Skip to content

Dataset · snorkelai

Agent traces for 10-K financial questions

snorkelai/agent-finance-reasoning

ReAct agent traces and question-answer pairs based on company 10-K financial reports.

What's inside

The dataset contains traces and metadata of agentic interactions for the financial reasoning task. The agent is built using langgraph and ReAct: it receives a question and must answer using data from one or more 10-Ks available through tools. Question-answer pairs were created in collaboration with the Snorkel expert network. The language is English, with a volume of fewer than 1K examples.

What you can do with it

  • Fine-tune an agent for QA on financial documents
  • Evaluate a ReAct system with 10-K access tools
  • Gather SFT examples for agentic financial reasoning
Tasks
Question answering
Languages
English
Size
up to 1K rows
Downloads this month
1.3K
Likes
69

How to load it

from datasets import load_dataset

ds = load_dataset("snorkelai/agent-finance-reasoning")