misc 0 0 0

Si1w/SWE-Lego-Data-gpt-6-luna-medium

published by @Si1w

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
194
Last 30 days
111
Likes
-
Size
993 KB
Created on HF
2026-09-06
Age
1 months ago

Description

SWE-Lego-Data-gpt-6-luna-medium

Step-level pruning annotations for the resolved splits of Lego-X/SWE-Lego-Real-Data and Lego-X/SWE-Lego-Synthetic-Data, annotated using gpt-6-luna with medium reasoning effort. The Real branch was annotated on 2026-10-07 and the Synthetic branch on 2026-10-10, both with the same prompt and verdict definitions: essential contributes to the final fix, neutral does not contribute but is a reasonable, harmless step worth imitating, and redundant is… See the full description on the dataset page: https://huggingface.co/datasets/Si1w/SWE-Lego-Data-gpt-6-luna-medium.

Tags

size_categories:10K<n<100Kformat:parquetmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantregion:us

README current version from Hugging Face


pretty_name: SWE-Lego-Data-gpt-6-luna-medium
configs:

  • config_name: default
    data_files:
    • split: train
      path: data/llm-annotation-00000-of-00001.parquet

SWE-Lego-Data-gpt-6-luna-medium

Step-level pruning annotations for the resolved splits of Lego-X/SWE-Lego-Real-Data and Lego-X/SWE-Lego-Synthetic-Data, annotated using gpt-6-luna with medium reasoning effort. The Real branch was annotated on 2026-10-07 and the Synthetic branch on 2026-10-10, both with the same prompt and verdict definitions: essential contributes to the final fix, neutral does not contribute but is a reasonable, harmless step worth imitating, and redundant is not worth imitating and names one supported reason. Non-contributing exploration before the fix is neutral, after the fix it is post_fix_exploration. For the Synthetic branch the reference patch is the fix (debug_patch), not the injected bug.

This repository was previously named SWE-Lego-Real-Data-gpt-6-luna-medium; the earlier revisions hold the Real branch only and the old name redirects here.

This repository contains annotations only. Join records to source trajectories using trajectory_id, which carries the branch as the swe-lego-real:: or swe-lego-synthetic:: prefix. For compatibility, the default configuration retains the train split and the existing Parquet path; the upstream source split is resolved.

Coverage

Measure Real Synthetic Total
Trajectories 5,009 9,127 14,136
Annotated steps 296,788 490,703 787,491
Essential 154,129 210,918 365,047
Neutral 68,050 144,516 212,566
Redundant 74,609 135,269 209,878

Redundant reasons

Reason Steps
post_fix_verification 89,384
post_fix_exploration 45,406
failed_attempt 31,394
restated_reasoning 26,585
repeated_failure 7,463
rewritten_edit 6,320
stalled_cycle 3,326

README history 8 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2026-10-10Add the Synthetic resolved branch annotations (gpt-6-luna medium)5a2e56a2.1 KB
    Loading...
  2. 2026-10-07Re-annotate with the revised neutral definition (ADR 0012)769541b1.6 KB
    Loading...
  3. 2026-10-01Update GPT-6 Luna medium annotations4aafef1892 B
    Loading...
  4. 2026-09-18Update LLM annotations with reviewed neutral labels106180d3 KB
    Loading...
  5. 2026-09-06Move annotations to numbered Parquet shard0e46a951.4 KB
    Loading...
  6. 2026-09-06Store step annotations in Parquet format04d46481.3 KB
    Loading...
  7. 2026-09-06Upload reviewed SWE-Lego step annotationsb1513b11.3 KB
    Loading...
  8. 2026-09-06initial commit43585400 B
    Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Community discussions 1 thread · 1 open · 1 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

  1. 2026-09-07[bot] Conversion to Parquetopen1 💬#1
    Loading...

Related datasets same stage · misc

Read further

Download the dataset

0

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python)
from datasets import load_dataset
ds = load_dataset("Si1w/SWE-Lego-Data-gpt-6-luna-medium")
Raw snapshot (Python)
from huggingface_hub import snapshot_download
snapshot_download(repo_id="Si1w/SWE-Lego-Data-gpt-6-luna-medium", repo_type="dataset")
git clone (requires git-lfs)
git clone https://huggingface.co/datasets/Si1w/SWE-Lego-Data-gpt-6-luna-medium
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration