Datasets:

SofiTesfay2010
/

SWE-Traces

messages listlengths 19 322
[ { "content": "Task Instance ID: aaugustin__websockets-627\n\nProblem Statement:\n", "role": "user" }, { "content": "You are a helpful assistant that can interact with a computer to solve tasks.", "role": "user" }, { "content": "<uploaded_files>\n/testbed\n</uploaded_files>\nI've uploaded...
[{"content":"Task Instance ID: theacodes__nox-482\n\nProblem Statement:\n","role":"user"},{"content"(...TRUNCATED)
[{"content":"Task Instance ID: weaveworks__flagger-715\n\nProblem Statement:\n","role":"user"},{"con(...TRUNCATED)
[{"content":"Task Instance ID: abuiles__ember-watson-53\n\nProblem Statement:\n","role":"user"},{"co(...TRUNCATED)
[{"content":"Task Instance ID: gcanti__fp-ts-1066\n\nProblem Statement:\n","role":"user"},{"content"(...TRUNCATED)
[{"content":"Task Instance ID: eslint__eslint-8459\n\nProblem Statement:\n","role":"user"},{"content(...TRUNCATED)
[{"content":"Task Instance ID: jlowin__fastmcp-63\n\nProblem Statement:\n","role":"user"},{"content"(...TRUNCATED)
[{"content":"Task Instance ID: ucan-lab__laravel-dacapo-176\n\nProblem Statement:\n","role":"user"},(...TRUNCATED)
[{"content":"Task Instance ID: cdk8s-team__cdk8s-plus-4483\n\nProblem Statement:\n","role":"user"},{(...TRUNCATED)
[{"content":"Task Instance ID: marp-team__marp-vscode-510\n\nProblem Statement:\n","role":"user"},{"(...TRUNCATED)

End of preview. Expand in Data Studio

SWE-Traces (Resolved & Formatted for SFT)

This dataset is a filtered and formatted variant of nvidia/Open-SWE-Traces. It is optimized for Supervised Fine-Tuning (SFT) of software engineering agents.

Dataset Details

Source Split: minimax_m25
Source Configuration: sweagent
Filtering Logic: Only retains successful executions where resolved == 1 (filtering out all failure paths, incomplete traces, and erroneous steps).
Format: Converted into a standardized multi-turn conversational format (messages array with role and content fields) compatible with standard SFT pipelines and chat templates.
Size: 19,551 instances (~1.17 GB raw parquet format).

Dataset Schema

Each row contains a single conversational sequence:

{
  "messages": [
    {
      "role": "user",
      "content": "Task Instance ID: <id>\n\nProblem Statement:\n<problem_statement>"
    },
    {
      "role": "assistant",
      "content": "<first_agent_action_or_thought>"
    },
    {
      "role": "user",
      "content": "<first_tool_output_or_environment_response>"
    },
    ...
  ]
}

Quick Start / Usage

from datasets import load_dataset

dataset = load_dataset("SofiTesfay2010/SWE-Traces")
print(dataset["train"][0])

Downloads last month: 44