SurfDoc AI (surfdoc:8b v1) Overview
The brady777/surfdoc-8b-v1 model is an 8 billion parameter language model developed by CloudSurf Software LLC, specifically fine-tuned from the Qwen/Qwen3-8B base model. Its primary function is to generate structurally valid SurfDoc (.surf) documents, making it a specialized tool for technical documentation and content creation workflows.
Key Capabilities
This model is designed to produce SurfDoc documents that adhere to specific structural and formatting requirements, including:
- YAML frontmatter: Generates correct
title, type, version, status, and tags fields. - Typed block directives: Incorporates
::summary, ::callout, ::data, and ::action-items with proper syntax. - Markdown content: Structures documents with Markdown headings and general content.
- Syntactic correctness: Ensures all generated SurfDoc documents are well-formed and valid.
Training and Performance
The model was fine-tuned using QLoRA on 9,241 instruction-output pairs derived from 26 CloudSurf repositories. Training was conducted on NVIDIA GB10 hardware for 5.7 hours. Evaluation on 20 diverse SurfDoc generation prompts yielded an average score of 79%, with high scores for block closure (100%), good length (100%), and no repetition (95%).
Unique Features and Usage
Unlike general-purpose LLMs, surfdoc-8b-v1 is highly specialized for a niche documentation format. Users should note that outputs may include <think> tags from the Qwen3 base model, which need to be stripped before display. This model is ideal for automating the creation of consistent and structured technical documentation within the SurfDoc ecosystem.