3results about How to "Accuracy unchanged" patented technology

This invention discloses a method and apparatus for accelerating distributed training of large AI models based on model prediction. The method includes: constructing a hierarchical parameter dataset; training a prediction model for predicting model parameters of a target model using an evaluation method based on prediction results; computing nodes updating the model parameters of some layers of the target model and then pushing them to a parameter server; on the parameter server, aggregating the received model parameters and predicting the model parameters of the remaining layers of the target model using the prediction model; concatenating the predicted and aggregated model parameters into a complete model parameter set and pushing it to all computing nodes; and replacing the local model parameters with the received complete model parameter set. The prediction model adopts a "convolution + channel attention mechanism + pooling" structure. This invention can significantly reduce communication volume without introducing a large amount of computational overhead while maintaining the accuracy of the target model.

The utility model provides an engine connecting rod manufacturing die convenient to demould, and relates to the field of manufacturing dies, the engine connecting rod manufacturing die convenient to demould comprises a processing box, a demoulding assembly is installed on the upper surface of the processing box, two clamping assemblies are installed on the inner wall of the processing box, and a storage assembly is installed on the front face of the processing box. An injection molding assembly is mounted on the upper surface of the machining box. The lower mold in the device is mounted in the rotating demolding assembly, so that after a workpiece is shaped and cooled, the two connecting rods and the rotating plates are driven to rotate through the driving motor, and the lower mold mounted between the two rotating plates can be turned over; by means of the structure, a workpiece in the lower mold can fall down due to the gravity of the workpiece, so that the workpiece can be demolded conveniently, the precision in the lower mold is not changed in the casting or demolding process, and therefore it can be guaranteed that the workpiece is demolded conveniently, and meanwhile the machining quality is guaranteed.

This disclosure relates to a method and electronic device for training an optical proximity correction model. The method includes: selecting a first number of data points as a training subset from a plurality of data points corresponding to measurement points in a layout, using a weighted random strategy, based on multiple index values ​​of the data points; generating updated index values ​​of the data points based on training the model using the training subset; selecting the first number of data points as an updated training subset based on the updated index values ​​of the data points using the weighted random strategy; and performing subsequent iterative training on the model based on the updated training subset. The method of this disclosure can effectively improve the training efficiency of the optical proximity correction model.