AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

POET-X: Memory-efficient LLM Training by Scaling Orthogonal Transformation

Zeju Qiu, Lixin Liu, Adrian Weller, Han Shi, Weiyang Liu

ICML 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. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use POET-X_fast instead of original POETPOET · Single-layer latencyfwd+bwd time (ms): 10.59ms to 1.38msImproved
Use gradient checkpointing (POET-X_mem) instead of standard autogradPOET-X · Single-layer latencyfwd+bwd time (ms): 10.59ms to 1.89msImproved
Add permutation reduction (merge 2 of 4 permutations into W) (ablation)POET-X · Permutation benchmark, block 256, uncompiled, hidden 4096time (ms): 15.858 -> 12.017 (1.32x)Improved
Add permutation reduction (merge 2 of 4 permutations into W) (ablation)POET-X · Permutation benchmark, block 512, compiled, hidden 16384time (ms): 96.798 -> 86.380 (1.12x)Improved
Use custom CUDA permutation operator instead of PyTorch-native (ablation)POET-X · Permutation, hidden dim 2048time (ms): 1.621 -> 0.086 (18.75x)Improved
Use custom CUDA permutation operator instead of PyTorch-native (ablation)POET-X · Permutation, hidden dim 16384time (ms): 13.246 -> 0.946 (14.00x)Improved
Use customized batch-parallel block-diagonal multiplication instead of PyTorch (ablation)POET-X · Seq length 2048time (ms): 0.790 -> 0.334 (2.37x)Improved
Store only half of skew-symmetric matrices in CNP (memory reduction) (ablation)POET-X · Seq length 2048memory (MB): 280 -> 192 (31.43%)Improved
Use Triton CNP kernels instead of PyTorch-native, N=64 (ablation)POET-X · CNP benchmark N=64time (ms): 0.316 -> 0.107 (2.96x)Improved
Use POET-X_b=512 instead of b=256 (larger block size)POET-X · C4 Llama-3B val PPLvalidation perplexity: 12.76 -> 12.05Improved
Use POET-XQ_b=512 instead of b=256POET-XQ · C4 Llama-3B quantized val PPLvalidation perplexity: 16.21 -> 14.78Improved
Replace AdamW with POET-X b=512Llama-3B · C4 Lmax=256validation perplexity: 12.69 -> 12.05Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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