undefinedRetrieval augmented generation (RAG) is a natural language processing (NLP) technique that combines the strengths of both retrieval- and generative-based artificial intelligence (AI) models. Based on the task assigned, the easiest and best suited way to extract insightful data without huge amounts of data analysis, in a short period of time, while being resistant to events as, changes in data formats, layouts, etc, was to implement a RAG based on a relatively lightweight model (Mistral 7B in our case).

The following notebook contains the code blocks required for the functioning of the CLI program that initializes a RAG at the backend and provides a chat based interactive interface to the user. The user has the ability to load data corresponding to a particular stock and then ask questions related to it. The user also has the ability to create and visualize important financial metrics via plots and get insights from the model itself.

Even after my best efforts to create the system completely locally so that it could be easily deployed to a server without the requirement of a GPU, it turned out that even after quantization of the model (reduction in weight precisions), the model still required a GPU to function. Hence the decision was later taken to shift to an interactive kaggle notebook and provide a chat interface that is as intuitive and easy to use as possible, while making sure that the required results were being generated.

undefinedDependency Installation: The following code block consists of the various libraries and frameworks required for the SectralCLI UI and Interface to function correctly.

Required Configuration

Load a model without downloading it - We leverage the Mistral model provided by Kaggle. In my case, my current setup, does not include a GPU which made it almost impossible even to train a quantized version of the Mistral 7B model. Hence, the required settings for this Kaggle Notebook to function correctly are,

• Internet: On (Toggle)
• Accelerator: GPU P100
• Model: mistral 7b-instruct-v0.1-hf/1 (by kaggle)
• Persistence: Files Only

Components

The entire notebook is divided into the following major sections: