Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Affiliation:
3
0
2
Equivariant tensor products can now saturate over 80% of peak memory bandwidth across both H100 GPUs and TPUv6e, beating cuEquivariance by up to 34% within an entirely open-source JAX framework.
Open-source mlip v2 slashes the computational cost of atomistic simulations while maintaining near ab initio accuracy, thanks to a high-performance backend and scalable Mixture-of-Experts architecture.
Unlock a 9x speedup in Clebsch-Gordan tensor product simulations by using the Vector Spherical Tensor Product with these newly derived integral formulas.