llama2-7b

Model Card for Meta LLAMA2-7B Custom Task-Oriented Chatbot

This model is a fine-tuned version of Meta's LLAMA2-7B model, adapted to function as a task-oriented chatbot that processes and answers questions related to financial 10K reports.

Model Details

Model Description

Developed by Yatharth Mahesh Sant, this model is a causal language model fine-tuned from Meta's LLAMA2-7B to specifically handle queries and tasks associated with 10K financial reports. It is designed to assist financial analysts and stakeholders by providing detailed, accurate answers to inquiries about company performances, financial standings, and other key metrics contained within 10K reports.

• Developed by: Yatharth Mahesh Sant
• Model type: Causal LM
• Language(s) (NLP): English
• Finetuned from: meta/llama2-7b
• Repository: Meta LLAMA2-7B

Uses

Intended Use

This model is meant to be used as a task-oriented bot to interact with users querying about details in 10K financial reports, enhancing the efficiency of financial analysis and decision-making processes.

Direct Use

The model can directly answer questions from financial reports, serving as an automated assistant to financial analysts, investors, and regulatory authorities who require quick, reliable interpretations of financial data.

Downstream Use

The model can be integrated into financial analysis software, used to power internal data review tools in corporations, or serve as a support system in investor relations departments to automate responses to common shareholder inquiries.

Out-of-Scope Use

This model is not designed for non-financial texts or languages other than English. It may not perform well in informal conversational settings or handle off-topic inquiries effectively.

Bias, Risks, and Limitations

The model's performance and responses are based on the data it was trained on, which primarily includes structured financial texts. As such, it may inherit biases from this data or fail to comprehend nuanced questions not directly related to financial reporting.

Recommendations

It is recommended that responses generated by this model be reviewed by a qualified financial analyst to confirm their accuracy before being used in critical decision-making processes.

How to Get Started with the Model

To get started with this model, please refer to the specific deployment guides and API documentation provided in the repository linked above.

Training Details

Training Data

The model was fine-tuned on a comprehensive dataset comprising several years' worth of 10K reports from companies across various industries, annotated for key financial metrics and queries.

Training Procedure

Preprocessing

Training data was preprocessed to normalize financial terminology and remove any non-relevant sections of the reports, focusing on the sections most pertinent to common queries.

Training Hyperparameters

The model was trained using a learning rate of 5e-5 with a batch size of 32 over 4 epochs, employing a transformer-based architecture optimized for natural language understanding tasks.

Evaluation

Testing Data

The model was evaluated on a separate validation set consisting of annotated 10K reports not seen during training to ensure it can generalize across different texts and query types.

Metrics

Evaluation metrics included accuracy, F1 score, and a custom metric for response relevance to financial queries.

Technical Specifications

Model Architecture

The model employs a transformer-based architecture, leveraging attention mechanisms to focus on relevant parts of the text when generating responses.

Compute Infrastructure

Training was conducted on cloud-based GPUs with support for high-throughput training sessions.