TorpedoSoftware/Luau-Devstral-24B-Instruct-v0.2

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 16, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

The TorpedoSoftware/Luau-Devstral-24B-Instruct-v0.2 is a 24 billion parameter instruction-tuned language model developed by Zack Williams (boatbomber) and sponsored by Torpedo Software LLC, specifically designed for state-of-the-art Luau code generation. Enhanced with Dr. GRPO reinforcement learning, it excels in producing high-quality, type-safe, and well-formatted Luau code for Roblox development. The model features a 32768-token context length and demonstrates superior performance in code formatting, linting, and conciseness on Luau programming challenges.

Loading preview...

Luau-Devstral-24B-Instruct-v0.2: Specialized Luau Code Generation

This model, developed by Zack Williams (boatbomber) and sponsored by Torpedo Software LLC, is a 24 billion parameter instruction-tuned language model specifically engineered for generating high-quality Luau code. It is an enhanced version of Luau-Devstral-24B-Instruct-v0.1, refined using Dr. GRPO (Group Relative Policy Optimization) for superior performance in Roblox development.

Key Capabilities & Performance:

  • State-of-the-art code quality: Achieves the lowest linter errors and warnings, and exceptional code formatting adherence.
  • High type safety: Ranks 2nd in strict mode typechecking compliance, favoring explicit type definitions.
  • Concise responses: Delivers direct solutions without unnecessary verbosity, making it efficient for integration.
  • Solid problem-solving: Ranks 4th overall in unit test pass rates on the LuauLeetcode benchmark, outperforming several OpenAI models.

Training & Optimization:

The model was trained using a curriculum learning approach on the TorpedoSoftware/LuauLeetcode dataset, progressing from easy to hard challenges. Optimization involved a unique reward function design incorporating correctness (unit testing), code quality (linting), type safety (Luau typechecking), and formatting (Stylua conformance).

Ideal Use Cases:

  • Generating robust and type-safe Luau code for Roblox projects.
  • Automating code formatting and linting tasks in Luau development workflows.
  • Developers seeking highly optimized and concise code solutions for Luau programming challenges.