CoolFace
Datasetpublic

yamaTK/merged_dataset_final_clean_v41

merged_dataset_final_clean_v41 English Rule-based cleaned SFT dataset for structured output generation (JSON / YAML / XML / TOML / CSV). Data Source This dataset was built from competition-provided datasets only. The cleaning pipeline loads the following source groups: u-10bei (6 datasets: source ids 1-1 to 1-6) daichira (3 datasets: source ids 2-1 to 2-3) After strict filtering and sampling for v4.1, the final retained rows are from u-10bei… See the full description on the dataset page: https://huggingface.co/datasets/yamaTK/merged_dataset_final_clean_v41.

sourceHugging Faceotherupdated 7mo agoView on Hugging Face
0likes21downloads
Dataset Card

mergeddatasetfinalcleanv41

English

Rule-based cleaned SFT dataset for structured output generation (JSON / YAML / XML / TOML / CSV).

Data Source

This dataset was built from competition-provided datasets only.

The cleaning pipeline loads the following source groups:

  • —u-10bei (6 datasets: source ids 1-1 to 1-6)
  • —daichira (3 datasets: source ids 2-1 to 2-3)

After strict filtering and sampling for v4.1, the final retained rows are from u-10bei sources (1-1 to 1-6).

Processing (What was done to original data)

merged_dataset_final_clean_v41.jsonl is generated by deterministic, rule-based preprocessing (no LLM rewriting).

Main steps:

  1. 1.Merge and normalize records into a unified message format (system, user, assistant).
  2. 2.Parse/validate assistant outputs by target format (CSV/JSON/YAML/XML/TOML) with rule checks.
  3. 3.Remove duplicates (exact hash + near-duplicate handling in the cleaning pipeline).
  4. 4.Remove extra non-structured text and malformed fenced outputs.
  5. 5.Apply consistency and outlier filtering.
  6. 6.Remove Approach: ... Output: style CoT prefix and keep only final output payload where applicable.
  7. 7.Apply aggressive, deterministic sampling to match training constraints and budget.

Mechanical-Only Transformation Statement

  • —No LLM was used to generate, rewrite, paraphrase, or synthesize training samples.
  • —Transformations are implemented with deterministic Python logic (parsers, validators, hashing, filtering, sampling).
  • —public_150.json was not modified during dataset construction.

Dataset Statistics (v41 file)

  • —File: merged_dataset_final_clean_v41.jsonl
  • —Total rows: 1000
  • —Target format distribution:
  • —json: 332
  • —yaml: 234
  • —toml: 164
  • —xml: 135
  • —csv: 135
  • —Source ids in final file: 1-1, 1-2, 1-3, 1-4, 1-5, 1-6

Columns

Each JSONL row contains:

  • —source
  • —series
  • —messages (chat-style list with roles/content)
  • —format
  • —source_format
  • —complexity
  • —schema
  • —type
  • —n_tokens

日本語

構造化出力(JSON / YAML / XML / TOML / CSV)向けSFT用に、ルールベースでクリーニングしたデータセットです。

元データ

コンペ提供データのみを使用しています。

クリーニングパイプラインでは次の系列を読み込みます。

  • —u-10bei(6データセット: 1-1〜1-6)
  • —daichira(3データセット: 2-1〜2-3)

ただし v4.1 の厳しめのフィルタとサンプリング後、最終的に残った行は u-10bei 系(1-1〜1-6)です。

元データに対して行った処理

merged_dataset_final_clean_v41.jsonl は、決定的なルール処理のみで生成しています(LLMによる書き換えなし)。

主な処理:

  1. 1.レコードを system / user / assistant の統一形式に正規化
  2. 2.出力フォーマット別(CSV/JSON/YAML/XML/TOML)に構文検証
  3. 3.重複除去(完全一致ハッシュ + 近似重複処理)
  4. 4.余計な説明文や不正なフェンス付き出力の除去
  5. 5.整合性チェックと外れ値フィルタ
  6. 6.Approach: ... Output: 形式のCoT接頭部を除去し、最終出力本体のみ保持
  7. 7.学習制約に合わせた決定的サンプリング

機械的変換のみであること

  • —学習データの生成・書き換え・言い換え・合成にLLMは使っていません。
  • —変換はすべて、Pythonによる決定的な処理(パース・検証・ハッシュ・フィルタ・サンプリング)です。
  • —データ構築時に public_150.json は変更していません。

データ統計(v41)

  • —ファイル: merged_dataset_final_clean_v41.jsonl
  • —総件数: 1000
  • —目標フォーマット内訳:
  • —json: 332
  • —yaml: 234
  • —toml: 164
  • —xml: 135
  • —csv: 135
  • —最終ファイル内の source id: 1-1, 1-2, 1-3, 1-4, 1-5, 1-6

カラム

各JSONL行の主なキー:

  • —source
  • —series
  • —messages
  • —format
  • —source_format
  • —complexity
  • —schema
  • —type
  • —n_tokens