rajivmehtapy/git-assistant-qwen2.5-coder-1.5b

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 6, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The rajivmehtapy/git-assistant-qwen2.5-coder-1.5b is an experimental 1.5 billion parameter model, fine-tuned by rajivmehtapy using GRPO on Qwen/Qwen2.5-Coder-1.5B-Instruct. It specializes in generating executable Git-operation instructions for common repository tasks like initialization, branching, committing, and merging. This model is designed to assist with small, well-scoped Git tasks, providing commands for human review or execution within a controlled environment.

Loading preview...

Git Assistant - Qwen2.5-Coder 1.5B

This model, developed by rajivmehtapy, is an experimental GRPO fine-tune of the Qwen/Qwen2.5-Coder-1.5B-Instruct base model. It is specifically designed to generate executable Git-operation instructions, making it a specialized tool for developers working with Git repositories. The model has 1.5 billion parameters and supports a maximum sequence length of 2,048 tokens during training.

Key Capabilities

  • Git Command Generation: Proposes executable Git commands for tasks such as:
    • Initializing repositories (init_commit)
    • Creating branches (branch_create)
    • Committing files and tagging releases (commit_and_tag)
    • Merging branches (merge_branch)
  • Coding Assistant Component: Intended for integration into coding assistants to suggest Git operations.

Training Details

The model was trained using LoRA + GRPO over 15 optimizer steps across 3 epochs. A unique aspect of its training involved a Git sandbox verifier that provided rewards based on response-format checks and successful execution within an isolated environment. This experimental run did not use external datasets beyond synthetic task prompts.

Limitations and Safety

Given its experimental nature and limited training (15 steps), the model's behavior may be inconsistent. Users must always review generated commands before execution, as the model can produce incorrect, incomplete, or unsafe shell commands. It does not provide security guarantees and should not be granted unrestricted shell access or credentials. No standardized benchmarks or human evaluations were conducted for this release.

Good for

  • Developers needing quick suggestions for common Git commands.
  • Integrating into tools that require automated Git instruction generation for review.
  • Experimenting with specialized, small-scale code generation tasks within a controlled environment.

Finetunes
1 models