MayankLad31/invoice_schema

Hugging Face
TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:May 3, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

MayankLad31/invoice_schema is a 1.5 billion parameter model, fine-tuned from Qwen2.5-Coder, specifically designed for extracting structured JSON data from OCR text based on user-defined schemas. It leverages a fine-tuned Qwen3.5 0.8B variant for improved performance. The model excels at parsing invoice-like documents and converting their content into a specified JSON format, making it suitable for automated data extraction tasks.

Loading preview...

Overview

MayankLad31/invoice_schema is a specialized model focused on extracting structured JSON data from OCR (Optical Character Recognition) text, particularly from documents like invoices. The model is built upon a fine-tuned version of Qwen3.5 0.8B, with previous iterations using a 1.5 billion parameter Qwen2.5-Coder model. Its core capability is to take an image (or its OCR-extracted text) and a user-defined JSON schema, then output the extracted information conforming to that schema.

Key Capabilities

  • Schema-driven Extraction: Extracts data into a JSON format specified by the user, allowing for flexible data structuring.
  • Invoice Processing: Demonstrated capability in parsing invoice details such as dates, invoice IDs, item descriptions, quantities, unit prices, and totals.
  • Local Deployment: Provides instructions for running the model locally using llama-server and gguf files, supporting CPU-based inference.
  • OpenAI SDK Compatibility: Integrates with the OpenAI SDK for ease of use, allowing developers to interact with the local server using familiar API calls.
  • Vision Integration: Processes image inputs (e.g., JPEG) to extract text, which is then structured according to the provided schema.

Good For

  • Automated Invoice Data Entry: Ideal for businesses looking to automate the extraction of key information from invoices into structured databases.
  • Document Processing Workflows: Useful in workflows requiring the conversion of unstructured text from scanned documents into structured, machine-readable formats.
  • Custom Data Extraction: Developers needing to extract specific fields from various document types based on their own custom schemas.
  • Local-first AI Applications: Suitable for applications where data privacy or offline processing is critical, as it can run entirely on local hardware.

Finetunes
1 models