bookxd/gemma-4-E2B-it-jmh-mutation-merged

VISIONConcurrent Unit Cost:1Model Size:5.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 13, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

The bookxd/gemma-4-E2B-it-jmh-mutation-merged model is a 5.1 billion parameter language model, based on the Google Gemma-4-E2B-it architecture, fine-tuned with a unique reward system. It incorporates five verifiable reward signals, including compile, runtime, SpotJMHBugs anti-pattern, RSD, and mutation, making it particularly adept at identifying and addressing planted mutants in code. With a context length of 32768 tokens, this model is specialized for code quality analysis and bug detection, especially within Java-based projects using JMH benchmarks.

Loading preview...

Model Overview

The bookxd/gemma-4-E2B-it-jmh-mutation-merged model is a specialized 5.1 billion parameter language model derived from the google/gemma-4-E2B-it base. It has been fine-tuned using a sophisticated reward system designed to enhance its capabilities in code analysis and bug detection, particularly in the context of Java Microbenchmark Harness (JMH) projects.

Key Capabilities

  • Advanced Reward System: The model's training incorporated five distinct and verifiable reward signals, each contributing 0.2 to the overall reward. These include:
    • Compile success
    • Runtime performance
    • SpotJMHBugs anti-pattern detection
    • RSD (Reward Signal for Defects)
    • Mutation testing: A unique focus on identifying and addressing planted mutants.
  • Mutation Testing Specialization: A significant aspect of its training involved a mutation corpus comprising 300 planted mutants across 99 out of 113 RL subject classes, making it highly effective at pinpointing code defects.
  • Extended Context Length: Supports a substantial context length of 32768 tokens, allowing for comprehensive analysis of larger code segments.

Good For

  • Code Quality Analysis: Excellent for evaluating and improving the quality of Java codebases.
  • Bug Detection: Particularly strong in identifying subtle bugs and anti-patterns, especially those related to JMH benchmarks.
  • Mutation Testing: Ideal for scenarios requiring automated detection and correction of code mutations.

Finetunes
1 models