Abstract: A field programmable gate array (FPGA) may be used for inference of a trained deep neural network (DNN). The trained DNN may comprise a set of parameters and the FPGA may have a first precision configuration defining first number representations of the set of parameters. The FPGA may determine different precision configurations of the trained DNN. A precision configuration of the precision configurations may define second number representations of a subset of the set of parameters. For each precision configuration of the determined precision configurations a bitstream file may be provided. The bitstream files may be stored so that the FPGA may be programmed using one of the stored bitstream files for inference of the trained DNN.

Abstract: The invention is notably directed to a method, computer program product, and computer system for running software inside containers. The method relies on a computerized system that includes a composable disaggregated infrastructure, in addition to general-purpose hardware. The computerized system is configured to dynamically allocate computerized resources, which include both general resources and specialized resources. The former are enabled by the general-purpose hardware, while the latter are enabled by specialized network-attached hardware components of the composable disaggregated infrastructure. The method maintains a table capturing specializations of the specialized network-attached hardware components. At runtime, software is run inside each container by executing corresponding functions.

Abstract: Field-programmable gate array and method to implement an artificial neural network. A trained model of the neural network is processed, in which weights are defined in a floating-point format, to quantize each set of weights to a respective reduced-precision format in dependence on effect of quantization on accuracy of the model. For each set of weights, a partitioning scheme is defined for a set of block memories of the apparatus such that a plurality k of those weights can be stored in each addressable location of the set of memories, wherein k differs for different sets of weights. The apparatus can be programmed to implement the neural network such that weights in each set are persistently stored in a set of block memories partitioned according to the partitioning scheme for that set of weights.

Abstract: The invention is notably directed to a method, computer program product, and computer system for running software inside containers. The method relies on a computerized system that includes a composable disaggregated infrastructure, in addition to general-purpose hardware. The computerized system is configured to dynamically allocate computerized resources, which include both general resources and specialized resources. The former are enabled by the general-purpose hardware, while the latter are enabled by specialized network-attached hardware components of the composable disaggregated infrastructure. The method maintains a table capturing specializations of the specialized network-attached hardware components. At runtime, software is run inside each container by executing corresponding functions.

Abstract: The present disclosure relates to a messaging method for a hardware acceleration system. The method includes determining exchange message types to be exchanged with a hardware accelerator in accordance with an application performed by the hardware acceleration system. The exchange message types indicate a number of variables, and a type of the variables, of the messages. The method also includes selecting schemas from a schema database. The message type schemas indicates a precision representation of variables of messages associated with the schema. The selected schemas correspond to the determined exchange message types. Further, the method includes configuring a serial interface of the hardware accelerator in accordance with the selected schemas, to enable a message exchange including the messages.

Abstract: The present disclosure relates to a hardware accelerator for executing a computation task composed of a set of operations. The hardware accelerator comprises a controller and a set of computation units. Each computation unit of the set of computation units is configured to receive input data of an operation of the set of operations and to perform the operation, wherein the input data is represented with a distinct bit length associated with each computation unit. The controller is configured to receive the input data represented with a certain bit length of the bit lengths and to select one of the set of computation units that can deliver a valid result and that is associated with a bit length smaller than or equal to the certain bit length.

Abstract: The present disclosure relates to a messaging method for a hardware acceleration system. The method includes determining exchange message types to be exchanged with a hardware accelerator in accordance with an application performed by the hardware acceleration system. The exchange message types indicate a number of variables, and a type of the variables, of the messages. The method also includes selecting schemas from a schema database. The message type schemas indicates a precision representation of variables of messages associated with the schema. The selected schemas correspond to the determined exchange message types. Further, the method includes configuring a serial interface of the hardware accelerator in accordance with the selected schemas, to enable a message exchange including the messages.

Abstract: A field programmable gate array (FPGA) may be used for inference of a trained deep neural network (DNN). The trained DNN may comprise a set of parameters and the FPGA may have a first precision configuration defining first number representations of the set of parameters. The FPGA may determine different precision configurations of the trained DNN. A precision configuration of the precision configurations may define second number representations of a subset of the set of parameters. For each precision configuration of the determined precision configurations a bitstream file may be provided. The bitstream files may be stored so that the FPGA may be programmed using one of the stored bitstream files for inference of the trained DNN.

Abstract: Field-programmable gate array and method to implement an artificial neural network. A trained model of the neural network is processed, in which weights are defined in a floating-point format, to quantize each set of weights to a respective reduced-precision format in dependence on effect of quantization on accuracy of the model. For each set of weights, a partitioning scheme is defined for a set of block memories of the apparatus such that a plurality k of those weights can be stored in each addressable location of the set of memories, wherein k differs for different sets of weights. The apparatus can be programmed to implement the neural network such that weights in each set are persistently stored in a set of block memories partitioned according to the partitioning scheme for that set of weights.