Abstract: A method for processing a model operator includes: determining an operator set for model networking, wherein the operator set comprises a plurality of operators; determining a storage amount occupied by an output tensor of each operator in the operator set and a computation time period consumed in a forward computation of each operator in the operator set; and determining a first operator participating in recomputation in a model from the operator set, based on the storage amounts and the computation time periods of the plurality of operators.

Abstract: Provided is a performance optimization method for a model training device, an electronic device, and a storage medium, relating to the fields of deep learning, large model training, and distributed parallel strategies. The method includes: determining communication timing of a current model training device with respect to a target model block at a target sorting position, so as to be able to perform synchronously collective communication with other model training devices of a plurality of model training devices with respect to model blocks at the target sorting position; and performing the collective communication on a backward gradient of the target model block at the communication timing.

Abstract: Cached decoding systems and techniques are described. A system (e.g., decoder) receives an input token (e.g., input vector). The system applies a projection tensor (e.g., a projection matrix) to the input token to generate a feature tensor (e.g., a key tensor or a value tensor). The system processes at least the feature tensor and at least one previous feature tensor using at least one attention calculation to generate an output token. The at least one previous feature tensor is retrieved from a buffer. The at least one previous feature tensor can be stored in the buffer after having been previously calculated based on application of the projection tensor to a previous input token (e.g., from a previous iteration before the iteration in which the input token is received).

Abstract: The present disclosure provides a mixture-of-experts (MoE) model implementation method and system, an electronic device, and a storage medium, and relates to the field of artificial intelligence (AI) such as deep learning and distributed storage. The method includes: constructing a communication group, the communication group including a tensor-parallelism communication group, the tensor-parallelism communication group including at least two computing devices, tensor-parallelism segmentation being adopted for sparse parameters of each of the computing devices in a same tensor-parallelism communication group; and training an MoE model based on the communication group. By use of the solutions of the present disclosure, normal operation of model training can be guaranteed.

Abstract: The present invention discloses a pipeline defect detecting and positioning method based on multi-sensor information fusion, which belongs to the technical field of pipeline non-destructive testing. The method comprises the following steps of performing state segmentation processing on a pipeline according to a pipeline defect detection signal and/or a three-axis attitude signal of a pipeline detection equipment in the pipeline, positioning a pipeline defect signal in the pipeline defect detection signal to a pipeline section in a corresponding state, and calculating a position of the pipeline defect signal in pipeline section in each state according to a running speed, the three-axis attitude signal, and mileage information of the pipeline detection equipment in the pipeline section in the corresponding state. According to the invention, the pipeline is segmented so that the pipeline defect can be preliminarily positioned in the pipeline section in a certain state.

Abstract: The invention discloses an eddy current testing system for nondestructive testing of a pipeline, which belongs to the field of nondestructive testing technology. The eddy current testing system includes a data processing unit, a first signal conditioning unit, a second signal conditioning unit, and a testing probe, wherein the testing probe comprises an excitation coil, a receiving coil, and a passive resonance coil; and the passive resonance coil is arranged between the excitation coil and the receiving coil.

Abstract: The invention discloses an NDT device for pipeline, belonging to the field of NDT, including a mobile carrier, which moves with fluid in pipeline or is moved by an actuator; a probe testing assembly, which includes a testing component installed on the mobile carrier and having a testing element in which testing probe is encapsulated; a data processing unit, a first signal conditioning unit and a second signal conditioning unit, wherein the testing probe includes an excitation coil, a receiving coil and a passive resonance coil between the excitation coil and the receiving coil. For the invention, no more magnetizing treatment device is needed for testing, so that the system volume is greatly reduced, thereby reducing the requirement of the testing system of the invention on the internal cleanliness of a pipeline, improving flexibility of equipment in the pipeline of the testing system, and greatly reducing the system cost.

Abstract: A vehicle power monitoring and auxiliary starting device is provided. The device includes a case, a supercapacitor module, a first circuit, and a second circuit. The at least one supercapacitor module is connected in series through the first circuit, the second circuit includes a balanced charging and controlling circuit, a display module, a relay control circuit and a boost circuit. The first circuit has a positive pole and a negative pole connected with the vehicle battery's positive pole and negative pole. The power of the vehicle battery maintains a certain amount of reserve power, and the balanced charging and controlling circuit of the second circuit monitor the vehicle battery.

Abstract: A smart cooling fan system is disclosed, and comprises temperature sensors, a first cooling fan connected to a main processor of an electronic device, a second cooling fan disposed in a housing case of the electronic device, and a microcontroller. The microcontroller is configured to, based on immediate fan rotation speed, find out a first volumetric flow rate and a second volumetric flow rate by looking up a cubic feet per minute (CFM)-RPM lookup table (LUT), a first pulse width modulation (PWM) LUT and a second PWM LUT, the microcontroller controls the first cooling fan to produce a first airflow having the first volumetric flow rate, and also controls the second cooling fan to produce a second airflow having the second volumetric flow rate.

Abstract: A determining method and apparatus for a neutral system working voltage includes a power system having three phase wires and three capacitor banks. Each bank has one terminal connected to one phase wire, and the other terminals form a neutral point. A phase voltage of each phase wire is periodically sampled, obtaining sampling value groups of real-time phase voltages. Each phase voltage group includes a sampling value of a phase voltage of each phase wire by measuring each phase wire. Unbalance rates between the banks are based on the groups of sampling values. The unbalance rate is a ratio between capacitances of every two phase wires. The isolated neutral system working voltage is based on the unbalance rates. The unbalance rates between banks are based on the sampling values measured in real time. The working voltage is based on the unbalance rates, permitting subsequent operations to be accurately performed.

Abstract: A method for updating encrypted firmware is performed by a terminal device. The method includes the following steps: receiving an encrypted differential upgrade package or a differential upgrade package through a communication channel, wherein when the terminal device receives the encrypted differential upgrade package, the terminal device decrypts the encrypted differential upgrade package to obtain the differential upgrade package; decrypting an encrypted to-be-upgrade firmware of the terminal device to obtain a decrypted to-be-upgraded firmware; generating a decrypted target firmware according to the decrypted to-be-upgraded firmware and the differential upgrade package; and encrypting the decrypted target firmware to generate an encrypted target firmware, and storing the encrypted target firmware in the terminal device.

Abstract: The present invention relates to the field of nucleic acid sequencing. Specifically, provided in the present invention is a nucleotide analogue for sequencing. The nucleotide analogue carries a linker on the base or the ribose ring 3?-OH, and can be used for NGS sequencing. The present invention further relates to a kit containing the nucleotide analogue, and a sequencing method based on the nucleotide analogue.

Abstract: A firmware update method having a self-restoration mechanism is performed by a terminal device. The firmware update method includes the following steps: downloading a differential upgrade package and a differential restoration package through a communication network channel; performing an upgrade operation to current firmware of the terminal device with the differential upgrade package to upgrade the current firmware to a target firmware, wherein the current firmware is changed into incompletely-upgraded firmware after the upgrade operation starts and before the upgrade operation finishes successfully; and verifying whether the upgrade operation fails, and when the upgrade operation fails, performing a restoration operation to the incompletely-upgraded firmware with the differential restoration package to restore the incompletely-upgraded firmware to the current firmware.

Abstract: A processor-implemented method for a neural architecture search (NAS) starts by generating an over-parameterized super network having multiple layers. The super network has multiple operator types. Each of the layers includes a largest super kernel corresponding to a search space. The method also includes performing gradient descent to evolve a largest super kernel to a small kernel corresponding to the search space in order to generate a range of kernel encodings. The method further includes identifying a subset of kernel encodings from the range of kernel encodings, for each layer of the super network, based on the gradient descent. The method determines a set of candidate architectures based on the subset of kernel encodings, each of the candidate architectures having a different model size. The method selects a target model, from the set of architectures, based on meeting hardware specifications, and then applies the target model.

Abstract: A composition including particular amounts of a polyetherimide or a poly(arylene ether sulfone) and an inorganic filler is described herein. Molded samples of the composition can exhibit an advantageous combination of properties, and the composition can be used in various articles. Methods of making the composition are also described.

Abstract: A modified nucleoside or nucleotide, the 3?-OH of the modified nucleoside or nucleotide being reversibly blocked; meanwhile, the present invention also relates to a kit comprising the nucleoside or nucleotide, and a sequencing method based on the nucleoside or nucleotide.

Abstract: A composition including particular amounts of a polyetherimide or a poly(arylene ether sulfone), a second polymer and an inorganic filler is described herein. Molded samples of the composition can exhibit an advantageous combination of properties, and the composition can be used in various articles. Methods of making the composition are also described.

Abstract: A method, device and computer readable medium are provided for determining the closing time of a circuit breaker. In an embodiment, the method includes: sampling a voltage signal of a target phase in the three-phase voltage signal of the circuit breaker; determining a first time period according to a voltage value of the sampled voltage signal, a voltage signal in the first time period including a voltage jump point of the voltage value; finding a voltage jump point of a voltage value in the first time period according to a change in the voltage value of the voltage signal in the first time period; and determining the closing time of a target phase of the circuit breaker according to the found sampling time of the voltage jump point.

Abstract: A node synchronization method, includes: receiving a synchronization message by a first synchronization node in a network; determining whether to update a first local time of the first synchronization node according to the synchronization message by the first synchronization node; updating the first local time according to a synchronization time of the synchronization message by the first synchronization node when determining to update the first local time; and updating the synchronization time of the received synchronization message with the current first local time and forwarding the updated synchronization message by the first synchronization node.