kstorm77/quick-add-qwen3-0.6b

TEXT GENERATIONConcurrency Cost:1Model Size:0.8BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 7, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The kstorm77/quick-add-qwen3-0.6b is a 0.8 billion parameter, text-only model based on the Qwen3 architecture, fine-tuned for on-device use. It specializes in converting short English or Danish text into structured JSON for two distinct tasks: capturing notes/tasks and extracting calendar events. This model achieves high accuracy in valid JSON output (99.9%) and event extraction (84.3% exact, 93.7% F1), making it suitable for applications requiring efficient, structured data generation from natural language.

Loading preview...

Model Overview

The kstorm77/quick-add-qwen3-0.6b is a compact, 0.8 billion parameter language model, a full fine-tune of Qwen/Qwen3-0.6B. It is specifically designed for on-device, text-only processing to transform short English or Danish natural language inputs into structured JSON outputs. The model supports two primary functions, differentiated by a leading tag in the user message: [capture] for general notes, tasks, and events, and [event_extract] for parsing upcoming calendar events from longer texts.

Key Capabilities

  • Dual-Task JSON Generation: Handles both general note/task capture and specific event extraction, guided by a mode tag.
  • High Accuracy: Achieves 99.9% valid JSON output, with 81.5% exact match and 89.3% item-F1 score across tasks. Event extraction specifically boasts 84.3% exact and 93.7% F1 scores.
  • Verbatim Date Handling: For event extraction, dates and times are emitted exactly as written in the source text (e.g., "next Thursday", "5/5 at 18"), requiring the host application to resolve them to absolute datetimes.
  • On-Device Deployment: Optimized for deployment as a Qwen3-0.6B.Q4_K_M.gguf file (~379 MB) for llama.cpp or Ollama, or directly using its Hugging Face weights.

Use Cases

This model is ideal for applications requiring efficient, structured data extraction from user inputs or OCR'd text, particularly in resource-constrained environments. It excels at:

  • Quick Note/Task Capture: Converting brief user inputs into categorized JSON items (tasks, events, notes) with optional time, location, and priority details.
  • Event Extraction from Text: Parsing calendar events from longer, often OCR'd, texts while preserving original date/time phrasing for host-side resolution.

Users must provide a specific system prompt and ensure max_new_tokens is set appropriately (≥ 768) to prevent truncated JSON. Long inputs for event extraction should be chunked and processed iteratively by the host application.