c7s89r/HazLLM-1.5B

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 7, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

HazLLM-1.5B by c7s89r is a 1.5 billion parameter language model, built on Qwen2.5-Coder-1.5B-Instruct, specifically fine-tuned for Roblox Luau and modern web development (HTML, CSS, JavaScript). It features a 32K context length and is engineered to deliver concise, expert-level code by eliminating repetition degeneration common in smaller models. This model excels at generating accurate and non-repetitive code for game development on Roblox and standard web applications.

Loading preview...

HazLLM-1.5B: Specialized Code Generation for Luau and Web Development

HazLLM-1.5B is a highly specialized 1.5 billion parameter language model developed by c7s89r, explicitly engineered for Roblox Luau development and modern web development (HTML, CSS, JS). Built upon unsloth/Qwen2.5-Coder-1.5B-Instruct, this model addresses the common repetition degeneration bug in smaller models by delivering concise, expert-level code without endless looping.

Key Capabilities

  • Roblox Luau Expert: Deep understanding of modern Luau paradigms, including TweenService, ScreenGui, DataStoreService, and strict type-checking.
  • Web Developer: Proficient in writing semantic HTML5, CSS3 (Flexbox/Grid), and vanilla ES6+ JavaScript.
  • Strictly Concise: Trained with custom system prompts and data to ensure the AI stops generating once the correct answer is provided, preventing repetition.

Training Methodology

The model was fine-tuned locally on a single NVIDIA RTX 4060 using 6,636 highly-curated instruction pairs. The training data specifically targeted modern Luau and web development practices, alongside dedicated "anti-repetition" conversations. This approach, combined with a repetition penalty during inference, ensures high-quality, non-repetitive code output. The training achieved a final loss of 0.5201, demonstrating efficient convergence without overfitting.

Good For

  • Developers requiring specialized code generation for Roblox Luau projects.
  • Generating modern, semantic HTML, CSS, and JavaScript code snippets.
  • Use cases where concise, non-repetitive code output from a smaller model is critical.

Finetunes
1 models