tetracta/llm-xray-twin-study-1b
Updated • 2
Layer-by-layer scans of open checkpoints: what fine-tuning, quantization and merging actually change. Free open beta at tetracta.ai/xray.html
Note Twin study: same data, two architectures — structural portraits
Note Case study subject (base) — see the layer map in our scan
Note Scanned vs base: change starts at layer 4, ~89% of depth; 0 facts broken; fabrication-avoidance 13/20 -> 18/20 (discussion #19)
Note Lesion scans: controlled damage, measured response
Compare two AI model versions and locate their differences
Note Sample reports, exactly as the tool produces them (trial watermark included)