16 results about "Second quantization" patented technology

The invention discloses a model quantification method and device, equipment, a storage medium and a computer program product, and relates to the field of model quantification, and the method comprises the steps: carrying out the training and quantification of a full-precision model disposed on edge equipment, and obtaining a first quantification model which comprises a first layer set; according to a sensitivity index and training cost corresponding to each model layer in the first layer set, a target training layer is selected from the first layer set, and the sensitivity index is used for representing the influence degree of the quantization operation on the model layer; performing quantitative perception training on the target training layer in the first quantitative model to obtain a second quantitative model, the second quantitative model including a second layer set; and performing bit width allocation on each model layer in the second layer set to obtain a target quantitative model. According to the invention, while the precision of the quantitative model is ensured, the cost required by quantitative training is reduced, so that the contradiction between insufficient quantitative precision after training and too high quantitative perception training cost in the prior art is solved.

The invention provides a video processing method, receiving end equipment and transmitting end equipment, and relates to the technical field of video transmission. The video processing method comprises the following steps: in response to a condition that a first health degree score of a network at a current moment is less than or equal to a first threshold value, discarding at least part of video coding frames in a buffer area, and sending a request for obtaining an instant decoding refreshing frame to a sending end; in response to receiving an instant decoding refresh frame, decoding the instant decoding refresh frame, the instant decoding refresh frame being obtained by encoding a target region in the original image frame based on a first quantization parameter and encoding a background region in the original image frame based on a second quantization parameter, the first quantization parameter does not change along with the change of the first health degree score, and the second quantization parameter is negatively correlated with the first health degree score. According to the method and the device, the delay of the video picture can be reduced, and the information of the target area can be clearly reserved, so that the stability, accuracy and safety of remote operation can be improved.

The present disclosure relates to an LCEVC-based enhancement layer coding method, device, equipment and medium, applied to the technical field of video coding, which can improve the coding quality of video coding. The method comprises: in the case of using temporal prediction for a to-be-coded enhancement layer sequence, obtaining the code rate of the current frame base layer corresponding to the current frame enhancement layer and the average code rate of the C frame base layer before the current frame base layer, the to-be-coded enhancement layer sequence comprising a plurality of enhancement layers, and the current frame enhancement layer being any frame in the plurality of enhancement layers; in the case that the code rate of the current frame base layer is less than or equal to the average code rate, coding the current frame enhancement layer based on a first quantization parameter; in the case that the code rate of the current frame base layer is greater than the average code rate, updating the quantization parameter of the current frame enhancement layer from the first quantization parameter to a second quantization parameter, and coding the current frame enhancement layer based on the second quantization parameter, the second quantization parameter being less than the first quantization parameter.

The application provides a binarization network training method and device based on gradient quantization, which comprises the following steps: in a full-precision model full-quantization training process taking a sample image as a training sample and taking a class label of the sample image as a training label, quantizing weights and input activation values of the full-precision model to obtain first quantization data with a target bit width; in a case of outputting activation value gradients by performing back propagation on the first quantization data with the target bit width, quantizing the activation value gradients to obtain second quantization data with the target bit width; and updating the activation value gradients and weight gradients according to the first quantization data and the second quantization data to obtain a trained binarization network. The method realizes binarization gradient training, effectively reduces the training power consumption of the model and the requirements for memory and computing power, and further improves the training efficiency and inference performance of the classification model in a complex scene.

The invention relates to the technical field of computer systems, and provides a fine-grained quantization matrix multiplication device and method based on systolic arrays, and the device comprises a first systolic array which is used for receiving element streams of a first quantization factor matrix and a second quantization factor matrix, and carrying out the fusion calculation of quantization factors according to the element streams of the quantization factors, generating a quantization factor product; the second systolic array is used for receiving element streams of the first data matrix and the second data matrix, and performing multiplication and accumulation calculation according to the element streams of the matrixes; each first processing unit in the first systolic array is configured to send a quantization factor product obtained by calculation to a corresponding anchor processing unit in the second systolic array according to a mapping proportion; and the anchor processing unit is configured to receive the quantization factor product and serve as a propagation source point, the quantization factor product is propagated in the second systolic array in the row direction and the column direction within the quantization block size range corresponding to the quantization factor product, and the calculation efficiency is improved.

The application discloses a language task processing method and electronic equipment, and relates to the technical field of artificial intelligence. The method comprises the following steps: selecting target weight parameters according to task calibration data and an activation function type of an original task processing model; determining a fitting processing result of a quantization parameter according to each target weight parameter, a maximum weight parameter and the task calibration data; determining a first quantization error function and a second quantization error function carrying a scaling factor according to the fitting processing result, and obtaining a value of the scaling factor by minimizing the ratio of the two; performing scaling processing on each target weight parameter by using the scaling factor, and quantizing each scaling weight parameter based on a quantization precision parameter to obtain a task processing model used for executing a to-be-processed language task. The application can solve the problem that related technologies cannot guarantee high accuracy of a language task processing result on the basis of reducing resources used in a language task execution process, and effectively improve language task processing precision.

A video decoding method according to the present invention includes: decoding a bitstream, to obtain P quantization coefficients of a current region; determining, according to the P quantization coefficients, parity of a quantization coefficient whose parity is hidden of a first quantization coefficient; determining a target context model corresponding to the first quantization coefficient, and decoding the first quantization coefficient which is encoded based on context using the target context model, to obtain a decoded first quantization coefficient; and determining the second quantization coefficient according to the parity of the quantization coefficient whose parity is hidden and the decoded first quantization coefficient.

The invention provides a method for determining a quantization strategy, a model quantization system and related equipment, and the method comprises the following steps: carrying out the free combination of a to-be-quantized object supported by a model and a data type which can be quantized, obtaining a plurality of different quantization strategies, determining the sensitivities of different quantization strategies based on a sensitivity algorithm, and carrying out the quantification of the to-be-quantized object. The sensitivity is used for indicating the influence of the quantization strategies on the precision of the to-be-quantized model, determining a first quantization strategy according to the sensitivity of each quantization strategy, and if the quantization evaluation of the first quantization strategy is qualified, using the first quantization strategy to quantize the model, so that the whole quantization process does not need quantization search; and if the quantization evaluation is unqualified, the precision of part of the to-be-quantized objects in the first quantization strategy is improved in a quantization rollback mode until the second quantization strategy which is evaluated to be qualified is obtained, so that compared with a traditional mode of performing quantization search based on large-scale parameters, the search space can be reduced, and the model quantization efficiency can be improved.

The invention discloses a large model quantification method and device, computer equipment and a storage medium, which can be applied to various scenes such as financial investment management, automatic driving and face recognition, and the method comprises the following steps: obtaining a first large model; based on the type of the first large model, obtaining a first quantification method corresponding to the first large model; performing quantification processing on the first large model by using a first quantification method to obtain a second large model; and performing quantization processing on the second large model by using a second quantization method to obtain a target large model. According to the method, the first quantification method is utilized to quantify the first large model, the reasoning speed of the obtained second large model is improved by improving the quantification precision in the quantification process, and then the second quantification method is adopted to quantify the second large model, so that the occupation of a reasoning video memory is reduced, and the reasoning efficiency is improved. The reasoning speed of the obtained target large model is further improved, and the reasoning speed is improved while the small precision loss of the large model is kept.

A device for decoding a bitstream of encoded mesh data is configured to in response to determining that first transform parameters are overridden, infer a value of a second syntax element to be equal to a first value indicating that second transform parameters are present in the bitstream of encoded mesh data; in response to the value of the second syntax element being equal to the first value, determine the second transform parameters from a second syntax structure; in response to determining that the first quantization parameters are overridden, infer a value of a third syntax element to be equal to a first value indicating that second quantization parameters are present in the bitstream of encoded mesh data; and in response to the value of the third syntax element being equal to the first value, determine second quantization parameters from a third syntax structure.

A processor-implemented method including obtaining first data in an integer (INT) form by performing a first quantization on input activation data, applying a block-wise orthogonal matrix to the first data to obtain second data, and performing a second quantization on the second data to obtain third data, the block-wise orthogonal matrix including a plurality of orthogonal matrices arranged diagonally.

The application discloses a TDC circuit and module for SRAM in-memory computing quantization circuit. The TDC circuit comprises a plurality of delay units, an asynchronous counter, a plurality of flip-flops and an encoding circuit. The plurality of delay units are connected in series and connected in a ring structure. The asynchronous counter is used for counting the number of laps of the first pulse signal in the ring structure and outputting a first quantization result. All flip-flops receive a second pulse signal, latch the output state of each delay unit in the ring structure under the triggering of the signal, and output a multi-bit latching result. The encoding circuit is used for encoding the multi-bit latching result and outputting a second quantization result. The first quantization result and the second quantization result jointly constitute the quantization result of the TDC circuit to the time difference between the first pulse signal and the second pulse signal. The time domain quantization method can better adapt to the development of integrated circuit technology, and achieve a better balance among power consumption, speed and accuracy.

The invention discloses a large language model quantification method and device, electronic equipment and a storage medium. The method comprises the following steps: adaptively determining respective target quantification granularity for each quantifiable layer of a specified large language model; based on each target quantization granularity, performing grouping quantization on the large language model under the first quantization precision to obtain a first quantization model; performing reasoning precision verification on the first quantitative model on the target hardware equipment by utilizing a preset verification data set; if the reasoning precision does not reach the preset threshold value, identifying the sensitive layer, and adjusting the quantization precision of the sensitive layer from the first quantization precision to a second quantization precision to obtain a second quantization model; and performing reasoning precision verification on the second quantification model, iteratively executing sensitive layer identification and quantification precision adjustment operation according to a verification result until the reasoning precision reaches a preset threshold value, and outputting a final quantification model. According to the method, the reasoning precision and the hardware efficiency of the large language model in actual deployment can be considered.

An example apparatus includes interface circuitry, machine-readable instructions, and at least one processor circuit to be programmed by the machine-readable instructions to quantize a group of input data using a first quantization algorithm to form a first group of quantized data, quantize the group of input data using a second quantization algorithm to form a second group of quantized data, quantize the group of input data using a third quantization algorithm to form a third group of quantized data, the third quantization algorithm different from the first quantization algorithm and the second quantization algorithm, determine a first quantization error of the first group of quantized data, determine a second quantization error of the second group of quantized data, and determine a third quantization error of the third group of quantized data.

The invention discloses a model quantification method and device, and belongs to the technical field of data processing. The method comprises the following steps: quantizing a weight matrix of a large language model to obtain a first quantized matrix; determining a low-rank compensation matrix according to the weight matrix and the first quantization matrix; and compensating the first quantization matrix according to the parameter value of the elastic connection parameter and the low-rank compensation matrix to obtain a second quantization matrix.