PLP_llama2_v4 / README.md
yatharth97's picture
Create README.md
9c1994e verified

Dataset Card for Structured Financial Document Queries Dataset

Dataset Description

This dataset consists of structured text entries designed to train natural language processing models on financial document analysis tasks. Each entry includes system prompts and user queries formatted with specific tags that provide contextual instructions to the model.

Tags and Formatting

  • <s>: Marks the beginning of a new segment, encapsulating an individual entry.
  • <<SYS>>: System prompt providing context or specific instruction for processing.
  • [INST]: A combined tag that includes both the system prompt and a user query related to a financial document.

Intended Use

The dataset is intended for the development and evaluation of models that can process and respond to queries related to financial reports, specifically 10K reports. It is tailored to enhance model capabilities in understanding and generating context-aware responses based on structured instructional data.

Limitations

The dataset is highly specialized towards financial document analysis, particularly 10K reports. Models trained on this dataset might not generalize well to other types of documents or queries outside the financial domain. Additionally, the structured nature of the data may limit the model's flexibility in handling less structured queries.

Dataset Size and Structure

  • Number of rows: Approximately 42.3k
  • Data format: Single text field per row, with each text encapsulated within <s> tags and containing system and user instructions ([INST]).

Example Entry

[INST] <> As a highly intelligent assistant, your primary goal is to provide accurate, relevant, and context-aware responses to user queries based on the provided information. Ensure your answers are factual, free from bias, and avoid promoting violence, ... Just uploaded the 10k report for Schlumberger Limited for 2022. Can you check it out? [/INST]

Data Fields

  • text: A string that includes the full instruction set for the model, composed of system prompts and user queries specifically formatted to facilitate model training in financial document query processing.

Source

This dataset is compiled from a variety of financial reports and associated queries, formatted to include tags for training conversational models in a controlled environment.

Usage

This dataset is suitable for training models that need to interact with financial databases or assist in financial decision-making processes. It can be used to train models for chatbots, automated report summarizers, and other AI tools that require an understanding of financial reports and the ability to respond to related queries.