TorpedoSoftware/R1-Distill-Qwen-14B-Roblox-Luau

Model Overview

TorpedoSoftware/R1-Distill-Qwen-14B-Roblox-Luau is a specialized 14.8 billion parameter language model, fine-tuned from the deepseek-ai/DeepSeek-R1-Distill-Qwen-14B base model. Its primary focus is on Roblox domain knowledge and Luau programming, achieved through training on specific datasets such as boatbomber/roblox-info-dump and boatbomber/the-luau-stack.

Key Capabilities & Features

• Roblox Development Expertise: Specifically trained to understand and generate content related to the Roblox platform.
• Luau Programming Proficiency: Enhanced capabilities for Luau code generation, analysis, and problem-solving.
• Optimized for Specific Prompts: Fine-tuned with a recommended system prompt: "You are an expert Roblox developer and Luau software engineer."
• Inference Settings: Recommended inference parameters include a temperature of 0.5-0.7 (with 0.55 suggested for best results) and a top_p of 0.95.
• Quantization Options: Available in various quantization levels (F16, Q8_O, Q6_K, Q5_K_M, Q4_K_M, Q3_K_M) using Unsloth, with Q5_K_M and Q4_K_M being recommended for a balance of performance and resource usage.

Ideal Use Cases

This model is particularly well-suited for:

• Roblox game development: Assisting with scripting, logic, and platform-specific queries.
• Luau code generation and debugging: Providing code snippets, refactoring suggestions, or identifying issues in Luau scripts.
• Educational tools: Helping new Roblox developers understand concepts and best practices.
• Content creation: Generating documentation or tutorials related to Roblox and Luau.