gr33r/ux-writing-1

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 10, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

gr33r/ux-writing-1 is a 27-billion parameter language model, fine-tuned from Qwen/Qwen3.6-27B, specifically designed for reviewing and rewriting UI copy in product code. It generates compact JSON outputs (rewrite, reason, risk) for given UI strings and their code context, focusing on purposefulness, conciseness, conversational tone, clarity, and accessibility. This model excels at scanning massive codebases for UX writing improvements at a low cost, preserving product intent and critical variables while avoiding weakening safety-critical copy.

Loading preview...

Overview

gr33r/ux-writing-1 is a specialized 27B parameter model, fine-tuned from the Qwen/Qwen3.6-27B base, designed to act as an open UX writing reviewer. It processes UI strings and their code context, returning compact JSON output containing a rewrite, reason, and risk assessment. Developed for the Hugging Face "Build Small" hackathon, it aims to provide a cost-effective, private, and reusable solution for large-scale UX writing review.

Key Capabilities

  • Purposeful Rewrites: Generates concise, conversational, clear, and accessible UI copy.
  • Context-Aware: Preserves product intent, {{ variables }}, and locale-specific terms.
  • Safety-Critical Copy Protection: Trained not to weaken destructive, payment, privacy, or security messages.
  • Efficient Output: Produces lean JSON responses, significantly reducing inference token count compared to the base model.
  • Human-Validated Performance: Achieved an 83% preference rate over the base model in blinded human expert reviews across various categories like inline errors, destructive confirmations, and buttons.
  • Codebase Scanning: Includes tooling to extract UI strings from various file types (JS/TSX/Vue/Svelte/HTML/i18n) and review them against an OpenAI-compatible endpoint.

Good for

  • Automated UX Writing Review: Ideal for scanning and improving UI copy across large codebases.
  • Cost-Effective Operations: Offers a private, unlimited-reuse solution at a fraction of frontier-API costs.
  • Custom Fine-tuning: Provides a robust base for further fine-tuning to align with specific team voice and style guides, with a detailed guide and tooling for verification.
  • English Product UI Microcopy: Optimized for short-form UI content like buttons, errors, notifications, and labels. Long-form content is out of scope.

Limitations

  • Primarily English-centric and focused on product-UI microcopy.
  • Functions as a reviewer's assistant; human oversight is essential, especially for risk notes.
  • Requires code context to avoid over-specification on vague inputs.
  • The fine-tune is text-only, despite the base model's vision capabilities.

Finetunes
3 models