Get the app
extract 0 0 0

latkes/inside-out-replication-v2-pamqfix-canary-v1

published by @latkes

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
107
Last 30 days
17
Likes
-
Size
546 KB
Created on HF
2026-05-19
Age
5 months ago

Description

Inside-Out Replication V2 — P(a|q) A.8.1-fix CANARY

Corrected externals (04_external_scores.py with the A.8.1 in-context

tokenization fix) over a deterministic 25-question-id subset of

Mistral-7B-Instruct-v0.3 / P264 / test (5832 candidate (q,a) rows).

Bounded subset used to validate the fix end-to-end on MLL before the full

3×4 re-score (per preflight rev2 spec; see red_team_brief.md).

Canary self-validation

C2 rows scored: 5832 (no build_qa_sequence ValueError;… See the full description on the dataset page: https://huggingface.co/datasets/latkes/inside-out-replication-v2-pamqfix-canary-v1.

Tags

size_categories:1K<n<10Kformat:parquetformat:optimized-parquetmodality:tabularmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantregion:us

README current version from Hugging Face

Inside-Out Replication V2 — P(a|q) A.8.1-fix CANARY

Corrected externals (04_external_scores.py with the A.8.1 in-context
tokenization fix) over a deterministic 25-question-id subset of
Mistral-7B-Instruct-v0.3 / P264 / test (5832 candidate (q,a) rows).
Bounded subset used to validate the fix end-to-end on MLL before the full
3×4 re-score (per preflight rev2 spec; see red_team_brief.md).

Canary self-validation

  • C2 rows scored: 5832 (no build_qa_sequence ValueError; would have
    crashed the job otherwise).
  • C5 rows failing eot ≤ content / norm ≤ 0 / NaN: 0 / 5832.
  • C1 rows whose log_p_a_q changed vs the OLD buggy-concat phase4 (same
    (qid,answer) keys): 360 / 5832 (6.2%).

Subset K vs committed full-pool corrected baseline (Mistral P264)

Note: subset K (25 qids) vs baseline K (442 qids) are NOT directly
comparable — magnitudes drift with subset choice. The apples-to-apples
check is C1 above (per-row diff on identical keys).

metric full-pool baseline A.8.1-fixed subset
log_p_a_q (paper-faithful) 0.5080 0.5004
log_p_norm_a_q 0.3154 0.2461
p_true (unchanged code) 0.4475 0.3861
log_p_a_q_eot (NEW sensitivity) — 0.4614
log_p_norm_a_q_eot (NEW) — 0.2401

Probe is DEFERRED in this phase (see NEXT_STEPS_TODO.md).

Reproducibility

The fix is on branch fix/judge-heuristic of the project repo (commits
60dc183 → 6a12f03). The pre-flight + canary harness lives in
sbatch/pamqfix/. The no-GPU correctness gate is tests/test_qa_sequence.py.

README history 3 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. 2026-05-19Upload README.md with huggingface_hub9fdee981.5 KB
    Loading...
  2. 2026-05-19Upload dataset6f0afa21.1 KB
    Loading...
  3. 2026-05-19initial commit46800060 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.

Related datasets same stage · extract

Anthropic/hh-rlhf
2.1M dl 2442 likes 90.4 MB
Dataset Card for HH-RLHF Dataset Summary This repository provides access to two different kinds of data: Human ...
djain95/sae-jailbreaks-cache
468.6K dl 1 likes 30.6 GB
mosaicml/dolly_hhrlhf
309.5K dl 113 likes 23.7 MB
Dataset Card for "dolly_hhrlhf" This dataset is a combination of Databrick's dolly-15k dataset and a filtered subset of Anthrop...
Dahoas/full-hh-rlhf
266.3K dl 104 likes 129.8 MB
Dataset Card for "full-hh-rlhf" Anthropic's HH dataset reformatted into prompt, chosen, rejected samples.
mlabonne/harmful_behaviors
198.2K dl 169 likes 23 KB
mlabonne/harmless_alpaca
190.2K dl 51 likes 1.2 MB
walledai/AdvBench
176.3K dl 123 likes 38 KB
Dataset Card for AdvBench Paper: Universal and Transferable Adversarial Attacks on Aligned Language Models Data: AdvBench Datas...
JailbreakV-28K/JailBreakV-28k
116.4K dl 80 likes 288.8 MB
⛓‍💥 JailBreakV-28K: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks 🌐 ...

Read further

Community best practices (extraction)
5 rules the field converged on for building the pair.
Building your own extraction pair
When the reference pair is not enough. Sizing, framing, and evaluation.
Datasets that shaped abliteration
The five corpora that did most of the work.
Non-English abliteration communities
How Chinese, Russian, Japanese, and Korean practitioners build extraction pairs.

Repository

DIR data FILE .gitattributes FILE README.md

File contents live on Hugging Face. Click a file or folder to open it there.

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("latkes/inside-out-replication-v2-pamqfix-canary-v1")
Raw snapshot (Python) 
from huggingface_hub import snapshot_download
snapshot_download(repo_id="latkes/inside-out-replication-v2-pamqfix-canary-v1", repo_type="dataset")
git clone (requires git-lfs) 
git clone https://huggingface.co/datasets/latkes/inside-out-replication-v2-pamqfix-canary-v1
Browse repository on Hugging Face ->
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.

Get the app Browse the catalog
Open in Abliteration