30 results about "Channel quantization" patented technology

The invention provides a quantification method for a large language model, and the method comprises the steps: for each linear layer, calculating the weight quantification sensitivity of the linear layer, and determining the high-precision bit width ratio of each linear layer according to the weight quantification sensitivity of each linear layer and a target precision mixing ratio; for a plurality of channels of each linear layer, based on the input activation of the linear layer, obtaining the channel quantization sensitivity of each channel in the linear layer; sequencing a plurality of channels in the linear layer according to the channel quantization sensitivity, clustering the sequenced channels in combination with the high-precision bit width ratio of the linear layer, and distributing quantization bit widths with different precisions for weight parameters corresponding to the channels of different clusters; and rearranging a plurality of channels of the same cluster according to distribution similarity indexes, grouping the rearranged channels, and synchronously quantizing weight parameters of each group of channels according to the allocated quantization bit width to obtain a quantized large language model.

The invention provides an information transmission method and device and a storage medium. The information transmission method applied to first communication equipment comprises the following steps: performing channel measurement by adopting a received reference signal to obtain a first channel measurement result; converting the first channel measurement result according to a preprocessing mode to obtain a second channel measurement result; quantizing and feeding back channel information according to the second channel measurement result; and sending the channel quantization feedback indication information to the second communication equipment.

The invention discloses a dual-channel collaborative unknown signal reconstruction error control method, and belongs to the technical field of wireless communication. The method solves the problem that the existing method cannot give consideration to both the signal reconstruction precision and the information content contained in the signal. The invention provides a scheme of combining dual-channel collaborative quantization and error control of a receiving end, a dual-channel quantization mechanism realizes information complementation through different quantization characteristics, single-channel error accumulation is inhibited, and the signal detail retention capability is improved. And in combination with an interpolation processing method, estimating and compensating quantization information, optimizing reconstruction output, and completing high-fidelity signal recovery. According to the method, the reconstruction precision and fidelity of unknown signals can be remarkably improved, the robustness and adaptability of the system under the condition of lack of prior information are ensured, meanwhile, extreme dependence on the performance of a single-channel quantizer can be reduced, and then efficient balance between computing resources and reconstruction quality is achieved. The method can be applied to the field of wireless communication.

Provided in the present application are a channel information feedback method, an electronic device, and a storage medium. The method comprises: acquiring channel parameters of a measurement channel; determining a codebook set for channel quantization feedback of the channel; selecting from the codebook set a codeword Ws used for channel quantization characterization of a sub-band s; and determining an indication parameter of the codeword Ws, and feeding the indication parameter back to a sending end, wherein the codeword Ws is a matrix having r columns, the codeword Ws is composed of r column vectors, at least one of the r column vectors is composed of a base vector uS, and the base vector uS satisfies the following model: [equation 1], there being at least one column vector determined from among v^1^s, v^2^s,..., v^k^s by means of extracting a column from a matrix XS, and the matrix XS being determined on the basis of the frequency-domain position of the sub-band or a sub-band index s.

A large language model quantization method based on channel arrangement and activation adaptive smoothing belongs to the technical field of large language model compression and efficient inference. In order to solve the problem that the existing post-training quantization method is difficult to simultaneously consider the difference of different channel activation outliers, the group mode quantization has large dynamic range in the group, and the low bit weight and high bit activation alignment efficiency is low, which leads to the problem of precision decline and inference efficiency limitation, the scheme calibrates the maximum activation amplitude of the input channel by statistical data, divides the significant and non significant activation outlier channels according to the percentile threshold, and respectively uses the logarithmic domain and power function domain smoothing strategy to generate channel smoothing scaling factor, and fuses it into the model parameter to extract the weight multidimensional feature to construct the channel quantization sensitivity index and execute the input channel rearrangement, and further generate the joint quantization scaling factor to realize the integer domain alignment inference. It is suitable for low bit compression deployment and efficient inference scene of large language model in resource limited environment.

The invention provides a post-quantization method and system for end-side deployment of a large language model. The method comprises the following steps: determining a channel value range corresponding to an input channel of any to-be-quantized current matrix of the large language model; according to the channel value range, channel smoothing is carried out based on a preset channel smoothing algorithm, and quantization parameters of the smoothed channel are calculated; performing channel quantization according to the quantization parameter of the smoothed channel to obtain a quantized matrix; and the quantized matrix is used for deployment on end side equipment, so that low-cost and high-efficiency operation of the large language model is realized. According to the method, efficient compression of the large language model can be achieved, the method is suitable for being deployed on end-side equipment, calculation and communication overhead is saved, and the overall precision of the quantification process is remarkably improved.

Provided in the present application are information transmission methods, devices, and a storage medium. An information transmission method applied to a first communication device comprises: using a received reference signal for channel measurement, so as to obtain a first channel measurement result; converting the first channel measurement result according to a preprocessing mode, so as to obtain a second channel measurement result; performing quantization and feedback on channel information according to the second channel measurement result; and sending channel quantization and feedback indication information to a second communication device.

The invention discloses a reconfigurable intelligent surface assisted backscattering signal coding method and system, and solves the problems of low multi-signal superposition processing efficiency, low signal identification degree and low transmission efficiency in the prior art. Each quantized signal is input into one RIS unit; the method comprises the following steps: preprocessing an input quantized signal, dividing signals of K channels into two equalization subsets, and distributing different weights for the quantized signal of each channel; calculating the phase shift and backscattering information of each channel quantized signal in the subset by using the weight and the channel number K and combining the opening and closing condition of each RIS unit, and carrying out backscattering signal coding; and based on the obtained coded signal, combining gain in a signal transmission process to obtain a superposed signal. The problem of signal superposition in air computing is solved, the signal identification degree and the transmission efficiency are improved, and the complexity and the cost of an air computing system are reduced.

This invention discloses a method and system for evaluating the effect of wellbore grouting based on change attribution, relating to the field of grouting quality evaluation technology. The method includes: acquiring pre-grouting and post-grouting data for the same detection well; extracting baseline defect characteristics from the pre-grouting data and establishing a pre-grouting defect template; constructing a non-grouting reference response based on the pre-grouting defect template and combining the response patterns of stable sections and adjacent background changes; extracting measured response characteristics from the post-grouting data; calculating the grouting action and residual defect amount using dual-channel quantization based on the measured response characteristics, the non-grouting reference response, and the pre-grouting defect template; identifying change attribution based on the grouting action and residual defect amount to obtain initial attribution results; and performing section-level continuity optimization based on the initial attribution results to obtain the grouting effect evaluation results.

Systems, apparatuses and methods may provide technology for optimizing an inference neural network model that performs asymmetric quantization by generating a quantized neural network, wherein model weights of the neural network are quantized as signed integer values, and wherein an input layer of the neural network is configured to quantize input values as unsigned integer values, generating a weights accumulation table based on the quantized model weights and a kernel size for the neural network, and generating an output restoration function for an output layer of the neural network based on the weights accumulation table and the kernel size. The technology may also perform per-input channel quantization. The technology may also perform mixed-precision auto-tuning.

The application provides an offline model optimization method, device, equipment, storage medium and program product, and relates to the technical field of data processing. The offline model optimization method constructs weight distribution feature statistical results through each weight channel corresponding to each network layer in the to-be-optimized offline model, calculates the channel importance index and the target channel quantization factor corresponding to each weight channel based on the same data framework, realizes deep linkage and cooperation of weight quantization and channel pruning, further performs channel pruning and weight quantization optimization processing on the to-be-optimized offline model based on the channel importance index and the target channel quantization factor corresponding to each weight channel, can reduce the decision mismatch caused by the separation design of weight quantization and channel pruning, reduce the risk of deleting key weight channels by mistake, improve the adaptability of the reserved weight channels to low-precision representation, and significantly reduce the model precision loss, the model parameter quantity, the reasoning time consumption and the calculation amount without retraining.

Proposed in one example of the present disclosure is a video coding for machines (VCM) decoding device. The VCM decoding device comprises: an inner decoder which generates a feature map by decoding a bitstream and parses information about the feature map by decoding the bitstream; and an image reconstructor which reconstructs at least one skipped channel in the decoded feature map on the basis of the information about the feature map, wherein the information about the feature map may include at least one among group-wise coding information including whether each channel of the feature map is skipped, quantization compensation parameters, feature map truncation information, temporal reconstruction information, spatial reconstruction information, and packing information.

The invention belongs to the technical field of analog integrated circuits, and particularly relates to an arbitrary period overshoot loop delay compensation CTDSM circuit based on cascade adder feedback. According to the invention, an output signal Y of the CTDSM is delayed in a manner that the delay period is gradually increased from 1 to m, and the quantization results of the CTDSM of the first n channels which are completely quantized are fed back to the input end of the TIADC in an analog domain manner; the quantization result of the channel CTDSM which is not completely quantized is fed back to the output end of the TIADC in a digital domain mode, and compensation of overshoot loop delay of m unit periods is completed, so that the circuit time sequence design difficulty is reduced. Meanwhile, signals corresponding to quantization results of the first n channels after additional one-to-one corresponding gain amplification are subtracted from the input end of the TI ADC, and the input swing is reduced; the signals subtracted from the input end of the TI ADC are added back at the output end of the TI ADC through digital domain addition, so that the signals are ensured not to be distorted. According to the invention, the overshoot loop delay of any period in the CTDSM using the TIADC as the quantizer can be effectively compensated, the circuit stability is improved, and support is provided for the development of the CTDSM in the high-speed and high-precision field.

The invention provides a method for improving post-quantization model similarity. The method comprises the following steps: S1, storing original model features; s2, counting feature quantization parameters and quantizing features; s3, storing the input characteristics of the current layer of the quantitative model; s4, quantizing the weight, wherein the weight is suitable for a convolution operator, and the weight data dimension is (convolution kernel width, convolution kernel height, the number of input channels and the number of output channels); the method further comprises the following steps: S4.1, obtaining the maximum value and the minimum value of the weight; s4.2, quantifying the weight channel by channel; the method comprises the following steps: calculating a similarity score; and S5, storing weight quantization parameters. According to the method, quantization parameters are calculated through scaling of group channels of weights, so that a group of relatively suitable quantization weight methods can be obtained, and the similarity of quantization models can be further improved.

Embodiments of the present application provide an information transmission method, an electronic device, and a storage medium. The method comprises: receiving a reference signal for channel measurement; determining a preprocessing mode of the reference signal, and on the basis of the preprocessing mode, preprocessing the reference signal; and performing channel estimation on the basis of the preprocessed reference signal, and feeding back channel quantization indication information to a configuration end.

The embodiment of the invention provides a channel information feedback method, electronic equipment and a storage medium, and the method comprises the steps: obtaining a channel parameter of a measurement channel; determining a codebook set of channel quantization feedback of the channel; selecting a code word Ws for channel quantization representation of the sub-band s from the codebook set; determining an indication parameter of the code word Ws, and feeding back the indication parameter to the sending end; wherein the code word Ws is a matrix with r columns; the code word Ws is composed of r column vectors # imgabs0 #, at least one of the r column vectors # imgabs1 # is composed of a base vector us, and the base vector us meets the following models that at least one column vector in # imgabs2 # # imgabs3 # is determined by extracting columns from a matrix Xs, the matrix Xs is determined according to the frequency domain position of a sub-band or a sub-band index s, the channel information feedback problem can be solved, and the channel information feedback efficiency can be improved. And the code word is determined according to the frequency domain position or the sub-band index, so that the channel quantization accuracy is improved, and the channel feedback error is reduced.

Embodiments of the present application provide a channel information feedback method, an electronic device, and a storage medium. The method comprises: acquiring a channel parameter of a measured channel; determining a codebook set for channel quantization feedback of the channel; selecting from the codebook set a codeword for quantitatively characterizing the channel parameter; and determining an indication parameter of the codeword, and feeding back the indication parameter to a sending end, wherein the codebook set comprises at least one type of codewords which meet the following conditions: each codeword is a vector or matrix formed by N elements; and the n-th element un of the codeword has the following feature: un is generated by a parameter set x(n) and a bias z, wherein the parameter set x(n) consists of at least one parameter, and at least one parameter in the parameter set x(n) is determined on the basis of the index value n of an element to be generated.

Provided in an embodiment of the present application are a channel state feedback method, an electronic device and a storage medium, the method comprising: determining the configuration of a channel state information measurement pilot frequency, the channel state information measurement pilot frequency comprising M ports and Mgt; 1; receiving a channel state information measurement pilot frequency and performing channel measurement; selecting N ports from the M channel state information measurement pilot frequency ports to form a port set Ps for sub-band s channel characterization; determining a channel corresponding to the port set Ps; determining a codebook for quantitative feedback of the channel; selecting code words used for sub-band s channel quantization representation from the codebook; determining an indication parameter of the code word, and feeding back the indication parameter to the sending end; wherein the port set Ps is determined according to the frequency domain position of the sub-band or the sub-band index. According to the embodiment of the invention, the channel information feedback problem of the wireless communication channel is solved, the matrix can be determined according to the frequency domain position or the sub-band index, the channel quantization accuracy is improved, and the channel feedback error is reduced.

A circuit for channel randomization based on time-interleaved ADC includes: a channel selection module for outputting M clock reception control signals and encoded N data reception control signals based on a main clock and a generated random number; a multi-phase clock distribution module for generating N multi-phase clocks according to a sampling main clock, redistributing the multi-phase clocks according to the clock reception control signals, and outputting M redistributed clock signals; a time-interleaved ADC module for outputting M output data and a corresponding number of channel quantization completion signals according to the redistributed clock signals; an adjustable delay module for setting a delay length for the data reception control signals; and a timing distribution control module for controlling, according to delayed data reception control signals and the channel quantization completion signals, the output data to be output sequentially in chronological order.

The embodiment of the invention provides a channel information feedback method, electronic equipment and a storage medium. The method comprises the following steps: acquiring channel parameters of a measurement channel; determining a codebook set of channel quantization feedback of the channel; selecting a codeword for quantizing the characterization channel parameter from the codebook set; determining an indication parameter of the code word, and feeding back the indication parameter to the sending end; wherein at least one type of code words included in the codebook set meet the following conditions: the code words are vectors or matrixes formed by N elements; the nth element un of the code word has the following characteristics: un is generated by a parameter set x (n) and an offset quantity z, the parameter set x (n) is composed of at least one parameter, and at least one parameter exists in the parameter set x (n) and is determined according to an index value n of an element to be generated, so that the problem of channel information feedback of a wireless communication channel is solved, and the offset quantity z of the element to be generated is solved. The codebook for feeding back the channel information is generated according to the offset value, a codeword generation model can be changed, the adaptation degree of the codebook and the channel is improved, and the quantization error of the channel is reduced.

The invention discloses a quantification method of a model, a reasoning method of the model and electronic equipment. In the method, the first model is quantized based on the quantization parameter of the weight, the quantization parameter of the activation value, the first bias vector and the second bias vector, on one hand, the channel-by-channel quantization of the weight and the lexical element-by-lexical quantization of the activation value can be realized, and the quantization precision is relatively high; and on the other hand, the original bias vector is decomposed into two low-rank bias vectors, and the two low-rank bias vectors are spliced with the weight and the activation value respectively during quantization, so that the parameter quantity can be reduced, the memory occupation is reduced, and the acceleration effect of the quantization model is realized.

The present invention proposes a channel randomization circuit and method based on a time-interleaved ADC, comprising: a channel selection module, configured to output M clock reception control signals and N encoded data reception control signals according to a master clock and a generated random number; a multi-phase clock distribution module, configured to generate N multi-phase clocks according to a sampling master clock, and redistribute the multi-phase clocks according to the clock reception control signals to output M redistributed clock signals; a time-interleaved ADC module, configured to output M output data and a corresponding number of channel quantization completion signals according to the redistributed clock signals; an adjustable delay module, configured to set the delay size of the data reception control signal; and a timing distribution control module, configured to control the output data to be arranged and output in chronological order according to the delayed data reception control signal and the channel quantization completion signal.

The embodiment of the invention provides a communication method and related equipment, and relates to the field of communication. In the method, transmission layers are divided into a first transmission layer group and a second transmission layer group according to the channel quality of the transmission layers, then the first transmission layer group and the second transmission layer group are mapped to a first code word and a second code word, a first cooperative transmission resource set is configured for the first code word, and a second cooperative transmission resource set is configured for the second code word. The number of transmission resources of the first cooperative transmission resource set is smaller than the number of transmission resources of the third cooperative transmission resource set, and / or the number of transmission resources of the second cooperative transmission resource set is smaller than the number of transmission resources of the third cooperative transmission resource set, so that when channel quantization is carried out based on the codebook, the channel quantization efficiency is improved compared with the prior art. According to the invention, the number of channels needing channel quantization is reduced, the data volume during CSI feedback can be reduced, and the communication overhead of CSI feedback is further reduced.

Systems, apparatuses and methods may provide technology for optimizing an inference neural network model that performs asymmetric quantization by generating a quantized neural network, wherein model weights of the neural network are quantized as signed integer values, and wherein an input layer of the neural network is configured to quantize input values as unsigned integer values, generating a weights accumulation table based on the quantized model weights and a kernel size for the neural network, and generating an output restoration function for an output layer of the neural network based on the weights accumulation table and the kernel size. The technology may also perform per-input channel quantization. The technology may also perform mixed-precision auto-tuning.

The invention discloses an end-side operating system-oriented large model adaptive quantitative deployment method and device, and belongs to the field of artificial intelligence. Firstly, the layer importance index of each layer is evaluated through a Jaccard coefficient, the quantization precision strategy is adaptively adjusted according to the memory capacity of the end side equipment, and a quantization scheme is dynamically formulated according to the layer importance index and the equipment resource condition. And finally, performing per-channel quantization on the layer needing to be quantized in the large model according to the adjusted quantization precision strategy and the corresponding precision, and deploying the model on the end side equipment. Through the intelligent self-adaptive quantification and deployment method, the model storage requirement can be remarkably reduced on the premise that the large model precision is kept as much as possible, and then efficient deployment of the model on different hardware platforms and under different use scenes of the same hardware platform is achieved. In the quantification process of the large model, the importance of the layers is considered, and a more flexible and efficient quantification strategy is adopted in combination with the equipment resource condition. Compared with a traditional unified quantification method, dynamic optimization can be carried out according to the importance of different layers and hardware resource conditions, so that the storage requirement of the model is reduced and reasoning is accelerated under the condition that precision is not obviously lost.

Embodiments of the present application relate to the field of communications, and provide a communication method and a related device. In the method, on the basis of the channel quality of transport layers, the transport layers are divided into a first transport layer group and a second transport layer group, and then the first transport layer group and the second transport layer group are mapped to a first code word and a second code word; and a first cooperative transmission resource set is configured for the first code word, and a second cooperative transmission resource set is configured for the second code word. The number of transmission resources in the first cooperative transmission resource set is less than the number of transmission resources in a third cooperative transmission resource set, and / or the number of transmission resources in the second cooperative transmission resource set is less than the number of transmission resources in the third cooperative transmission resource set, and thus, when channel quantization is to be performed on the basis of a codebook, the number of channels requiring channel quantization is reduced compared with the prior art, such that the data volume involved in feeding back CSI can be reduced, thereby reducing the communication overhead of feeding back CSI.

This invention proposes a method and apparatus for efficient vector quantization based on pre-quantization and post-correction, relating to the field of image data discrete compression and reconstruction technology. The method includes: encoding input data into a continuous representation using a pre-trained variational autoencoder; dividing the continuous representation into multiple channel groups using a multi-channel quantization strategy, assigning an independent codebook to each channel group and generating corresponding quantization features; constructing an EfficientViT-based post-corrector to optimize the quantization features, generating optimized features; and using a pre-trained variational autoencoder decoder to decode the optimized quantization features to obtain reconstructed data.

This disclosure relates to optimizing low-precision inference models for deployment in deep neural networks. Systems, apparatus, and methods provide techniques for performing asymmetric quantization for optimizing inference neural network models, achieved by: generating a quantized neural network, wherein the model weights of the neural network are quantized to signed integer values, and wherein the input layer of the neural network is configured to quantize input values ​​to unsigned integer values; generating a weight accumulation table based on the quantized model weights and the kernel size of the neural network; and generating an output recovery function for the output layer of the neural network based on the weight accumulation table and the kernel size. The technique also performs per-input channel quantization. The technique also performs mixed-precision autotuning.

The invention discloses a commodity information management method and system based on big data analysis. The method relates to the technical field of commodity information management, and comprises the following steps: multi-source data acquisition and synchronous quality quantification, data synchronous adaptive regulation and control, supply chain information penetration and risk quantification, and risk intelligent tuning and collaborative response. According to the method, a multi-source commodity information acquisition channel is established, the accuracy and timeliness of supplier service data synchronization are quantified, and whether the information visualization link is entered after self-adaptive regulation and control is judged according to the accuracy and timeliness; and then, obtaining supply chain information penetration and risk parameters, and judging whether to enter a supply chain collaboration link after intelligent adjustment and optimization so as to balance data, risks, resources and efficiency, thereby improving the commodity information management reliability and improving the commodity information management efficiency. The problem of low commodity information management reliability caused by islanding of the business process and the cooperation mode in the prior art is solved.

This present disclosure provides a communication method and apparatus used to adaptively determine CSI channel quantization parameters for EDs at different sensing environments, so that the feedback accuracy of the CSI may be improved. The method includes: receiving a reference signal; obtaining first channel state information by performing channel measurement based on the reference signal; transmitting a first channel state information report, where the first channel state information report includes quantized first channel state information that is quantized based on first channel quantization information, the first channel quantization information indicates a channel quantization parameter corresponding to a first sensing environment parameter set, and a first sensing environment of a ED corresponds to the first sensing environment parameter set.