AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Know What You Need: Efficient Activation Checkpointing

Danyal Ahmed, Edouard Oyallon

NeurIPS 2026 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

The paper targets the compute overhead of activation checkpointing for transformer linear layers during LLM training (abstract truncated, so the proposed method is not specified).

Problem

LLM training is memory-bound; activation checkpointing saves memory by recomputing activations in backprop, but rematerializing linear layers of transformers is prohibitively expensive in FLOPs.

Approach

Abstract truncated before the proposed approach is stated; the paper's specific mechanism for efficient checkpointing of linear layers cannot be determined from the provided text.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use Top-k selection instead of random selection (2/3 kept) (ablation)KEAP on FFN (Llama-488M) · FineWeb, token-levelperplexity: 12.84 -> 12.23Improved
use Top-k instead of variance-optimal Opt selection (2/3 kept) (ablation)KEAP on attention+FFN · FineWeb, token-levelperplexity: 12.81 -> 12.33Improved
use Opt instead of Rand selection (ablation)KEAP on FFN · FineWeb, token-level, 1/3perplexity: 14.70 -> 13.66Improved
use token-level instead of batch-level granularity (ablation)KEAP Top-2/3 on attention+FFN · FineWebperplexity: 12.85 -> 12.33Improved
use KEAP Top-2/3 on FFN instead of compute-matched smaller modelLlama-488M · FineWeb, token-levelperplexity: 12.44 -> 12.23Improved
use KEAP Top-2/3 on attention instead of compute-matched smaller modelLlama-488M · FineWeb, token-levelperplexity: 12.17 vs 12.17No clear effect
use batch-level KEAP instead of compute-matched smaller modelLlama-488M · FineWeb, batch-levelperplexity: 12.44 -> 12.61Hurt
recompute only 2/3 of FFN components (KEAP Top-2/3)Llama-488M · FineWebperplexity: 12.03 -> 12.23Hurt
reduce kept fraction from 2/3 to 1/3 (Top-k) (ablation)KEAP on attention+FFN · FineWeb, token-levelperplexity: 12.33 -> 12.90Hurt
apply KEAP to FFN rather than to attention+FFN (ablation)KEAP Top-2/3 · FineWeb, token-levelperplexity: 12.33 -> 12.23Improved
scale to 953M model with Top-2/3 FFNKEAP · FineWebperplexity: 1.66% (488M) -> 1.02% (953M) degradationImproved
batch-level FFN KEAP vs full ACKEAP Top-1/3 (953M, H100) · training step timestep time: -26% (Top-1/3), -18% (Top-2/3)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy