repairs: Capable of generating complete fixes for bugs within individual Java methods.
  • Consistent Performance: Achieves a stable BLEU score of 0.5876 on the CodeXGLUE benchmark, indicating good similarity to human-written fixes.
  • Production Ready: The model is fully merged, eliminating the need for adapter loading and simplifying deployment.
  • Optimized for Modern Hardware: Supports BF16 precision for efficient inference on contemporary GPUs like H100.

Good for

  • Developer Assistance: Providing real-time bug fix suggestions during Java code development.
  • Code Review Enhancement: Aiding in the identification and correction of bugs during the code review process.
  • Automated Program Repair Research: Serving as a robust baseline or component in research focused on automated code repair.
  • Educational Tool: Helping developers understand common Java bug patterns and their solutions.

While highly effective for its specialized task, users should always review and test AI-generated fixes, as the model does not perform code execution or verification. It is not recommended for fully autonomous bug fixing in security-critical applications without human oversight.