Rostik7272/semiconductor_datasheets_JSON_extraction
This dataset consist of the test cases to test the LLMs ability to extracted structured JSON information from semiconductor products datasheet pages. The LLM is asked to extract numeric values into the JSON file to represent the graph with nodes and relationships. Graph Schema looks like this: Nodes: Attribute (id, description) Measurement (value, unit, test_conditions) Relationships: Measurement-FOR_ATTRIBUTE->Attribute In dataset there are 70 test cases: 35 for FLASH memory products and… See the full description on the dataset page: https://huggingface.co/datasets/Rostik7272/semiconductor_datasheets_JSON_extraction.
This dataset consist of the test cases to test the LLMs ability to extracted structured JSON information from semiconductor products datasheet pages. The LLM is asked to extract numeric values into the JSON file to represent the graph with nodes and relationships.
Graph Schema looks like this:
Nodes:
- Attribute (id, description)
- Measurement (value, unit, test_conditions)
Relationships:
- Measurement-FOR_ATTRIBUTE->Attribute
In dataset there are 70 test cases: 35 for FLASH memory products and 35 for transceivers.
Each test cases is represented by folder with the following files in it:
- product_data.txt - file containing info about test case: datasheet url| manufacturer name|product name|package|category(Digikey based)|page number
- image.png - screenshot of the page
- markdown.md - file with extracted from page image markdown using PyMuPDF4LLM library
- mistal_markdown.md - file with extracted from page image markdown using Mistral OCR
- result - expected JSON that was composed using SOTA LLM + human refinement