Update README.md

README.md

@@ -134,7 +134,7 @@ thumbnail: "https://github.com/user-attachments/assets/8ae905f3-6884-45cb-bd50-0
             <img src="https://cdn-uploads.huggingface.co/production/uploads/6589d7e6586088fd2784a12c/NckjqdBE-gOPt8r0L_Apr.png" alt="Caduceus Project" style="border-radius: 10px;">
             <p><strong>Creator:</strong> <a href="https://github.com/Kquant03" target="_blank">Kquant03</a></p>
             <div>
-                <p><strong>About the Dataset:</strong> The Caduceus Project Dataset is a curated collection of scientific and medical protocols sourced from <a href="https://github.com/protocolsio/protocols" target="_blank">protocols.io</a> and converted from PDF to markdown format. This dataset aims to enhance the accessibility and usability of complex scientific and medical information by providing a structured and easily readable version of the protocols. You can find the repository for the pipeline <a href="https://github.com/Kquant03/caduceus" target="_blank">here</a>.</p>
+                <p><strong>About the Dataset:</strong> The Caduceus Project Dataset is a curated collection of scientific and medical protocols sourced from <a href="https://github.com/protocolsio/protocols" target="_blank">protocols.io</a> and converted from PDF to markdown format. This dataset aims to help models learn to read complicated PDFs by either using computer vision on the PDF file, or through processing the raw text directly. You can find the repository for the pipeline <a href="https://github.com/Kquant03/caduceus" target="_blank">here</a>.</p>
                 <p><strong>Source Data:</strong></p>
                 <ul>
                     <li>Protocols from <a href="https://github.com/protocolsio/protocols" target="_blank">protocols.io</a></li>
@@ -142,16 +142,14 @@ thumbnail: "https://github.com/user-attachments/assets/8ae905f3-6884-45cb-bd50-0
                 <p><strong>Key Features:</strong></p>
                 <ul>
                     <li>Carefully selected high-quality protocols</li>
-                    <li>Converted from PDF to markdown format for improved readability</li>
-                    <li>Structured organization with individual markdown files and a JSONL file</li>
-                    <li>Quality assessment based on word count and measurements ratio</li>
-                    <li>Density maps visualizing word counts and measurements ratios</li>
+                    <li>Base64 encodings for potential vision training</li>
+                    <li>Guaranteed quality through hand processing the resulting data</li>
                 </ul>
                 <p><strong>Dataset Structure:</strong></p>
                 <ul>
-                    <li><code>original_pdfs/</code>: Contains the original PDF files of the selected protocols.</li>
+                    <li><code>pdf_files/</code>: Contains the original PDF files of the selected protocols.</li>
                     <li><code>markdown_files/</code>: Contains the individual markdown files converted from the selected PDF files.</li>
-                    <li><code>Caduceus_Data.jsonl/</code>: Contains the JSONL files with the text versions of the PDFs and the converted markdown content of all the selected protocols.</li>
+                    <li><code>Caduceus_Data.jsonl/</code>: A JSONL file including an input field, a Base64 encoding of the PDF file, the raw text from the PDF, and the formatted markdown output.</li>
                 </ul>
                 <p><strong>License:</strong> The Caduceus Project Dataset is released under the <a href="https://creativecommons.org/licenses/by/4.0/" target="_blank">Creative Commons Attribution 4.0 International (CC BY 4.0) License</a>.</p>
                 <p><strong>Acknowledgments:</strong> We would like to express our gratitude to the contributors of <a href="https://github.com/protocolsio/protocols" target="_blank">protocols.io</a> for providing the open-source repository of scientific and medical protocols that served as the foundation for this dataset.</p>