Research map / Architectures and efficiency / RNNs and normalization
E2Former-V2: On-the-Fly Equivariant Attention with Linear Activation Memory
Lin Huang, Chengxiang Huang, Ziang Wang, Yiyue Du, Chu Wang, Haocheng Lu, Yunyang Li, Xiaoli LIU, Arthur JIANG, Jia Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add the fused Triton on-the-fly attention kernel (ablation) | E2Former-V2 conservative · End-to-end, 5k atoms | samples/s: 1.40 -> 3.31 | Improved |
| add EAAS (ablation) | E2Former-V2 conservative · End-to-end, 5k atoms | samples/s: 1.40 -> 1.55 | Improved |
| add EAAS on top of the Triton kernel (ablation) | E2Former-V2 conservative · End-to-end, 5k atoms | samples/s: 3.31 -> 4.24 | Improved |
| add EAAS on top of the Triton kernel (ablation) | E2Former-V2 conservative · End-to-end, 10k atoms | samples/s: 1.65 -> 2.12 | Improved |
| add the Triton kernel to avoid out-of-memory (ablation) | E2Former-V2 conservative · End-to-end, 10k atoms | samples/s: OOM -> 1.65 | Improved |
| replace OpenEquivariance with EAAS SO(2) | tensor product kernel · Forward runtime, 32k TPs | ms: 6.20 -> 2.25 | Improved |
| replace cuEquivariance with EAAS SO(2) | tensor product kernel · Forward runtime, 32k TPs | ms: 3.06 -> 2.25 | Improved |
| use conservative instead of direct force head | E2Former-V2 · SPICE (All) | Force MAE: 8.27 -> 7.05 | Improved |
| use conservative instead of direct force head | E2Former-V2 · OMol25 Total | Energy MAE: 1.24 -> 1.16 | Improved |
| replace MACE Large with E2V2-Cons | SPICE Dimers · SPICE (Dimers) | Energy MAE: 0.54 -> 0.18 | Improved |
| replace UMA-S with E2V2-Cons | OMol25 · OMol25 Total | Energy MAE: 3.62 -> 1.16 | Improved |
| replace E2Former-V1 with E2Former-V2 conservative | inference throughput · 10,000 atoms conservative | steps/s: 0.53 -> 2.12 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MACE Medium — MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields (NeurIPS 2022) · better on 4 of 4 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
- E2Former: An Efficient and Equivariant Transformer with Linear-Scaling Tensor Products (NeurIPS 2025 · similarity 0.91)
- EquiformerV2: Improved Equivariant Transformer for Scaling to Higher-Degree Representations (ICLR 2024 · similarity 0.90)
- Compact SO(3) Equivariant Atomistic Foundation Models via Structural Pruning (NeurIPS 2026 · similarity 0.88)
- Equivariant Graph Neural Operator for Modeling 3D Dynamics (ICML 2024 · similarity 0.88)
- The Price of Freedom: Exploring Expressivity and Runtime Tradeoffs in Equivariant Tensor Products (ICML 2025 · similarity 0.86)
- Efficient and Scalable Density Functional Theory Hamiltonian Prediction through Adaptive Sparsity (ICML 2025 · similarity 0.86)
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.
