Dataset Viewer
Auto-converted to Parquet Duplicate
instance_id
stringlengths
12
54
repo
stringclasses
628 values
license
stringclasses
5 values
trajectory_id
stringlengths
36
36
trajectory
listlengths
19
201
model_patch
stringlengths
88
63.4M
dataset
stringclasses
3 values
pytorch__vision-6883
pytorch/vision
BSD-3-Clause
c6e83ba0-793a-4afc-8e4a-428037baad87
[ { "content": "You are OpenHands agent, a helpful AI assistant that can interact with a computer to solve tasks.\n\n<ROLE>\nYour primary role is to assist users by executing commands, modifying code, and solving technical problems effectively. You should be thorough, methodical, and prioritize quality over speed...
diff --git a/torchvision/models/_api.py b/torchvision/models/_api.py index 52ac070e6d..2f239b4cf1 100644 --- a/torchvision/models/_api.py +++ b/torchvision/models/_api.py @@ -75,6 +75,11 @@ class WeightsEnum(StrEnum): return object.__getattribute__(self.value, name) return super().__getattr__(...
internlm/SWE-Fixer-Train-110K
scikit-image__scikit-image-4493
scikit-image/scikit-image
MIT
3bfdd0b4-1473-4030-a6cf-343913928299
[ { "content": "You are OpenHands agent, a helpful AI assistant that can interact with a computer to solve tasks.\n\n<ROLE>\nYour primary role is to assist users by executing commands, modifying code, and solving technical problems effectively. You should be thorough, methodical, and prioritize quality over speed...
diff --git a/skimage/draw/_random_shapes.py b/skimage/draw/_random_shapes.py index d3b3c8919..6485f241d 100644 --- a/skimage/draw/_random_shapes.py +++ b/skimage/draw/_random_shapes.py @@ -1,6 +1,6 @@ import numpy as np -from . import polygon as draw_polygon, circle as draw_circle +from . import polygon as draw_poly...
SWE-Gym/SWE-Gym-Raw
Netflix__lemur-267
Netflix/lemur
Apache-2.0
ae3180db-82c9-43a4-ada6-0fbd59f89694
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/lemur/authorities/service.py b/lemur/authorities/service.py\nindex 9a7ff6a2..a49c52a4 (...TRUNCATED)
internlm/SWE-Fixer-Train-110K
ipython__ipython-7479
ipython/ipython
BSD-3-Clause
f39e5f88-c7be-4568-afc8-4e915d523716
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/IPython/html/services/contents/fileio.py b/IPython/html/services/contents/fileio.py\ni(...TRUNCATED)
internlm/SWE-Fixer-Train-110K
mesonbuild__meson-5572
mesonbuild/meson
Apache-2.0
3739eb46-92ba-4430-96f2-c3b8501a92f3
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/mesonbuild/backend/backends.py b/mesonbuild/backend/backends.py\nindex b9bf166d8..fc51(...TRUNCATED)
internlm/SWE-Fixer-Train-110K
keras-team__keras-nlp-385
keras-team/keras-nlp
Apache-2.0
46278b12-5c28-4739-bde6-009c183b7b53
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/keras_nlp/models/bert/bert_models.py b/keras_nlp/models/bert/bert_models.py\nindex 016(...TRUNCATED)
nebius/SWE-rebench
jazzband__django-debug-toolbar-1832
jazzband/django-debug-toolbar
BSD-3-Clause
86cbe658-3573-46b9-a507-8d12ac30ea76
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/debug_toolbar/panels/sql/tracking.py b/debug_toolbar/panels/sql/tracking.py\nindex 0c5(...TRUNCATED)
internlm/SWE-Fixer-Train-110K
pre-commit__pre-commit-566
pre-commit/pre-commit
MIT
f4f775d6-6403-40c1-be17-2723add956dc
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/pre_commit/commands/install_uninstall.py b/pre_commit/commands/install_uninstall.py\ni(...TRUNCATED)
nebius/SWE-rebench
googleapis__google-cloud-python-6262
googleapis/google-cloud-python
Apache-2.0
3f80e282-d1c3-4ba8-a48e-f7ecda90f097
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/redis/README.rst b/redis/README.rst\nindex 4df28b2439..50e38148a5 100644\n--- a/redis/(...TRUNCATED)
internlm/SWE-Fixer-Train-110K
pypi__warehouse-2978
pypi/warehouse
Apache-2.0
d507b54e-9316-4493-96b8-a7ea1433a5b5
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED)
"diff --git a/warehouse/email.py b/warehouse/email.py\nindex ac00bb574..0938af4de 100644\n--- a/ware(...TRUNCATED)
SWE-Gym/SWE-Gym-Raw
End of preview. Expand in Data Studio

nvidia/SWE-Zero-openhands-trajectories(複製)

データの内容

GitHub の issue を解くソフトウェア工学エージェントの実行軌跡のデータ。1 行が 1 回の試行で、会話全体を trajectory、最終的な修正を unified diff 形式の model_patch に持ち、対象の課題は repo と instance_id で示される。課題は SWE-Gym・R2E-Gym-Subset・SWE-rebench など 5 つのデータセットから集め、軌跡は Qwen3-Coder-480B-A35B-Instruct で合成した。使うリポジトリは MIT・Apache-2.0・BSD 系の寛容なライセンスのものに限られる。

データの概要

項目 値
行数 318,115(1 構成・1 分割の合計)
列数 7
容量 12.21 GB(ファイル 67 本)
ファイルの形式 parquet 64、json 1
公開 public

構成と分割

構成(config) 分割(split)ごとの行数 列数
default train 318,115 7

列(構成 default・分割 train)

列 型 先頭の行の値
instance_id string pytorch__vision-6883
repo string pytorch/vision
license string BSD-3-Clause
trajectory_id string c6e83ba0-793a-4afc-8e4a-428037baad87
trajectory list<struct{content: string, role: string, tool_calls: list<struct{…}>}> [31 件]
model_patch string diff --git a/torchvision/models/_api.py b/torchv…
dataset string internlm/SWE-Fixer-Train-110K

行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は HYUKI5/dataset-catalog にある。

複製について

nvidia/SWE-Zero-openhands-trajectories をリビジョン 7b3cd106d00f60918e722d33a1d74bc67072a7ea の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。

出典

論文

アノテーションの方法

(元のカードに該当の見出しが無いので、ラベルの付け方に触れた段落を抜き出した)

SWE-ZERO Trajectories は、ソフトウェア工学における LLM の能力を高めることを目的とした、エージェント向けの指示チューニング用データセットである。このデータセットは、OpenHands フレームワークで収集した 318k 件のエージェントの軌跡(trajectory)からなる。軌跡は Qwen3-Coder-480B-A35B-Instruct で合成され、教師あり微調整(SFT)向けに整備されている。SWE-Bench 型のタスクでのモデル性能の向上を狙う。issue の記述(PR)は、許容的なライセンスのもとにある SWE-Gym、SWE-Gym-Raw、R2E-Gym-Subset、SWE-Fixer-Train-110K、SWE-rebench から取られている。issue の記述は、R2E-Gym-Subset のプロンプトを使い、Qwen3-Coder-480B-A35B-Instruct で生成した。

** ラベル付けの方法

  • [ハイブリッド: 自動, 合成]

微調整に使うときの注意

  • 調査メモの台帳での区分: A(許諾型のライセンスだけで、疑義の注記が無い)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
  • Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
  • 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。

調査メモの記述

メモ 節 ライセンス(記載) 確認 注記
nvidia_nemotron.md 付録. 取得した nvidia 組織のデータセット全件の簡易一覧 (L383) cc-by-4.0 確認済 確認印なし(各ファイルの凡例により確認済扱い) / 原文注記: × 数学・コード・科学・計測

元のカード(日本語訳)

以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。


SWE-Zero Trajectories: 実行を伴わない微調整によるソフトウェア工学エージェント

データ概要

SWE-ZERO Trajectories は、ソフトウェア工学における LLM の能力を高めることを目的とした、エージェント向けの指示チューニング用データセットである。このデータセットは、OpenHands フレームワークで収集した 318k 件のエージェントの軌跡(trajectory)からなる。軌跡は Qwen3-Coder-480B-A35B-Instruct で合成され、教師あり微調整(SFT)向けに整備されている。SWE-Bench 型のタスクでのモデル性能の向上を狙う。issue の記述(PR)は、許容的なライセンスのもとにある SWE-Gym、SWE-Gym-Raw、R2E-Gym-Subset、SWE-Fixer-Train-110K、SWE-rebench から取られている。issue の記述は、R2E-Gym-Subset のプロンプトを使い、Qwen3-Coder-480B-A35B-Instruct で生成した。

  • 技術報告 を読むと、SWE-ZERO Trajectories の方法論と技術的詳細がわかる。
  • 使用するライセンスは MIT、Apache-2.0、BSD-2-Claude、BSD-3-Clause のみである。
  • このデータセットは商用・非商用のどちらでも利用できる。

データの分布

データのフィールド

Field Type Description
instance_id string インスタンスを一意に識別する ID。R2E-Gym の場合は、リポジトリ名とコミットのハッシュからなる。
repo string owner/repo 形式の GitHub リポジトリ
license string リポジトリの SPDX ライセンス識別子
trajectory_id string 収集した軌跡ごとに一意な識別子
trajectory list 役割 'system' と 'user'(初期プロンプト)、'assistant'(モデルの推論と行動)、'tool'(環境の観測)を含む会話履歴の全体
model_patch string エージェントが作成した最終的なコード修正。unified diff 形式
dataset string インスタンスを収集元のデータセットの名前

データセットの特性

** データ収集方法

  • [ハイブリッド: 自動, 合成]

** ラベル付けの方法

  • [ハイブリッド: 自動, 合成]

ライセンス・利用条件

このデータセットは Creative Commons Attribution 4.0 International License(CC BY 4.0)のもとで提供される。追加のライセンスとして、MIT License、Apache 2.0 License、BSD 2 Clause License、BSD 3 Clause License が適用される。

データの開発者: NVIDIA

データのバージョン

1.0

想定される用途

このデータセットは、実世界の issue の記述を解釈し、複数段階のツール利用を計画し、コードベースを探索して、SWE-Bench 型の設定で修正を実装しなければならない、自律的なソフトウェア工学エージェントやコード志向のアシスタントを作る LLM エンジニアと研究チームを想定している。SFT や、そのようなモデルの蒸留に向いている。軌跡は、エージェントの方策の評価とデバッグ、リポジトリを踏まえた推論の改善、回帰を起こさないコード編集の振る舞いの研究にも使える。学術と実運用の両方の環境が対象となる。

倫理的配慮:

NVIDIA は、信頼できる AI は共同の責任であると考えており、多様な AI 応用の開発を可能にする方針と慣行を整えている。ダウンロードして利用規約に従って使う場合、開発者は社内のモデルチームと協力して、そのモデルが関連する業界と用途の要件を満たすこと、そして想定外の製品上の誤用に対処することを確かめるべきである。

セキュリティの脆弱性や NVIDIA の AI に関する懸念は こちら から報告してほしい。

引用

データが有用だった場合は、次の文献を引用してほしい。

@article{ahmad2026swezerotohero,
      title={From SWE-ZERO to SWE-HERO: Execution-free to Execution-based Fine-tuning for Software Engineering Agents}, 
      author={Nikolai Ludwig, Wasi Uddin Ahmad, Somshubra Majumdar, Boris Ginsburg},
      year={2026},
      eprint={2604.01496},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2604.01496}, 
}
原文(英語)

SWE-Zero Trajectories: Execution-free Fine-tuning for Software Engineering Agents

Data Overview

SWE-ZERO Trajectories is an agentic instruction tuning dataset designed to advance the capabilities of LLMs in software engineering. This dataset comprises 318k agent trajectories collected using the OpenHands framework. The trajectories were synthesized using Qwen3-Coder-480B-A35B-Instruct, specifically curated for supervised fine-tuning (SFT), aiming to improve model performance on SWE-Bench style tasks. The issue statements (PRs) are sourced from SWE-Gym, SWE-Gym-Raw, R2E-Gym-Subset, SWE-Fixer-Train-110K, SWE-rebench with permissive licenses. We use the prompts from R2E-Gym-Subset to generate issue statements using Qwen3-Coder-480B-A35B-Instruct.

  • Read the technical report to learn the methodology and technical details behind SWE-ZERO Trajectories.
  • We use MIT, Apache-2.0, BSD-2-Claude, and BSD-3-Clause licenses only.
  • This dataset is ready for commercial/non-commercial use.

Data distribution

Data Fields

Field Type Description
instance_id string Unique identifier for the instance. For R2E-Gym, the identifier consists of the repository name and the commit hash.
repo string GitHub repository in owner/repo format
license string SPDX license identifier of the repository
trajectory_id string The identifier unique for each collected trajectory
trajectory list Complete conversation history with roles: 'system' and 'user' (initial prompts), 'assistant' (model reasoning/actions) and 'tool' (environment observations)
model_patch string Final code modifications produced by the agent in unified diff format
dataset string The name of the dataset the instance is collected from

Dataset Characterization

** Data Collection Method

  • [Hybrid: Automated, Synthetic]

** Labeling Method

  • [Hybrid: Automated, Synthetic]

License/Terms of Use

This dataset is governed by the Creative Commons Attribution 4.0 International License (CC BY 4.0). Additional Information: MIT License, Apache 2.0 License, BSD 2 Clause License, and BSD 3 Clause License.

Data Developer: NVIDIA

Data Version

1.0

Intended use

This dataset is intended for LLM engineers and research teams building autonomous software engineering agents and code-focused assistants. It is suitable for supervised fine-tuning and distillation of models that must interpret real-world issue statements, plan multi-step tool use, navigate codebases, and implement fixes in a SWE-Bench–style setting. The trajectories can also be used to benchmark and debug agent policies, improve repository-aware reasoning, and study robust, regression-free code editing behaviors in both academic and production environments.

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Please report security vulnerabilities or NVIDIA AI Concerns here.

Citation

If you find the data useful, please cite:

@article{ahmad2026swezerotohero,
      title={From SWE-ZERO to SWE-HERO: Execution-free to Execution-based Fine-tuning for Software Engineering Agents}, 
      author={Nikolai Ludwig, Wasi Uddin Ahmad, Somshubra Majumdar, Boris Ginsburg},
      year={2026},
      eprint={2604.01496},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2604.01496}, 
}
Downloads last month
55

Paper for HYUKI5/nvidia__SWE-Zero-openhands-trajectories