hwding/forge-coder-qwen-v1.21.11-merged

Warm
Public
7.6B
FP8
32768
1
License: apache-2.0
Hugging Face

Forge Coder Qwen v1.21.11 is a fine-tuned Qwen2.5-Coder-7B-Instruct model developed by hwding, specialized for generating Minecraft Forge mod code. It features a 128K context window, enabling it to handle large codebases and complex mod structures. This model is specifically trained on extensive Forge mod source code and documentation for Minecraft 1.21.11, excelling at accurate and idiomatic Java code generation for mod development.

Overview

Forge Coder Qwen v1.21.11: Specialized Minecraft Forge Mod Code Generation

Forge Coder Qwen is a fine-tuned large language model built upon Qwen2.5-Coder-7B-Instruct, specifically engineered to assist in creating Minecraft Forge mods. Developed by hwding, this model leverages a 128K context window, allowing it to process and generate code for extensive mod structures and large codebases efficiently.

Key Capabilities

  • Deep Specialization: Possesses a deep understanding of the Forge API, registry systems, and common modding patterns, specifically for Minecraft 1.21.11.
  • Comprehensive Code Generation: Excels at generating various Forge mod components, including blocks, items, entities, GUIs, networking, world generation, events, and data generation.
  • Version Alignment: Optimized for Minecraft 1.21.11 and Forge 1.21.11, ensuring generated code adheres to current standards and best practices for this version.
  • Efficient Fine-tuning: Utilizes QLoRA (4-bit quantization + LoRA) with 161.5M trainable parameters, achieving a token accuracy of 92.17% on its specialized dataset.

Use Cases

This model is ideal for:

  • Learning Forge modding patterns and best practices.
  • Rapid prototyping of mod components from natural language descriptions.
  • Code completion and intelligent suggestions within a modding context.
  • Understanding and implementing Forge API usage.
  • Working with large mod codebases due to its extended context window.

It is important to note that the model is version-specific (1.21.11) and primarily generates Java code, not supporting other languages like Kotlin or JSON resources. Generated code should always be reviewed and tested.