Abstract: A computing system including a plurality of processing devices configured to execute a Mixture-of-Experts (MoE) layer included in an MoE model. The processing devices are configured to, in each of a plurality of iterations, at each of the processing devices, receive a respective plurality of input tokens. Executing the MoE layer further includes, at each of the processing devices, selecting one or more destination expert sub-models associated with the input tokens. Respective numbers k of expert sub-models selected differ across the iterations. At each of the processing devices, executing the MoE layer further includes conveying the input tokens to the one or more destination expert sub-models. Executing the MoE layer further includes generating one or more respective expert sub-model outputs at the one or more destination expert sub-models. Executing the MoE layer further includes generating and outputting an MoE layer output based on the one or more expert sub-model outputs.

Abstract: A method for preparing a rare earth anisotropic bonded magnetic powder, comprises the following steps: (1) preparing raw powder with RTBH as the main component, wherein, R is Nd or Pr/Nd, and T is a transition metal containing Fe; (2) adding La hydride or Ce hydride and copper powder to the raw powder to form a mixture; (3) subjecting the mixture to atmosphere diffusion heat treatment to give the rare earth anisotropic bonded magnetic powder.

Abstract: A computing system including a plurality of processing devices configured to execute a Mixture-of-Experts (MoE) layer included in an MoE model. The processing devices are configured to execute the MoE layer at least in part by, during a first collective communication phase between the processing devices, splitting each of a plurality of first input tensors along a first dimension to obtain first output tensors. Executing the MoE layer further includes processing the first output tensors at a respective a plurality of expert sub-models to obtain a plurality of second input tensors. Executing the MoE layer further includes, during a second collective communication phase between the processing devices, receiving the second input tensors from the expert sub-models and concatenating the second input tensors along the first dimension to obtain second output tensors. Executing the MoE layer further includes outputting the second output tensors as output of the MoE layer.

Abstract: A computing system is provided, including a plurality of processing devices configured to execute a Mixture-of-Experts (MoE) layer included in an MoE model. The MoE layer includes a plurality of expert sub-models that each have a respective plurality of parameter values. The MoE layer is configured to be switchable between a data parallel mode and an expert-data-model parallel mode without conveying the respective parameter values of the expert sub-models among the plurality of processing devices.

Abstract: The invention discloses a composite rare earth anisotropic bonded magnet and a preparation method thereof. The composite rare earth anisotropic bonded magnet comprises a Nd—Fe—B magnetic powder, a Sm—Fe—N magnetic powder, a binder and an inorganic nano-dispersant. The preparation method comprises steps of preparing a Nd—Fe—B magnetic powder by a HDDR method, preparing a Sm—Fe—N magnetic powder by a powder metallurgy method, mixing the Nd—Fe—B magnetic powder, the Sm—Fe—N magnetic powder, the binder and the inorganic nano-dispersant at a specific ratio to finally obtain the composite rare earth anisotropic bonded magnet. The invention, by adding an inorganic nano-dispersant, enables the full dispersion of the fine Sm—Fe—N powder during the mixing process of the binder, the Nd—Fe—B magnetic powder and the Sm—Fe—N powder, and thus makes the fine Sm—Fe—N powder and the binder evenly coated on the surface of the anisotropic Nd—Fe—B magnetic powder.

Abstract: A computing system including a plurality of processing devices configured to execute a Mixture-of-Experts (MoE) layer. The processing devices are configured to execute the MoE layer at least in part by receiving an input tensor including input tokens. Executing the MoE layer further includes computing a gating function output vector based on the input tensor and computing a sparse encoding of the input tensor and the gating function output vector. The sparse encoding indicates one or more destination expert sub-models. Executing the MoE layer further includes dispatching the input tensor for processing at the one or more destination expert sub-models, and further includes computing an expert output tensor. Executing the MoE layer further includes computing an MoE layer output at least in part by computing a sparse decoding of the expert output tensor. Executing the MoE layer further includes conveying the MoE layer output to an additional computing process.

Abstract: A method of describing a temporal event, including receiving a video sequence of the temporal event, extracting at least one physical characteristic of an at least one occupant within the video sequence, extracting at least one action of the at least one occupant within the video sequence, extracting at least one interaction of the at least one occupant with a secondary occupant within the video sequence, determining a safety level of the temporal event within a vehicle based on at least one of the at least one action and the at least one interaction and describing the at least one physical characteristic of the at least one occupant and at least one of the at least one action and the at least one interaction of the at least one occupant.

Abstract: A method of face anti-spoofing, comprising, receiving a near infra-red facial image, having a near infrared channel, receiving a red-green-blue facial image, having a red channel, a green channel and a blue channel, generating a synthetic three channel image based on the near infrared channel, the red channel, the green channel and the blue channel and training a deep neural network based on the synthetic three channel image.

Abstract: A method of describing a temporal event, including receiving a video sequence of the temporal event, extracting at least one physical characteristic of an at least one occupant within the video sequence, extracting at least one action of the at least one occupant within the video sequence, extracting at least one interaction of the at least one occupant with a secondary occupant within the video sequence, determining a safety level of the temporal event within a vehicle based on at least one of the at least one action and the at least one interaction and describing the at least one physical characteristic of the at least one occupant and at least one of the at least one action and the at least one interaction of the at least one occupant.

Abstract: The present invention discloses an integrated composite overload injection system and a working method thereof. The feeding mechanism preliminarily mixes water with a main agent and an auxiliary agent of an intelligent energy-gathered oil-displacing agent according to the ratio, the outlet of the feeding mechanism is communicated with the input port of the composite overload mechanism through a pipeline, the composite overload mechanism stirs, mixes, dissolves and overload ripens the preliminarily mixed solution to form mother solution, the mother solution is input from the output port of the composite overload mechanism to the inlet of the booster pump through a pipeline, the booster pump injects the boosted mother solution into the mixer, the mixer mixes the mother solution and the diluted high-pressure water and injects it into an oil-water well, and the power shafts of the composite overload mechanism and the booster pump are both driven by the driving mechanism.