Expert Parallelism
Mixture-of-Experts routing and communication benchmarks
View on GitHubOverview
Benchmarks for evaluating all-to-all communication patterns used in Mixture-of-Experts (MoE) model architectures.
Mixture-of-Experts routing and communication benchmarks
View on GitHubBenchmarks for evaluating all-to-all communication patterns used in Mixture-of-Experts (MoE) model architectures.