AI Architecture & Systems

Dynamic Shape and Runtime Specialization

3.10.1Symbolic Shape, Shape Constraint, and Shape Polymorphism#

3.10.2Shape Bucketing, Padding, and Multi-Versioning#

3.10.3Autotuning Cache and Compilation Cache#

3.10.4Dynamic Sparsity, Dynamic Routing, and Ragged Workloads#

3.10.5Compilation Latency, Cold Start, and Steady-State Performance#