Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

430 results about "Linear layer" patented technology

Method for analyzing vibration of fluid conveying pipe on basis of fourier featured pinn

Provided in the present invention is a method for analyzing the vibration of a fluid conveying pipe on the basis of a Fourier featured PINN. The method comprises: constructing a physics-informed neural network (PINN), determining initial conditions and boundary conditions on the basis of a fluid conveying pipe model, and obtaining loss functions corresponding to the initial conditions and the boundary conditions; using a "soft" constraint to express a second-type boundary condition and initial excitation of a pipe into PINN loss functions, and using a "hard" constraint to code a first-type boundary condition and an initial amplitude into a DNN system structure; adding an observation anchor point to the network for correction when a peak value is insufficient during a network model training process; constructing a Fourier feature mapping function, adjusting a neural tangent kernel of the network, training two mapping results by means of a set network, and finally, in the network, connecting spatial and temporal hidden layers by means of point multiplication, so as to establish an FFNN; and by means of training and learning, finally outputting a prediction result by means of a linear layer connection.
Owner:FUZHOU UNIV

Multi-modal semi-supervised medical image segmentation method based on SAM adaptive fine tuning

The invention discloses a multi-modal semi-supervised medical image segmentation method based on SAM adaptive fine tuning, and the method comprises the steps: obtaining an original multi-modal image, adjusting the original multi-modal image to a preset size, and carrying out the image preprocessing, thereby obtaining a multi-modal image pair; simultaneously inputting the multi-modal image pair into a student branch model and a teacher branch model which are completely identical in structure, wherein each of the student branch model and the teacher branch model comprises a fusion module, an SAM encoder and a layered decoder; the multi-modal image pair passes through a fusion module formed by a convolutional network to obtain fusion features, the fusion features are input to the SAM encoder, and output features of different Transform layers are obtained; and inputting the plurality of output features into a hierarchical decoder, performing similarity calculation by the hierarchical decoder according to the plurality of output features to obtain prototype classification, and obtaining a prediction mask through a linear layer. According to the method, the multi-level features can be well utilized to realize fine-grained segmentation, and a good segmentation effect is realized.
Owner:SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI

Abnormal value-based weight activation quantification method after large model training

The invention discloses a weight activation quantification method based on abnormal value large model training. The weight activation quantification method comprises a weight quantification process and an activation value quantification process during reasoning. In weight quantization, model bit width is converted into low bit representation in an off-line mode, and in weight activation value quantization during reasoning, a quantized weight model is initialized, input of a linear layer is pre-scaled during reasoning operation, and finally 8-bit quantization operation is executed. According to the method, mathematical equality transformation of weight and activation is combined, a strategy of lossless quantization and low-delay execution is provided, and the significant weight is considered in an off-line weight quantization process, so that the influence of a smooth activation abnormal value on the significant weight is reduced, and the quantization error of the abnormal value is reduced in a model reasoning process. According to the method, effective quantification during large-scale model reasoning is realized, the memory required for running a large model is remarkably reduced, and the running speed is increased.
Owner:THE ACAD OF TIANJIN UNIV HEFEI

Construction method and device of AI text detection model, equipment and storage medium

The invention relates to a construction method of an AI text detection model, and the method comprises the steps: determining a pre-training BERT model through constructing a training data set comprising a text generated by AI and a text created artificially; and coding the training data in the training data set to obtain corresponding mark data. And adding a linear layer classifier on the pre-trained BERT model to obtain an initial classifier, and training the initial classifier by using the cross entropy loss function, the optimizer and the marked data to obtain an AI text detection model. According to the method, the common sense of the pre-trained BERT model is utilized, the higher convergence speed and better generalization ability can be obtained, the detection accuracy and the user confidence are improved, the spreading of misleading information is avoided, an effective and reliable solution is provided for detecting the AI to generate the Chinese text, and the method is suitable for popularization and application. And development and application of a natural language processing technology are further promoted.
Owner:PING AN TECH (SHENZHEN) CO LTD

Aero-engine missing value filling method based on multi-scale time sequence and attribute feature extraction

The invention discloses an aero-engine missing value filling method based on multi-scale time sequence and attribute feature extraction. Firstly, for data collected by each sensor of an aero-engine, a temporary replacement value of a missing value is generated by adopting linear interpolation; extracting a plurality of significant cycles of the aero-engine sequence data as a time scale, decomposing the sequence into trend terms and season terms according to the significant scale, and splicing all the trend terms and season terms in the time dimension; then, in-scale and inter-scale time sequence dependence is captured in a time dimension by using cross-scale graph convolution so as to carry out missing value filling in time sequence. In the attribute dimension, local neighborhood information is accumulated by adopting a cross-attribute graph volume, global attribute features are calculated by utilizing a relational network, node features are updated, and attribute association is supplemented on the basis of a time sequence filling result. And finally, aggregating the node features of a specific scale through a maximum pooling layer to represent the features, and calculating by adopting two linear layers and an activation function to obtain a final filling result.
Owner:DALIAN UNIV OF TECH

DETR target detection model compression method based on combination of pruning and knowledge distillation

The invention belongs to the field of computer vision, and particularly relates to a DETR target detection model compression method based on combination of pruning and knowledge distillation, and the method comprises the steps: calculating an attention mechanism of each layer of a Transform structure, and constructing a mask by using the attention mechanism; performing adaptive pruning on parameters of a linear layer and an attention layer of the Transform structure by using a mask to obtain a first compression model of the original model; taking a to-be-compressed encoder with a Transform structure as a student network, and taking the original model as a teacher network to carry out distillation learning; the student network performs distillation updating according to the global distillation loss and the intersection distillation loss to complete compression; according to the method, the calculated amount of the model can be remarkably reduced and the reasoning speed of the model can be greatly improved on the premise that the model does not have obvious precision glide.
Owner:CHONGQING STATE-OWNED BIG DATA INTELLIGENT TECHNOLOGY CO LTD

Large model compression method and device, task processing method and equipment and storage medium

The invention relates to the technical field of model compression, and provides a large model compression method and device, a task processing method and equipment and a storage medium, and the large model compression method comprises the steps: carrying out the layer-by-layer quantification of a linear layer of a to-be-compressed initial large model, and obtaining a first large model; the initial large model is a pre-trained large language model constructed based on an expert hybrid architecture; performing route calibration on each expert sub-model in the first large model to obtain a second large model; in the reasoning process of the second large model, based on the task type of a to-be-executed target task, the importance of each expert sub-model in the task type is evaluated; and performing dynamic pruning on each expert sub-model based on importance so as to compress the second big model. Through a compression mode of combining static quantification and dynamic pruning, on the basis of ensuring the model performance, the memory and calculation overhead required by large model reasoning can be reduced, and efficient operation of the large model on light-weight equipment with limited video memory resources is facilitated.
Owner:INST OF AUTOMATION CHINESE ACAD OF SCI

Social robot detection method based on KAN graph neural network model

The invention provides a social robot detection method based on a KAN graph neural network model, and the method comprises the steps: carrying out the data preprocessing of an identified social account data set, carrying out the adaptive synthesis sampling through employing an ADASYN method, and carrying out the balance processing of the data set; after the balanced data set is input into the model, graph structure data is processed through two layers of convolution, the first layer extracts preliminary features, the second layer further learns multi-level graph features, in each layer of convolution, a KAN linear layer enhances the nonlinear expression ability of the features by using B spline interpolation, and the feature extraction efficiency is improved; the model is helped to better capture various social interaction type relationships in the graph data; neighborhood between nodes is used for learning representation of the nodes, the nodes with higher representation capability can be learned through two convolutional layers for training, classification prediction is further performed through a softmax function, and a final prediction result is obtained, so that social robot detection is realized.
Owner:KUNMING UNIV OF SCI & TECH

Large model reasoning acceleration method and device and medium

PendingCN119204222AImprove parallel processing capabilitiesImprove inference speedInference methodsPhysical realisationData setAlgorithm
The invention discloses a large model reasoning acceleration method and device and a medium, and the method comprises the steps: constructing a training set and a test set for training a large model according to scene data; expanding a prediction head in the large model to obtain an expanded target large model, and freezing original prediction head parameters in the large model so as to update the prediction head parameters corresponding to the expanded prediction head in the training process of the target large model; training and evaluating the target large model through the training set and the test set to obtain a trained target large model; constructing a calibration data set formed by partial scene data, and inputting the calibration data set into the target large model for forward reasoning to obtain a calibration matrix corresponding to each linear layer; and determining an input matrix and an output matrix corresponding to the linear layer according to the calibration matrix, and quantizing a weight corresponding to the linear layer based on the input matrix to obtain a quantized model weight.
Owner:SHANDONG INSPUR SCI RES INST CO LTD

Quantization method and reasoning method of large language model and electronic equipment

The invention discloses a quantification method and reasoning method of a large language model and electronic equipment, and belongs to the technical field of large language models.The quantification method of the large language model comprises the steps that each linear layer to be quantized in the large language model is quantized; dividing a channel of the linear layer in the hidden layer dimension into a normal channel and an outlier channel; performing INT8 quantization on the first activation matrix corresponding to the normal channel in a word segmentation token dimension to obtain a second activation matrix, and performing INT4 quantization on the first weight matrix corresponding to the normal channel according to an output channel to obtain a second weight matrix; and determining an output result of the linear layer according to the second activation matrix, the second weight matrix, a third activation matrix corresponding to the outlier channel and a third weight matrix corresponding to the outlier channel.
Owner:ZTE CORP

Model training method and device, equipment and storage medium

The invention provides a model training method and device, equipment and a storage medium, and relates to the technical field of computers, in particular to the technical field of neural network models and model training. The specific implementation scheme is as follows: a calculation unit executes quantization matrix multiplication based on Hadamard pre-transformation on an activation tensor and a weight tensor of a target model stored in a memory so as to generate an output tensor of a linear layer based on a low-precision tensor with smaller data bit width; using the output tensor and a subsequent network layer of the target model to complete forward propagation so as to obtain a loss value; and according to the loss value, updating model parameters of the target model stored in a memory through a back propagation algorithm. By means of the technical scheme, on the premise that the model training precision is guaranteed, memory resource occupation and the calculation amount in the calculation process can be remarkably reduced, and the training cost is reduced.
Owner:BEIJING BAIDU NETCOM SCI & TECH CO LTD

Large model quantification method and system

The invention relates to a large model quantification method and system. The method comprises the following steps: performing greedy grid search on each linear layer of an off-line large model to obtain an optimal smoothing factor corresponding to each linear layer; performing reverse scaling on the weight and the activation value of the corresponding linear layer in the channel dimension by using the obtained optimal smoothing factor corresponding to each linear layer so as to keep the multiplication result between the scaled weight and the activation value unchanged, thereby obtaining the smoothed weight and the smoothed activation value; and performing quantization processing on the smoothed weight and activation value in the large model so as to convert into a numerical format of INT8 or FP8 or a numerical format lower than INT8 or FP8.
Owner:BEIJING SILICONFLOW TECHNOLOGY CO LTD

Algorithm of tensor parallel computing large model based on cpu + gpu

The invention discloses a cpu + gpu-based tensor parallel computing algorithm for a large model, which comprises the following steps of: S1, combining row parallelism and column parallelism for a linear layer in the model by adopting a mixed-dimension tensor segmentation mode, and dynamically adjusting a segmentation proportion according to a model structure and hardware resources; s2, a CPU and GPU cooperative computing mechanism is constructed, part of tasks which have large video memory requirements and are relatively simple in computation are allocated to the CPU, the GPU is responsible for computing intensive tasks, communication between CPU-GPU is optimized, and overlapping of CPU-GPU communication and GPU computation is achieved; and S3, implementing a dynamic resource allocation and load balancing strategy, monitoring load conditions of the CPU and the GPU in real time, dynamically adjusting task allocation according to calculation requirements of different layers of the model and use conditions of hardware resources, and adopting a self-adaptive batch processing size adjustment strategy. The invention obviously reduces the occupation of the video memory, reduces the communication cost and improves the utilization rate of computing resources.
Owner:GUANGDONG UNIV OF TECH +1

Prediction method and device based on low-rank quantization large model, electronic equipment, storage medium and computer program product

The invention relates to a prediction method and device based on a low-rank quantization large model, electronic equipment, a storage medium and a computer program product, and the method comprises the steps: obtaining a prediction model which is a large model; performing low-rank decomposition on parameters of a linear layer in the prediction model to obtain a first low-rank matrix, a second low-rank matrix and a third low-rank matrix; based on the first low-rank matrix, the second low-rank matrix and the third low-rank matrix, a quantization low-rank module corresponding to the linear layer is constructed, and the quantization low-rank module sequentially comprises a first activation quantization layer, a first linear layer, a scale scaling layer, a second activation quantization layer and a second linear layer; replacing a linear layer in the prediction model with a quantization low-rank module to obtain a compressed prediction model; and inputting the input data into the compressed prediction model to obtain a prediction result corresponding to the input data. The problem that the model prediction reasoning speed is low can be solved.
Owner:INST OF AUTOMATION CHINESE ACAD OF SCI

Multi-wave-potential pulse radar target detection method and device based on deep learning, and medium

The invention discloses a multi-wave-potential pulse radar target detection method and device based on deep learning and a medium, and the method comprises the steps: obtaining radar original echo data in different detection scenes, and forming a radar original echo data set; training the target detection network by using the radar original echo data set to obtain a target detection model; in an input preprocessing layer of the target detection network, a real part and an imaginary part of radar original echo data are separated, linear layer networks with different weight matrixes are used for carrying out feature mapping on the real part and the imaginary part of different wave position data, the features of the real part and the imaginary part are calculated, then the phase spectrum features are calculated, and a multi-dimensional feature map is formed; training a target detection network by using the echo data set; and obtaining echo data to be detected, and inputting the echo data to the target detection model for detection. According to the method, high-precision and low-calculation-overhead target detection can be realized in a complex scene, and meanwhile, the detection performance and robustness are remarkably improved.
Owner:HUNAN NOVASKY ELECTRONICS TECH CO LTD

Model optimization method and device for low-resource scene, electronic equipment and medium

The invention relates to a model optimization method and device for a low-resource scene, electronic equipment and a medium, and the method comprises the steps: obtaining a pre-training model, the pre-training model being a Transform model, for any linear layer in the Transform model, connecting a pair of low-rank matrixes in parallel in the linear layer to form a LoRA increment path, and freezing a backbone network in the pre-training model; for any linear layer in the Transform model, a learnable gate is added on a LoRA increment path, and an improved LoRA module is formed; and obtaining training data, training the pre-training model according to the training data to obtain a target model, and performing semantic analysis on the to-be-processed text according to the target model. According to the method provided by the invention, through collaborative design of structural innovation and an optimization mechanism, the performance and stability of the efficient parameter fine tuning method in a low-resource scene are remarkably improved, and the method has a good industrial application prospect.
Owner:DIGITAL HEALTH CHINA TECHNOLOGIES CO LTD

Distributed optical fiber sound wave sensing data feature extraction method and device

The invention relates to a distributed optical fiber sound wave sensing data feature extraction method and device, and the method comprises the steps: carrying out the feature extraction of distributed optical fiber sound wave sensing data through a Transform network model which is pre-trained through a mask reconstruction task; the Transform network model is improved by utilizing an attention mechanism and comprises a coding network and a decoding network which are the same in structure, each of the coding network and the decoding network comprises a linear layer, a position coding layer and a multi-layer cascaded converter encoder, and the system is used for realizing the method. Compared with the prior art, the method does not need to use data labels for model training, has excellent generalization ability, and can be suitable for various tasks based on distributed optical fiber sound wave sensing data.
Owner:SHANGHAI JIAOTONG UNIV +1

Parameter fine tuning method and device of pre-training model, equipment and medium

The invention discloses a parameter fine tuning method and device of a pre-training model, equipment and a medium. The parameter fine tuning method is used for performing parameter fine tuning on a pre-training model for natural language processing based on a text data set, the pre-training model comprises a linear layer with an initial weight matrix, and the parameter fine tuning method comprises the following steps: applying orthogonal transformation based on at least one rotation matrix to the linear layer and obtaining an angle parameter matrix according to the orthogonal transformation, constructing an orthogonal fine tuning layer based on the initial weight matrix and the angle parameter matrix, and replacing a linear layer in the pre-training model with the orthogonal fine tuning layer to obtain a to-be-fine-tuned model; only updating the angle parameter matrix through a gradient back propagation algorithm based on the text data set so as to train the to-be-fine-tuned model; and outputting the trained model to be fine-tuned as a parameter fine-tuning result.
Owner:INST OF AUTOMATION CHINESE ACAD OF SCI

Traffic flow prediction method based on temporal-spatial feature fusion Transform in edge environment

The invention relates to a traffic flow prediction method based on temporal-spatial feature fusion Transform in an edge environment, and the method comprises the steps: carrying out the feature embedding and coding of original traffic flow data; then, constructing a multi-head convolution low-rank decomposition attention mechanism to capture a long-term time dependency relationship and obtain local context information; then, constructing an attention map convolution to capture a spatial dependency relationship; and finally, performing adaptive fusion on the spatial-temporal characteristics through a gating unit, and further realizing accurate prediction of future traffic flow by using a feedforward neural network and a linear layer. The method can improve the prediction precision, effectively reduce the resource overhead, and improve the prediction efficiency.
Owner:FUZHOU UNIV

Deep learning method and device for anomaly detection of industrial control system, storage medium and computer equipment

The invention discloses a deep learning method and device for anomaly detection of an industrial control system, a working medium and computer equipment. The neural network model constructed by the method utilizes one-dimensional convolution and a linear layer to extract time-dependent features of sensor nodes. A physical flow chart divides sensor nodes into sub-graphs, local dependency features are extracted by setting a complete graph and a graph attention network, physical flow embedding is generated in a pooling mode, and global dependency between flows is captured by applying the graph attention network. According to the control chart, communication characteristics of sensors in the controllers are extracted through a chart attention network, controller embedding is generated through pooling, and then information flow between the controllers is captured by applying the chart attention network. Through joint modeling of a physical flow chart and a controller communication chart, multi-level dependence is effectively captured. And an early stop mechanism is adopted to avoid overfitting, and abnormal positioning based on abnormal scores is supported. The method can efficiently detect and position the abnormity in the industrial control system, and guarantees the safe and stable operation of the system.
Owner:JINAN UNIVERSITY

Height measurement method based on residual network multi-mode deep learning model

The invention relates to the technical field of satellite altimetry, in particular to an altimetry method based on a residual network multi-mode deep learning model, which can remarkably improve the robustness of SSH inversion under sea condition noise interference, and is characterized in that a fusion self-attention mechanism and a residual network multi-mode deep learning model ViTResNetMDL is established, and the model ViTResNetMDL is applied to improve the robustness of the SSH inversion under sea condition noise interference. A satellite-borne GNSS-R sea surface height inversion geometric physical model is established, a ViTResNetMDL model is composed of ResNet50, ViT and MHL-NN, the first part adopts a ResNet50 convolution structure to extract effective features from an original DDM, the second part adopts a ViT layer which captures global features in an effective scattering area DDM based on a self-attention mechanism in a Transform module, the third part is composed of linear layers, and the fourth part is composed of linear layers. And using the MHL-NN to retrieve the sea surface height.
Owner:HARBIN INST OF TECH AT WEIHAI

High-flow-degree MambaV2 hardware acceleration system based on calculation sequence optimization of FPGA (Field Programmable Gate Array)

The invention, which relates to the field of FPGA and machine learning, discloses a high-flow-degree MambaV2 hardware acceleration system with an optimized calculation sequence based on FPGA, comprising a preceding stage module, a tensor expansion module, a hidden layer state calculation module, an output state calculation module, a residual multiplication normalization module and an output linear layer module. The calculation process of the MambaV2 algorithm is divided into a plurality of pipeline levels, and a data stream is output in a pipeline form. According to the method, an original algorithm structure is not changed, the operation sequence of a hidden layer state equation in an original algorithm is changed, an algorithm for optimizing the calculation sequence is provided, the number of times of operation is reduced, cache consumption is reduced, the MambaV2 model is deployed and reasoned on an FPGA, the edge scene deployment requirement is met, customized acceleration of the FPGA is utilized, and data streams are adjusted, so that the edge scene deployment efficiency is improved. The pipeline design of the whole algorithm is realized, the utilization rate of computing resources is improved, the problem of data blocking is solved, and computing delay and cache consumption are reduced.
Owner:SUN YAT SEN UNIV

Overlapping relation triple extraction method based on global detection and bidirectional marking

The invention discloses an overlapping relation triple extraction method based on global detection and bidirectional marking, and belongs to the technical field of natural language processing. Firstly, a data set is preprocessed, a model uses a pre-training model BERT as an encoder, initial semantic feature representation of sentences is obtained, and four linear layers are used for abstracting the encoded semantic feature representation so as to be suitable for different tasks; an entity pair identification module based on bidirectional marking extracts all entity pairs with potential relations from two directions; the global detection module filters the extracted entity pairs and screens out the entity pairs with low confidence; finally, a relation extraction module based on entity attention extracts a semantic relation between the entity pairs, and extraction of relation triples is completed; in the training stage, an entity pair negative sampling strategy is used, the generalization of the model is improved, the model is jointly extracted by training the entity relationship, the model is tested, and then the model is updated according to a test result. According to the invention, the knowledge of the unstructured text can be realized.
Owner:SHENYANG AEROSPACE UNIVERSITY

Audio optimization method and device based on artificial intelligence, computer equipment and medium

The invention relates to the technical field of audio processing, in particular to an audio optimization method and device based on artificial intelligence, computer equipment and a medium. According to the method, a linear layer is used for mapping a spectrum envelope of a to-be-optimized audio to obtain an envelope feature, an embedding layer is used for embedding a standard audio parameter into a parameter feature, a prediction model is used for predicting a fusion feature of the envelope feature and the parameter feature to obtain a prediction pitch curve, and a noise adding model is used for adding noise to a Mel spectrum of the to-be-optimized audio to obtain a predicted pitch curve. Calculating the noise of the noise addition result by using a noise estimation model to obtain predicted noise, updating the noise estimation model according to the predicted noise, the real noise and the predicted pitch curve, calculating the reference noise of the noise addition result by using the updated noise estimation model, and de-noising the noise addition result according to the reference noise. The noise estimation model is optimized in combination with pitch information, so that the denoising process meets the pitch requirement, and the audio optimization effect is improved.
Owner:PING AN TECH (SHENZHEN) CO LTD

Conservative collision detection method for industrial robot path planning

The present invention belongs to the field of industrial robot collision detection, and discloses a conservative collision detection method for industrial robot path planning, including 1: using a binary search algorithm based on Morton code to construct a linear hierarchical bounding volume; 2: proposing a GPU-accelerated SDF construction algorithm; 3: proposing an envelope sphere generation algorithm based on particle swarm; 4: using SDF and envelope sphere to achieve conservative collision detection. The present invention marks and subdivides the voxel grid of the model surface area, and only calculates the distance field function value in the area, thereby reducing the number of voxels that need to be calculated, thereby improving the calculation efficiency, and using sparse grids for storage to reduce memory overhead; all threads can independently access LBVH and calculate the shortest distance, thereby improving the parallelism of the calculation; using a particle swarm optimization algorithm to calculate a tight envelope sphere instead of robot joints to perform collision detection, thereby improving the efficiency of collision detection.
Owner:ZHEJIANG UNIV

Multi-modal abstract method and system based on visual information fusion

The invention discloses a multi-modal abstract method and system based on visual information fusion, and relates to the data processing and artificial intelligence technology, and the method comprises the steps: splicing extracted visual features and text features, and inputting the spliced visual-text features into a multi-head attention module and a feedforward network module of an encoder; taking the extracted visual features and the output of the feedforward network module as the input of a modal fusion layer of an encoder to obtain text features combined with visual information; inputting the obtained text features combined with the visual information and the abstract text features into a decoder; taking the extracted visual features and the output of a decoder feedforward network module as the input of a decoder modal fusion layer to obtain the text features output by the decoder; and enabling text features output by a decoder to pass through a linear layer, inputting an output vector into a Softmax layer, and obtaining a required abstract text according to probability distribution and a vocabulary. According to the method, the abstract generation quality and accuracy can be improved.
Owner:CHINA ACADEMY OF ELECTRONICS AND INFORMATION TECHNOLOGY OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION +2

Quantitative model degradation correction method and device, electronic equipment and storage medium

The embodiment of the invention discloses a quantitative model degradation correction method and device, electronic equipment and a storage medium, relates to the technical field of data processing, and is convenient for reducing errors generated by a quantitative model. The method comprises the following steps: acquiring a key operator of an original model; implementing the key operator by adopting a model reasoning framework to obtain a model reasoning operator of a quantized version; registering the model reasoning operator into a deep learning training framework, and replacing a corresponding operator in the deep learning training framework with the model reasoning operator to obtain a quantitative model; injecting a fine-tuning low-rank weight into each linear layer of the quantitative model to obtain an injected quantitative model; performing fine tuning on the post-injection quantification model to obtain a fine tuning parameter result; and loading the fine tuning parameter result to the quantitative model. The method is used for degeneration correction of the quantitative model.
Owner:DINGDAO ZHILIAN (BEIJING) TECH CO LTD

Model training method and device, equipment, storage medium and program product

The invention discloses a model training method and device, equipment, a storage medium and a program product, and the method comprises the steps: carrying out the type conversion of an input tensor of a second data type based on a zooming parameter corresponding to a linear layer in a forward process, and obtaining an input tensor of a first data type; after the calculation of the linear layer is completed, converting the output tensor of the first data type into an output tensor of a second data type; in a reverse process, performing type conversion on the model parameter of the first data type based on a scaling parameter corresponding to a linear layer to obtain a model parameter of a second data type; determining a gradient of the second data type based on the model parameter of the second data type and the loss value of the second data type; and updating model parameters of the model by using the gradient of the second data type to obtain a trained model. Therefore, under the condition that the model precision is guaranteed, the model training efficiency is improved, and the requirements for computing power and storage during model training are reduced.
Owner:MOORE THREADS TECH CO LTD

Dynamic scene re-operation mirror video generation method and system based on diffusion model

The invention discloses a dynamic scene re-operation mirror video generation method and system based on a diffusion model, and belongs to the technical field of computer vision and video generation. A diffusion generation architecture with a control branch is adopted, and the core is composed of an embedded layer, a main branch and a control branch. In the control branch, the output of each sub-block is added with the output of the corresponding block of the main branch after being processed by the zero initial linear layer, and the sum is input into the next block of the main branch. During training, generating a rendered video by using the target video and the reference video in the same scene; and inputting the target video latent variable after noise addition into the control branch, inputting the splicing result of the target video, the reference video and the rendering video latent variable into the main branch, and simultaneously providing the text latent variable of the reference video for the two branches as a condition. During generation, the model finally generates a target video latent variable through step-by-step denoising and decodes the target video latent variable into a target track video, and it is ensured that the motion of a moving object in a scene of a generated video and a reference video is consistent at the same time.
Owner:ZHEJIANG UNIV