Facebook, Inc.

Meta Platforms, Inc.

Roman Dzhabarov

Jong Soo Park

Abdulkadir Utku Diril

Mikhail Smelyanskiy

Nadav Rotem

Zhaoxia Deng

James Wesley Hegeman

Bertrand Allen Maher

Hector Yuen

Nadathur Rajagopalan Satish

Nicholas Gibson

Jordan Samuel Fix

Garret Ray Catron

A computer program compiled for a machine learning accelerator hardware and associated with a default input data size is received. An execution of an operation of the computer program is initiated. It is identified that a data size of an input data of the operation is smaller than the default input data size. The smaller data size of the input data of the operation rather than the default input data size is caused to be transferred to the machine learning accelerator hardware for the input data of the operation.

Reducing data transfer to machine learning accelerator hardware

The disclosed computer-implemented method may include receiving an input value and a floating-point scaling factor and determining (1) an integer scaling factor based on the floating-point scaling factor, (2) a pre-scaling adjustment value representative of a number of places by which to shift a binary representation of the input value prior to a scaling operation, and (3) a post-scaling adjustment value representative of a number of places by which to shift the binary representation of the input value following the scaling operation. The method may further include calculating a scaled result value by (1) shifting rightwards the binary representation of the input value by the pre-scaling adjustment value, (2) scaling the shifted binary representation of the input value by the integer scaling factor, and (3) shifting rightwards the shifted and scaled binary value by the post-scaling adjustment value. Various other methods, systems, and computer-readable media are also disclosed.

Systems and methods for efficient scaling of quantized integers

Growing community of inventors

Redwood City, CA, United States of America

Roman Dzhabarov