Width pruning in Llama-3.2 models reduces parametric knowledge while enhancing instruction-following and preserving reasoning.
Lambada (PPL) WikiText-2 (PPL) Selection Criteria Value∆vs
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Fragile Knowledge, Robust Instruction-Following: The Width Pruning Dichotomy in Llama-3.2
Width pruning in Llama-3.2 models reduces parametric knowledge while enhancing instruction-following and preserving reasoning.