GarvAgnihotri/Big-G-1

TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 2, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Big-G-1 by GarvAgnihotri is a 3 billion parameter model built on Qwen2.5-Coder-3B, specifically fine-tuned for Fill-in-the-Middle (FIM) code completion tasks. This lightweight model excels at predicting missing code segments given both preceding and succeeding context, making it highly effective for editor-based code assistance. It is optimized for local deployment and efficient performance on less powerful GPUs, providing robust code completion capabilities.

Loading preview...

Overview

Big-G-1 is a 3 billion parameter language model developed by GarvAgnihotri, specifically designed for code completion. It is built upon the Qwen2.5-Coder-3B architecture and fine-tuned to perform "Fill-in-the-Middle" (FIM) tasks, where it predicts missing code segments based on both the code before and after a gap. This makes it particularly useful for enhancing code editors and autocomplete tools, offering a lightweight solution for local development environments.

Key Capabilities

  • Fill-in-the-Middle (FIM) Code Completion: Predicts code within a given context, not just sequentially.
  • Lightweight and Efficient: Designed for use without requiring powerful GPUs, making it suitable for local deployment.
  • Specialized for Code: Optimized for understanding and generating programming code.

Benchmarks

Big-G-1 was tested on 15 custom FIM tasks, achieving a 93.3% score (14/15 passed). The single failure was attributed to a token limit cutoff rather than a logical error, suggesting higher performance with increased token limits.

Good for

  • Code completion within integrated development environments (IDEs).
  • Filling in missing lines or blocks of code.
  • Serving as a lightweight, local code assistant.

Not Great for

  • General conversational AI or question answering.
  • Complex, long-form reasoning tasks.
  • Code comments or content in languages other than English.