Dataset Viewer
Auto-converted to Parquet Duplicate
messages
stringlengths
7.95k
664k
instance_id
stringlengths
23
79
resolved
bool
2 classes
model
stringclasses
3 values
traj_id
stringlengths
32
88
patch
stringlengths
0
393M
[{"role": "system", "content": "You are a helpful assistant that can interact with a computer to solve tasks.", "agent": "main", "message_type": "system_prompt"}, {"role": "user", "content": [{"type": "text", "text": "<uploaded_files>\n/testbed\n</uploaded_files>\nI've uploaded a python code repository in the directory...
django-money__django-money.835c1ab8.func_pm_ctrl_shuffle__viqnyl9u
true
claude-3-7-sonnet-20250219
django-money__django-money.835c1ab8.func_pm_ctrl_shuffle__viqnyl9u.8qa84d2e
[{"role": "system", "content": "You are a helpful assistant that can interact with a computer to solve tasks.", "agent": "main", "message_type": "system_prompt"}, {"role": "user", "content": [{"type": "text", "text": "<uploaded_files>\n/testbed\n</uploaded_files>\nI've uploaded a python code repository in the directory...
marshmallow-code__apispec.8b421526.func_pm_remove_assign__kdkrbg6a
true
claude-3-7-sonnet-20250219
marshmallow-code__apispec.8b421526.func_pm_remove_assign__kdkrbg6a.8qa84d2e
diff --git a/flashtext/keyword.py b/flashtext/keyword.py index 673248d..fda82b2 100644 --- a/flashtext/keyword.py +++ b/flashtext/keyword.py @@ -411,18 +411,6 @@ class KeywordProcessor(object): self.remove_keyword(keyword) def get_all_keywords(self, term_so_far='', current_dict=None): - for k...
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
mozillazg__python-pinyin.e42dede5.func_pm_remove_assign__vn346ybe
true
claude-3-7-sonnet-20250219
mozillazg__python-pinyin.e42dede5.func_pm_remove_assign__vn346ybe.8qa84d2e
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
marshmallow-code__marshmallow.9716fc62.func_pm_class_rm_funcs__e35a2bbx
true
claude-3-7-sonnet-20250219
marshmallow-code__marshmallow.9716fc62.func_pm_class_rm_funcs__e35a2bbx.8qa84d2e
"diff --git a/LICENSE.txt b/LICENSE.txt\ndeleted file mode 100644\nindex 57bc88a..0000000\n--- a/LIC(...TRUNCATED)
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
luozhouyang__python-string-similarity.115acaac.func_pm_remove_cond__gch4emzr
true
claude-3-7-sonnet-20250219
luozhouyang__python-string-similarity.115acaac.func_pm_remove_cond__gch4emzr.8qa84d2e
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
tkrajina__gpxpy.09fc46b3.func_pm_ctrl_shuffle__grix6fbc
true
claude-3-7-sonnet-20250219
tkrajina__gpxpy.09fc46b3.func_pm_ctrl_shuffle__grix6fbc.8qa84d2e
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
sloria__environs.73c372df.func_pm_remove_assign__l35izs7k
true
claude-3-7-sonnet-20250219
sloria__environs.73c372df.func_pm_remove_assign__l35izs7k.8qa84d2e
"diff --git a/gpxpy/geo.py b/gpxpy/geo.py\nindex 4f8d7cc..0632140 100644\n--- a/gpxpy/geo.py\n+++ b/(...TRUNCATED)
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
amueller__word_cloud.ec24191c.func_pm_ctrl_shuffle__3gl981ew
true
claude-3-7-sonnet-20250219
amueller__word_cloud.ec24191c.func_pm_ctrl_shuffle__3gl981ew.8qa84d2e
"diff --git a/dvc/render/match.py b/dvc/render/match.py\nindex 108654d..f643577 100644\n--- a/dvc/re(...TRUNCATED)
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
facelessuser__soupsieve.a8080d97.func_pm_remove_cond__0nmyui1x
false
claude-3-7-sonnet-20250219
facelessuser__soupsieve.a8080d97.func_pm_remove_cond__0nmyui1x.8qa84d2e
"diff --git a/ptyprocess/ptyprocess.py b/ptyprocess/ptyprocess.py\nindex 03f8cd7..a5a06c1 100644\n--(...TRUNCATED)
"[{\"role\": \"system\", \"content\": \"You are a helpful assistant that can interact with a compute(...TRUNCATED)
joke2k__faker.8b401a7d.func_pm_ctrl_invert_if__bompadwm
false
claude-3-7-sonnet-20250219
joke2k__faker.8b401a7d.func_pm_ctrl_invert_if__bompadwm.8qa84d2e
"diff --git a/djmoney/money.py b/djmoney/money.py\nindex bc67644..ed426d9 100644\n--- a/djmoney/mone(...TRUNCATED)
End of preview. Expand in Data Studio

SWE-bench/SWE-smith-trajectories(複製)

データの内容

SWE-smith の課題を SWE-agent と Claude 3.7 Sonnet で解かせて生成した軌跡のデータ。主な列は、課題の instance_id、成否の resolved、生成されたパッチの patch、軌跡の messages(文字列)である。このカードによれば、この軌跡で Qwen 2.5 Coder Instruct を微調整し、SWE-agent-LM-32B を作った。

データの概要

項目 値
行数 76,002(1 構成・3 分割の合計)
列数 6
容量 4.22 GB(ファイル 34 本)
ファイルの形式 parquet 32
言語 en
公開 public

構成と分割

構成(config) 分割(split)ごとの行数 列数
default tool 24,100、xml 26,076、ticks 25,826 6

列(構成 default・分割 tool)

列 型 先頭の行の値
messages string [{"role": "system", "content": "You are a helpfu…
instance_id string django-money__django-money.835c1ab8.func_pm_ctrl…
resolved bool True
model string claude-3-7-sonnet-20250219
traj_id string django-money__django-money.835c1ab8.func_pm_ctrl…
patch string

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

SWE-bench/SWE-smith-trajectories をリビジョン 08e109b4a59eaeebf80e4675cd125d42e7ac99a4 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)

軌跡は、SWE-agent と Claude 3.7 Sonnet を動かして生成した。タスクのインスタンスは SWE-smith の データセット から取った。

微調整に使うときの注意

  • 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。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
agent_trajectories_thought_logs.md Q1. コーディングエージェントの軌跡で、自動テストの成否が付いたものは何があるか (L21) mit 確認済 生成元モデルの出力規約が別に及ぶ(Q4) / 原文注記: カード冒頭は「5,017 trajectories」と書くが、tool split は 24,100 件で食い違う(カードの 5,017 は SWE-agent-LM-32B の微調整に使った分と読めるが、確認していない

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。


コード • 論文 • サイト

このデータセットは、Qwen 2.5 Coder Instruct を微調整するために使った 5017 件の軌跡を含む。これにより SWE-agent-LM-32B が得られた。これは、コーディング用の LM エージェントであり、SWE-bench Verified で 40.2% を達成している(検証器や複数のロールアウトは使わず、インスタンスあたり 1 回の試行のみ)。

軌跡は、SWE-agent と Claude 3.7 Sonnet を動かして生成した。タスクのインスタンスは SWE-smith の データセット から取った。

原文(英語)

Code • Paper • Site

This dataset contains the 5017 trajectories we fine-tuned Qwen 2.5 Coder Instruct on, leading to SWE-agent-LM-32B, a coding LM agent that achieve 40.2% on SWE-bench Verified (no verifiers or multiple rollouts, just 1 attempt per instance).

Trajectories were generated by running SWE-agent + Claude 3.7 Sonnet on task instances from the SWE-smith dataset.

Downloads last month
52

Paper for HYUKI5/SWE-bench__SWE-smith-trajectories