coreweave / tensorizer

Module, Model, and Tensor Serialization/Deserialization
MIT License
180 stars 25 forks source link

perf: Optimize Serialization #43

Closed Eta0 closed 1 year ago

Eta0 commented 1 year ago

Serialization Optimization

This change optimizes the serialization code to multithread and overlap hash computations, reduce redundant reads, and write to disk more efficiently.

This provides a ~2x speedup on serialization tests when serializing gpt-neo-2.7B.