pith. sign in

Layerwise importance analysis of feed-forward networks in transformer-based language models.arXiv preprint arXiv:2508.17734,

Wataru Ikeda, Kazuki Yano, Ryosuke Takahashi, Jaesung Lee, Keigo Shibata, Jun Suzuki · 2025 · arXiv 2508.17734

3 Pith papers cite this work. Polarity classification is still indexing.

3 Pith papers citing it
read on arXiv browse 3 citing papers

fields

cs.LG 2 cs.CL 1

years

2026 3

verdicts

UNVERDICTED 2 CONDITIONAL 1

representative citing papers

Tapered Language Models

cs.LG · 2026-06-22 · unverdicted · novelty 7.0

Tapered Language Models monotonically decrease MLP width across depth with a cosine schedule, yielding better perplexity and downstream performance than uniform-width baselines across multiple architectures and scales at no extra cost.

Variable-Width Transformers

cs.CL · 2026-06-16 · conditional · novelty 6.0

×-shaped variable-width transformers outperform parameter-matched uniform baselines on language modeling loss with 22% fewer FLOPs and 15% smaller KV cache.

citing papers explorer

Showing 3 of 3 citing papers.

  • Tapered Language Models cs.LG · 2026-06-22 · unverdicted · none · ref 16

    Tapered Language Models monotonically decrease MLP width across depth with a cosine schedule, yielding better perplexity and downstream performance than uniform-width baselines across multiple architectures and scales at no extra cost.

  • Variable-Width Transformers cs.CL · 2026-06-16 · conditional · none · ref 17

    ×-shaped variable-width transformers outperform parameter-matched uniform baselines on language modeling loss with 22% fewer FLOPs and 15% smaller KV cache.

  • Discovering Millions of Interpretable Features with Sparse Autoencoders cs.LG · 2026-06-25 · unverdicted · none · ref 26

    Trains and releases SAEs for Qwen3-1.7B/4B/8B models with layer-wise coverage and demonstrates causal steering of refusal via selected features.