ADFP aligns fingerprint token sampling with student learning dynamics via a proxy model to achieve stronger distillation detection and lower utility loss than prior heuristic methods on GSM8K, OASST1, and MBPP.
- Weight after adding additional jelly beans = 6 + 2 = 8 lbs
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Antidistillation Fingerprinting
ADFP aligns fingerprint token sampling with student learning dynamics via a proxy model to achieve stronger distillation detection and lower utility loss than prior heuristic methods on GSM8K, OASST1, and MBPP.