Who reads PDF files better? As we noted in the last OpenAI dev day (more like RAGgers doomsday, [kidding]) GPT4 now has the ability to glean information from pdf files (that you can just drag and drop) and answer questions about it. I wanted to check how well it performs…

Whatever you can do, OpenAI can do better. :D In this blog post, we delve into the announcements made during OpenAI developer day on the 6th November, 2023. Special thanks to Mathew Berman for his insightful breakdown video. Lets take a quick look at the highlights. Six Game-Changers related to GPT-4 Turbo 1. Context Length GPT4 turbo supports a…

Unlocking the Power of Ollama Infrastructure for Local Execution of Open Source Models and Interacting with PDFs Ollama is the new docker like system that allows easy interfacing with different LLMs, set up a local LLM server, fine-tune and much more. What we are going to do is simple. I…

Answering complex queries spanning multiple documents A challenge encountered in the context of Retrieval Augmented Generation, where vector embeddings are utilized, is the limitation posed when seeking information that extends across multiple documents. This limitation arises because the Language Model (LLM) is restricted by the discrete information contained within separate…

Disclaimer: This one needs 8GB GPU USER: Describe the Image. ASSISTANT: The image features a group of horses running together in a field. The horses appear to be enjoying their time together, possibly galloping or running freely in the open space. USER: How many horses are there? ASSISTANT: There are…

No more simple inquiries — let’s put the chatbots to the test with complex, data-driven questions. Case Study: Given the annual financial report of an org, can we leverage llama_chains document retrieval along with OpenAI to extract the different Corporate Social Responsibility related activities taken up by the same. Disclaimer…

Using NSQL-Llama-2–7B, to generate SQL queries from free-flowing text instruction. Goal: Given a table schema as follows: CREATE TABLE stadium ( stadium_id number, location text, ... ) CREATE TABLE singer ( singer_id number, name text, country text, ... ) CREATE TABLE concert ( concert_id number, concert_name text, ... …

Treading dangerous waters of scraping information and then fine-tuning a llama2 chatbot over it. Thanks to Xiupeng for rectifying the title. This is a super quick guide to extract company information from the web (HTML and PDF files) and then training llama2 over it to get specialized information about the…

Running llama2 on private data on cpu. Hosting it on chainlit and reply with reference… Thanks to Eduardo Alvarez for rectifying the title. This is a super quick guide on setting up a chatbot on private data on CPU. …