ganesh-go-v0.1.0

Ganesh Go v0.1.0

Ganesh Go v0.1.0 is a 12B coding model focused on Go code generation, code repair and developer-assistant workflows. It is distributed as a standalone merged checkpoint based on Gemma 4 12B IT. It is developed and released by the ExecuteSpec R&D Lab.

Project status (2026-08-27): supported, frozen research release. A later unified v0.2 experiment failed its executable specialist-preservation gate and was not released. Model development is closed; this v0.1 specialist remains the final published Ganesh Go artifact.

Intended use

• Go code generation and completion
• Code repair and debugging assistance
• Refactoring and implementation exploration
• Coding-agent and developer-tool evaluation
• Research and experimentation with local or hosted inference

Model format

• Architecture family: Gemma 4 12B IT
• Weight formats: merged BF16 SafeTensors, GGUF Q8_0 and GGUF Q4_K_M
• Interface: text generation
• Version: v0.1.0

This repository contains a complete standalone model; no additional weight package is required for inference.

Usage

Use the tokenizer and processor files included in this repository. Runtime examples will be added after clean-environment compatibility validation for Transformers, vLLM and Unsloth.

For local llama.cpp inference, select a file from gguf/. Q8_0 prioritizes fidelity; Q4_K_M reduces memory and storage requirements.

Evaluation snapshot

On the retained 25-task Go executable heldout, the base scored 17/25 and this specialist scored 19/25, with no base-solved task lost. On the reported 154-task MultiPL-E Go surface, the base scored 119 and this specialist scored 120.

These are narrow, versioned evaluation surfaces, not a claim of general coding superiority. Scores can move with prompt, extraction, runtime, and task mix.

Limitations

This is an early coding-model release intended for evaluation. It may produce incorrect, incomplete, insecure or inefficient code; invent APIs; mishandle edge cases; or fail to follow repository-specific conventions. Generated code must be reviewed and tested before use. Do not rely on the model for security-critical, safety-critical or compliance-sensitive decisions without independent verification.

Performance may vary with prompt format, runtime, precision, sampling settings, context length and task distribution. Results from one runtime should not be assumed to transfer unchanged to another.

Versioning

Ganesh v0.1.0 publishes separate language-focused model identities. No unified Ganesh successor was released. This repository is frozen except for factual, safety, licensing, and documentation corrections.

Attribution

Ganesh Go v0.1.0 is developed and released by the ExecuteSpec R&D Lab and is based on Gemma 4 12B IT. Use of this model remains subject to the applicable Gemma Terms of Use and Gemma Prohibited Use Policy.