OBLITERATUS model weight modification process (EXCISE).
Raw Developer Origin & Technical Request
GitHub Issue
Mar 5, 2026
Hello, as the title says - Issue :(
COSMIC layer selection: bottom 3 by cosine similarity (range 0.7758..0.9997)
Layer selection: knee=7, COSMIC=3, fused=7
Selected 7 layers via knee_cosmic (threshold=3866751.0000)
Strong refusal layers: [27, 26, 25, 24, 23, 22, 21]
Refusal subspace extracted (4.1s)
Wrapping 100 prompts with chat template
chat template 100/100
Capturing baseline logits on 100 harmless prompts for KL...
Captured baseline logits: torch.Size([100, 152064])
✂️ EXCISE — Modifying weights...
Capping refinement_passes from 2 to 1: norm_preserve without re-probing causes compound amplification (directions are not re-extracted)
ERROR: result type Float can't be cast to the desired output type Byte
Using Cuda Nightly 12.8
Developer Debate & Comments
No active discussions extracted for this entry yet.
Adjacent Repository Pain Points
Other highly discussed features and pain points extracted from elder-plinius/OBLITERATUS.
Engagement Signals
Cross-Market Term Frequency
Quantifies the cross-market adoption of foundational terms like KL and COSMIC layer selection by tracking occurrence frequency across active SaaS architectures and enterprise developer debates.
Market Trends