ecairol/simpleqa-eval-qwen3.8-27b-obliterated
SimpleQA-Verified results: qwen3.8-27b, obliterated vs. normal Factual-accuracy evaluation of the abliterated ("obliterated") fine-tune OBLITERATUS/Qwen3.8-27B-OBLITERATED, compared against the normal (non-abliterated) base model Qwen/Qwen3.8-27B, against codelion/SimpleQA-Verified, run locally with Inspect via LM Studio. Result Model Samples (N) Accuracy Stderr qwen3.8-27b-obliterated 201 of 1000 11.4% 2.25% qwen3.8-27b (normal) 201 of 1000 29.4%… See the full description on the dataset page: https://huggingface.co/datasets/ecairol/simpleqa-eval-qwen3.8-27b-obliterated.
Librarian Bot: Add language metadata for dataset (#2)
Add obliterated vs. normal comparison; revise hedging-behavior finding
Add qwen3.8-27b (normal) SimpleQA-Verified run for comparison
Trigger dataset-viewer cache refresh after removing historyqa log
Remove historyqa sanity-check log (not part of the SimpleQA eval)
Add Space link
Add raw eval logs (simpleqa N=201, historyqa N=5)
Add README
initial commit