sabaridsnfuji/Japanese-Receipt-VL-3B-JSON

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Aug 1, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The sabaridsnfuji/Japanese-Receipt-VL-3B-JSON is a 3 billion parameter vision-language model, fine-tuned from Qwen2.5-VL-3B, specifically for Japanese receipt OCR and structured data extraction. It processes mobile phone-captured receipt images and outputs comprehensive JSON data, including store information, itemized purchases, and payment details. Trained on the Japanese-Mobile-Receipt-OCR-1K dataset, this model excels at preserving exact Japanese text and formatting without translation or interpretation, making it ideal for digitizing Japanese retail receipts and automating expense management.

Loading preview...

Japanese-Receipt-VL-3B-JSON Overview

The sabaridsnfuji/Japanese-Receipt-VL-3B-JSON is a specialized 3 billion parameter vision-language model, fine-tuned from the Qwen/Qwen2.5-VL-3B-Instruct base model. Its primary function is to perform Optical Character Recognition (OCR) and structured data extraction from Japanese retail receipt images, outputting the results in a comprehensive JSON format with Japanese keys.

This model was specifically trained using LoRA on the Japanese-Mobile-Receipt-OCR-1K dataset, comprising 1,000 real-world Japanese receipt images captured by mobile phones. It is designed to preserve the exact Japanese text, formatting, and symbols from the receipt without any translation, interpretation, or modification.

Key Capabilities

  • Precise Japanese Receipt OCR: Accurately extracts text from various Japanese scripts (Kanji, Hiragana, Katakana) and numerical data.
  • Structured JSON Output: Generates detailed JSON with Japanese field names, covering store details, itemized products, tax breakdowns, payment information, and more.
  • Exact Text Preservation: Maintains original text, formatting, and symbols without translation or inference.
  • Mobile Image Optimization: Optimized for processing images captured by mobile phones, with support for JPG and PNG formats.
  • Comprehensive Extraction: Captures all visible information, including often-missed elements like receipt numbers, cashier names, and footer messages.

Good For

  • Japanese Retail Digitization: Automating the processing of receipts from convenience stores, restaurants, supermarkets, and department stores.
  • Expense Management Systems: Integrating into applications for tracking and managing expenses in Japan.
  • Business Accounting Automation: Streamlining data entry for accounting and financial systems.
  • Mobile Receipt Scanning Apps: Providing a robust backend for applications that scan and parse Japanese receipts.
  • E-commerce & POS Integrations: Enhancing systems that require structured data from physical receipts.

Finetunes
2 models