5results about How to "Reduce calculation steps" patented technology

The invention provides a method for predicting hydration expansibility of a compacted bentonite material, which comprises the following steps of: generalizing soil particles into a spherical shape, and dividing a compacted bentonite structure into a characteristic structure element consisting of four particles based on a space accumulation configuration; according to constraint conditions of apparent density, elastic deformation energy minimization and structural entropy maximization of the soil body structure, the distribution probability of each characteristic structure element is solved; a Hertz contact model is adopted to calculate the expansive force between particles in the characteristic structure elements; calculating the equivalent expansive force of each characteristic structure element through a finite element method; based on the Hill lemma, the macroscopic hydration expansibility of the compacted bentonite structure is predicted. According to the method, the prediction process is simple, the calculated amount is moderate, the hydration expansibility of the compacted bentonite structures with different dry densities can be accurately and theoretically predicted, and theoretical guidance can be provided for design and prediction of the hydration expansibility of the compacted bentonite structures in the environments of refuse landfills, nuclear waste repositories and the like.

The application discloses a low-bit matrix calculation device with a floating-point scaling factor and a calculation method thereof. The device comprises a scaling factor processing module, an input module, a matrix calculation module, a floating-point accumulation module and an output module. The method first groups and quantizes weight and eigenvalue matrices to obtain quantization values and floating-point scaling factors; then, the scaling factor product is calculated in advance with a single accumulation length as a grouping unit; then, matrix multiplication and accumulation operation is performed, the result is multiplied by the scaling factor product after being converted into an fp16 or bfp16 format, and then accumulated into a final result; and finally, the result is output after being dequantized. The application simplifies the group-wise calculation process, saves hardware resources, guarantees calculation accuracy, meets different scene requirements and improves calculation efficiency by sharing the scaling factor product and using fp16 or bfp16 as an accumulator.

The invention discloses a self-adaptive question answering method based on a large language model and an intelligent large language model. The method comprises the following steps: receiving a query text input by a user; analyzing the query text by using a preset query analysis model to determine the complexity of the query text; if the complexity is a first category, the query text is directly input into a large language model to generate an answer, and the first category corresponds to a preset low-complexity query; if the complexity is a second category, the retrieval process is executed, a reference document related to the query text is obtained from a preset data source, the query text and the reference document are jointly input into the large language model to generate answers, and the second category corresponds to preset high-complexity query. According to the method, query analysis is introduced, so that shunting processing, simple direct answering and answering after complex retrieval on queries with different complexities are realized. The self-adaptive mechanism maximizes the operation efficiency of the system while ensuring the answer quality.

The invention relates to the technical field of storage, and discloses a data processing method and system, the method is applied to storage equipment, according to the method, copying operation is carried out by obtaining a data copying instruction, a verification value of original data corresponding to a logic data block is calculated while decompression processing is carried out on the logic data block, and based on the verification value and a source logic address, the original data corresponding to the logic data block is obtained. And generating second protection information, comparing the first protection information with the second protection information according to the protection control information, if the comparison is passed, generating third protection information based on the verification value and the destination logic address, combining the logic data block and the third protection information to obtain first target data, and taking the first target data as data corresponding to the destination address. And the preset process is skipped to store the first target data in the storage device, so that the calculation step of data decompression can be reduced, and the data processing efficiency is improved.

The invention relates to the technical field of pixel positioning, in particular to an IC chip sub-pixel precision positioning method and system corresponding to secondary matching, and the method comprises the steps: obtaining an initial image of an IC chip and a reference template image corresponding to the initial image; generating a rough positioning frame according to the initial image and the reference template image; cutting the reference template image by using the rough positioning frame to obtain a first target image area; calculating a gradient feature map of the first target image region; performing first matching in the reference template image according to the gradient feature map to obtain a corresponding point set and generate an initial offset; and updating a coordinate mapping table of the initial image by using the initial offset. According to the method, through back projection and real-time image slice extraction, the similarity between the positioning result and the standard template can be immediately verified after positioning, and the reliability of final positioning is ensured.