Executespec/ganesh-go-v0.1.0
Ganesh Go v0.1.0 is a 12 billion parameter coding model developed by the ExecuteSpec R&D Lab, based on Gemma 4 12B IT. It is specifically optimized for Go code generation, code repair, and developer-assistant workflows, offering a 32768 token context length. This model excels in Go-specific coding tasks, demonstrating improved performance over its base model on Go executable heldout and MultiPL-E Go surfaces.
Loading preview...
Overview
Ganesh Go v0.1.0 is a specialized 12 billion parameter coding model developed by the ExecuteSpec R&D Lab. Built upon the Gemma 4 12B IT architecture, this model is designed to enhance Go programming tasks. It is distributed as a standalone merged checkpoint, with development now closed, making this v0.1 release the final published artifact for Ganesh Go.
Key Capabilities
- Go Code Generation: Generates Go code and assists with code completion.
- Code Repair & Debugging: Aids in identifying and fixing issues within Go code.
- Refactoring & Exploration: Supports refactoring efforts and exploring implementation alternatives.
- Developer Assistant Workflows: Integrates into coding-agent and developer-tool evaluation scenarios.
- Research & Experimentation: Suitable for local or hosted inference research.
Performance Snapshot
On a retained 25-task Go executable heldout, Ganesh Go v0.1.0 scored 19/25, an improvement over the base model's 17/25. For the 154-task MultiPL-E Go surface, it achieved a score of 120, slightly surpassing the base model's 119. These evaluations highlight its specialized performance in Go-specific coding tasks.
Limitations
As an early coding-model release, Ganesh Go v0.1.0 may produce incorrect, incomplete, or insecure code. Users must review and test all generated code. Performance can vary based on prompt format, runtime, precision, and sampling settings.