Training Super Weights in isolation collapses LLM performance to random guessing, while full-layer low-rank updates succeed, showing parameter importance does not imply trainability.
The emergence of essential sparsity in large pre-trained models: The weights that matter
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Super Weights in LLMs and the Failure of Selective Training
Training Super Weights in isolation collapses LLM performance to random guessing, while full-layer low-rank updates succeed, showing parameter importance does not imply trainability.