Abstract: A method of performing automatic tuning on a deep learning model includes: utilizing an instruction-based learned cost model to estimate a first type of operational performance metrics based on a tuned configuration of layer fusion and tensor tiling; utilizing statistical data gathered during a compilation process of the deep learning model to determine a second type of operational performance metrics based on the tuned configuration of layer fusion and tensor tiling; performing an auto-tuning process to obtain a plurality of optimal configurations based on the first type of operational performance metrics and the second type of operational performance metrics; and configure the deep learning model according to one of the plurality of optimal configurations.

Abstract: Aspects of the present disclosure provide a method for training a predictor that predicts performance of a plurality of machine learning (ML) models on platforms. For example, the method can include converting each of the ML models into a plurality of instructions or the instructions and a plurality of intermediate representations (IRs). The method can also include simulating execution of the instructions corresponding to each of the ML models on a platform and generating instruction performance reports. Each of the instruction performance reports can be associated with performance of the instructions corresponding to one of the ML models that are executed on the platform. The method can also include training the predictor with the instructions or the IRs as learning features and the instruction performance reports as learning labels, compiling the predictor into a library file, and storing the library file in a storage device.

Abstract: Floating-point numbers are compressed for neural network computations. A compressor receives multiple operands, each operand having a floating-point representation of a sign bit, an exponent, and a fraction. The compressor re-orders the operands into a first sequence of consecutive sign bits, a second sequence of consecutive exponents, and a third sequence of consecutive fractions. The compressor then compresses the first sequence, the second sequence, and the third sequence to remove at least duplicate exponents. As a result, the compressor can losslessly generate a compressed data sequence.

Abstract: A temperature prediction system and method thereof use sensors to sense the temperature at a plurality of times, thereby calculating an actual temperature difference between the current time and the previous time. According to the result of actual temperature difference, an adaptive filter is used to predict a predicted temperature difference at a future time and to add the temperature of the current time and the predicted temperature difference to render a predicted temperature. By calculating the difference between the temperature at the next time and the predicted temperature at the next time, the filter parameter of the adaptive filter is adjusted. This mechanism helps improving the accuracy in temperature predictions.