Skip to content

Dataset · tonychenxyz

Sweeperbench run logs and traces database

tonychenxyz/sweeperbench-traces

Tabular and text data from Sweeperbench runs featuring prompts, traces, verdicts, and metadata in CSV and JSON.

What's inside

The dataset is a Sweeperbench run database organized around a central runs.csv table, where each row describes a prediction or evaluation step. Individual directories by run UUID contain corresponding metadata.json files, artifact manifests, prompts, execution traces, verdicts, and collected records. The field schema is described in schema.json.

What you can do with it

  • Debugging and analyzing LLM agent errors based on detailed logs and evaluation verdicts
  • Selecting successful execution traces for subsequent model fine-tuning
Size
10K–100K rows
Downloads this month
20.4K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("tonychenxyz/sweeperbench-traces")