Alignment datasets by siddharthmb

Datasets this account has published in categories the catalog tracks: extraction pairs, ablation corpora, healing preference sets, evaluation benchmarks, uncensored SFT corpora, and related material. Category badges link to the workflow stage.

6 in /datasets
Dataset
Stage
Downloads
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n52_plain 0 0
Refusal prefill-transplant test (n=52 prompts, plain template): transcripts 1144 transcripts from the prefill-transplant experiment of 20...
425
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n52_chat 0 0
Refusal prefill-transplant test (n=52 prompts, chat template): transcripts 1144 transcripts from the prefill-transplant experiment of 202...
191
2026.TA.experiments.refusal_tracing.refusal_ladder 0 0
Refusal ladder: transcripts Transcripts from the refusal-ladder evaluation of 2026-08-10. The same 93 prompts (63 strict-flip harmful Har...
78
2026.TA.experiments.refusal_tracing.strict_compliance_refusal 0 0
Strict base-compliance vs instruct-refusal screen (HarmBench, 200 prompts): transcripts 200 transcripts (one per HarmBench standard promp...
76
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n3 0 0
Refusal prefill-transplant test, 3-prompt pilot: transcripts The 2026-08-10 pilot of the prefill-transplant experiment on three harmful r...
73
2026.MI.Transcripts 0 0
Transcripts of the MATS-II experiments (June to September 2026) The transcripts/ tree of the MATS-II scratch volume, one tarball per expe...
48
Dataset
Stage
Downloads
2026.MI.Transcripts 0 0
Transcripts of the MATS-II experiments (June to September 2026) The transcripts/ tree of the MATS-II scratch volume, one tarball per expe...
48
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n3 0 0
Refusal prefill-transplant test, 3-prompt pilot: transcripts The 2026-08-10 pilot of the prefill-transplant experiment on three harmful r...
73
2026.TA.experiments.refusal_tracing.refusal_ladder 0 0
Refusal ladder: transcripts Transcripts from the refusal-ladder evaluation of 2026-08-10. The same 93 prompts (63 strict-flip harmful Har...
78
2026.TA.experiments.refusal_tracing.strict_compliance_refusal 0 0
Strict base-compliance vs instruct-refusal screen (HarmBench, 200 prompts): transcripts 200 transcripts (one per HarmBench standard promp...
76
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n52_plain 0 0
Refusal prefill-transplant test (n=52 prompts, plain template): transcripts 1144 transcripts from the prefill-transplant experiment of 20...
425
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n52_chat 0 0
Refusal prefill-transplant test (n=52 prompts, chat template): transcripts 1144 transcripts from the prefill-transplant experiment of 202...
191
Dataset
Stage
Downloads
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n52_plain 0 0
Refusal prefill-transplant test (n=52 prompts, plain template): transcripts 1144 transcripts from the prefill-transplant experiment of 20...
425
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n52_chat 0 0
Refusal prefill-transplant test (n=52 prompts, chat template): transcripts 1144 transcripts from the prefill-transplant experiment of 202...
191
2026.TA.experiments.refusal_tracing.refusal_ladder 0 0
Refusal ladder: transcripts Transcripts from the refusal-ladder evaluation of 2026-08-10. The same 93 prompts (63 strict-flip harmful Har...
78
2026.TA.experiments.refusal_tracing.strict_compliance_refusal 0 0
Strict base-compliance vs instruct-refusal screen (HarmBench, 200 prompts): transcripts 200 transcripts (one per HarmBench standard promp...
76
2026.TA.experiments.refusal_tracing.refusal_prefill_flip_n3 0 0
Refusal prefill-transplant test, 3-prompt pilot: transcripts The 2026-08-10 pilot of the prefill-transplant experiment on three harmful r...
73
2026.MI.Transcripts 0 0
Transcripts of the MATS-II experiments (June to September 2026) The transcripts/ tree of the MATS-II scratch volume, one tarball per expe...
48
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