Scriptease/colorhex-1b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Aug 23, 2026License:gemmaArchitecture:Transformer0.0K Featherless Exclusive Cold

Scriptease/colorhex-1b is a 1 billion parameter model, fine-tuned from Google's Gemma-3-1b-it, designed to map product color names to hex RGB codes. It excels at structured output, handling various languages including German, Spanish, Greek, Hungarian, and Turkish, even for compound color names. This model is optimized for single-call color code generation, making it suitable for applications requiring precise color mapping from textual descriptions.

Loading preview...

Model Overview

Scriptease/colorhex-1b is a 1 billion parameter model, fine-tuned using LoRA (rank 16) from google/gemma-3-1b-it. Its primary function is to accurately map product color names to their corresponding six-digit hexadecimal RGB codes. The repository includes both the merged safetensors weights and a Q8_0 GGUF export for llama.cpp compatibility.

Key Capabilities

  • Color Name to Hex Mapping: Converts product color names into precise hex RGB values.
  • Multilingual Support: Handles color names in German, Spanish, Greek, Hungarian, and Turkish, including complex compound and modifier prefixes (e.g., hellblau, dunkelgrün).
  • Structured Output: Designed for a strict chat format, returning results as a JSON object with input-value pairs.
  • Batch Processing: Optimized for batched input, processing 10 color names per call for efficiency.

Training and Performance

The model was distilled from a production color-mapping service, trained on approximately 25,000 unique product color names paired with representative hex values. Evaluation on unseen product color names shows an approximate 60% exact hex match, with most other results falling within the correct color family. The Q8_0 GGUF version maintains performance consistent with the merged weights.

Usage Guidelines

This model requires a specific chat format, including a detailed system prompt and numbered user inputs, batched 10 at a time. Deviations from this format may degrade accuracy. The model is licensed under the Gemma Terms of Use, as it is a derivative of the Gemma model.