Instruction-tuned language models stabilize their next-token predictions later in the forward pass than pretrained models, with late MLP layers providing the strongest tested control point under matched histories.
Title resolution pending
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
-
The Convergence Gap: Instruction-Tuned Language Models Stabilize Later in the Forward Pass
Instruction-tuned language models stabilize their next-token predictions later in the forward pass than pretrained models, with late MLP layers providing the strongest tested control point under matched histories.