LUMAMODEL/luma-1.5b

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 4, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

LUMA 1.5B by LUMAMODEL is a 1.5 billion parameter instruction-tuned causal language model, based on Qwen2.5-1.5B-Instruct, with a 32768 token context length. It is specifically fine-tuned for Russian-language dialogue and generating complete, ready-to-run HTML/CSS/JS code from descriptions. The model excels at producing accurate and well-structured web development code, particularly with CSS, and offers natural conversational abilities in Russian.

Loading preview...

LUMA 1.5B: Russian Assistant & Code Generator

LUMA 1.5B is a 1.5 billion parameter instruction-tuned model, built upon Qwen2.5-1.5B-Instruct, specifically enhanced for Russian-language interaction and web development code generation. It features a 32768 token context length, allowing for comprehensive code outputs and extended conversations.

Key Capabilities

  • Natural Russian Dialogue: Provides human-like responses and maintains conversational context.
  • Complete HTML/CSS/JS Code Generation: Generates full, ready-to-run code snippets from natural language descriptions, without abbreviations or partial outputs.
  • Advanced CSS Proficiency: Highly trained on over 12,000 HTML-to-CSS pairs, 2,500 pure CSS samples, and 1,690 CSS completion examples, resulting in accurate, well-styled, and adaptive CSS.
  • Efficient Performance: With 1.5 billion parameters, it runs effectively on both CPU and GPU.
  • Open-Source: Available under the Apache-2.0 license.

Training Details

The model was fine-tuned using QLoRA (4-bit NF4) on a comprehensive dataset of 31,860 examples. This dataset includes 4,700 description-to-HTML/CSS/JS pairs, 12,000 HTML-to-CSS examples, and extensive Russian dialogues and code blocks. Recent updates (v1.3.0) further boosted its HTML+CSS generation capabilities with an additional 7,016 examples, focusing on full code generation and CSS completion skills. The adapter is merged into the weights, providing a complete, ready-to-use model.

Good For

  • Developers needing a Russian-speaking assistant for web development tasks.
  • Generating complete HTML, CSS, and JavaScript code from textual descriptions.
  • Applications requiring natural and context-aware dialogue in Russian.

Finetunes
1 models