Chat with your PDFs with Open Source OCR (Optical Character Recognition) & Mistral Lite

Hi there! Harper Carroll from Brev.dev here. In this tutorial, we go through a pre-made Jupyter Notebook to run OCR (Optical Character Recognition) on our uploaded PDFs to extract the text, and then we use Amazon's MistralLite to ask questions about those PDFs. Amazon's MistralLite is its fine-tuned version of Mistral 7B, which allows for context lengths of up to 32K tokens... this means we can fit more data into the model's prompt i.e. memory (with some lossiness as the context length is more utilized). Notebook: https://github.com/brevdev/notebooks/blob/main/ocr-pdf-analysis.ipynb More…