23 results about "Model application" patented technology

This invention relates to the field of AI model applications, specifically to a method and system for assembling model input parameters based on multidimensional variable expressions. The method includes: using a first multidimensional variable expression to filter and select various types of user-input data based on a pre-built system context variable dataset to obtain multiple corresponding computational datasets; using a second multidimensional variable expression to preprocess the data in the computational datasets to obtain multiple corresponding assembly datasets; using a third multidimensional variable expression to merge and assemble the multiple assembly datasets to obtain an input parameter dataset; parsing a pre-built input parameter JSON template and solving it based on the input parameter dataset to obtain the input parameter JSON for the AI ​​model. This invention, based on a multi-level, multidimensional vector expression-based automated process, can achieve real-time dynamic construction of input parameter JSON.

The embodiment of the application provides a kind of artificial intelligence native application ecosystem, and related method and product;Artificial intelligence native application ecosystem includes: big model application module and multiple application programs;Big model application module is used to determine user intent information, and according to user intent information, determine target application program from multiple application programs;With target application program carries out data interaction, to provide target application program corresponding service to user.By the embodiment of the application, the application program of third party can be called, so that AI application can provide more rich functions to user, to reach the performance ability of human assistant.

The application relates to the technical field of big model computing power scheduling, in particular to a computing power cooperative intelligent scheduling method and system based on a big model application, which comprises the following steps: obtaining a global computing power task request containing multiple big model inference tasks, extracting multi-dimensional features according to task constraint conditions and forming a task demand vector. An improved greedy algorithm is adopted to match the task demand vector and a state vector of a dynamic heterogeneous computing power resource pool, a preliminary scheduling instruction set is generated through weighted distance, a multi-step forward-looking evaluation model with computing power resource balance and task communication overhead as joint targets is constructed, the running state of tasks and resources in multiple time slices is simulated, the scheduling instruction is corrected, and finally the instruction is issued to a hardware computing unit to complete task execution. The method can improve the matching accuracy of tasks and computing power resources, optimize the balance of resource allocation, and reduce the data communication overhead between tasks.

This application provides a model evaluation method, apparatus, device, storage medium, and program product. The method includes: acquiring multi-dimensional evaluation quantitative data, wherein the multi-dimensional evaluation quantitative data includes at least one of the following: task adaptability of a first model, model deployment feasibility, and model application value; and determining the maturity of the first model deployment applied to the target device based on the multi-dimensional evaluation quantitative data. This method, by utilizing model maturity, allows for reasonable model selection in the early stages of model deployment, ensuring that the model deployed to the device matches actual needs and avoiding resource waste caused by blindly and recklessly using large models.

1. Name of this design product: Graphical User Interface for Large-Scale Evaluation of Electronic Equipment. 2. Purpose of this design: An electronic device. 3. The key design feature of this product is its graphical user interface. 4. The image or photograph that best illustrates the design's key points: the front view. 5. GUI (Graphical User Interface), omitting the rear view, left view, right view, top view, and bottom view. 6. Purpose of the graphical user interface: To implement various evaluations throughout the lifecycle of a large model. 7. Human-computer interaction method of graphical user interface: Select or enter version information from the "Specific Version" field in the main view, and click "Submit for Review" below to enter interface change state diagram 1. Click "Next" in interface change state diagram 1 to enter interface change state diagram 2. Click "Start Evaluation" in interface change state diagram 2 to enter interface change state diagram 3. Click "Next" in interface change state diagram 3 to enter interface change state diagram 4. Enter the model application service image (group), model ID, and application model packaging description in the text box in interface change state diagram 4, and click "Next Evaluation" in interface change state diagram 4 to enter interface change state. Figure 5, click "Start Timeliness Evaluation" in the interface change state diagram 5, enter the interface change state diagram 6, click "Next" in the interface change state diagram 6, enter the interface change state diagram 7, click "Start Performance Evaluation" in the interface change state diagram 7, enter the interface change state diagram 8, click "Next" in the interface change state diagram 8, enter the interface change state diagram 9, click "Submit Application Experience Evaluation" in the interface change state diagram 9, enter the interface change state diagram 10, click "Generate Report" in the interface change state diagram 10, enter the interface change state diagram 11, click "Next" in the interface change state diagram 11, enter the interface change state diagram 12. 8. Other situations requiring explanation: The upper left corner of the interface is the content screen.

This application discloses a task execution method, apparatus, and electronic device. Relating to the field of data processing, the method includes: upon receiving a training task for a target model, determining the model application scenario of the target model; acquiring single-turn dialogue data within the model application scenario, and constructing multi-turn dialogue data based on the single-turn dialogue data, wherein the single-turn dialogue data includes a question-answer pair, and the question-answer pair includes two dialogue rounds; and using the multi-turn dialogue data to execute the training task of the target model. This application solves the problem in related technologies where the processing method of piling up single-turn dialogue data results in low data validity and efficiency of multi-turn dialogue data, leading to low task execution efficiency for model training.

A communication method and apparatus, and relates to the field of communication technologies. In the method, after indicating a first state of one or more first models to a model management device, a model application device may retain the first state within a first time window, so that the model application device does not frequently report the first state of the one or more first models to the model management device, to reduce information exchange between the model application device and the model management device.

The application discloses a model construction method and system based on digital twinning, and the method comprises the following steps: collecting multi-source heterogeneous data of a physical entity, wherein the multi-source heterogeneous data comprises running data, environment data and state data; based on data collection errors, transmission delays and environmental interference factors, the multi-source heterogeneous data is dynamically corrected in credibility; based on the corrected credibility result, the sensitivity parameters output by the digital twinning model in combination with the multi-source heterogeneous data and the real-time requirement parameters of the model application scene corresponding to the multi-source heterogeneous data are used for dynamic weight distribution; based on the allocated dynamic weight, the core parameters of the digital twinning model are updated by using a lightweight iteration strategy to construct the digital twinning model. The system corresponds to realize the above method. The application solves the problems of multi-source data distortion, fixed weight and iteration calculation power consumption, realizes accurate model construction, and adapts to dynamic scene requirements.

The application discloses a three-dimensional model lightweight processing method and device and electronic equipment, and relates to the technical field of model lightweight processing, and comprises the following steps: acquiring original three-dimensional models in multiple formats; performing multi-step lightweight processing on the original three-dimensional models to obtain standard three-dimensional models; calling target parameters, optimizing the standard three-dimensional models according to the target parameters to obtain target three-dimensional models; and the target parameters comprise target platform hardware configuration parameters and model application scene parameters. The application can acquire original three-dimensional models in multiple formats, perform multi-step lightweight processing on the original three-dimensional models, optimize model display effects for different application scenes and ensure the quality on the basis that the loading and rendering speeds of three-dimensional models in various different formats are not affected, thereby bringing more efficient and sustainable solutions for ground engineering projects.

The present disclosure is directed toward systems, methods, and non-transitory computer readable media for generating a model application comprising one or more discrete model functions classified into a function category comprising a sense category, a reason category, or an act category. In some embodiments, the disclosed systems can combine a discrete model function with one or more additional model functions to generate a model application that defines data processing for a customized instance of a large language model. The disclosed systems can surface a model interaction interface comprising selectable application elements to instantiate respective applications of a large language model and modify the model interaction interface to surface a recommended source content item within a source selection window to utilize with the large language model.

The application discloses a kind of construction methods of numerical control processing whole-process multi-modal data space-time correlation model, it is related to numerical control processing whole-process processing data correlation modeling field, comprising: firstly, part geometric model is discretized into point cloud, and multi-level granularity is divided according to point, feature, overall geometry;Then, feature level granularity point cloud entity is constructed based on tool position point sequence;Next, process data is based on machining feature, monitoring data is aligned based on time, and based on space mapping, detection data is based on spatial positioning, respectively associated to the point cloud entity of corresponding granularity;Finally, through the mutual reference of the unique identification and attribute of entity, the fusion of multi-modal, multi-scale, multi-granularity data such as geometry, process, monitoring, detection in unified space-time coordinates is realized;The application solves the problem that multi-source heterogeneous processing data is difficult to express and correlate uniformly, and provides technical support for constructing high-quality processing data set and industrial big model application.

The application discloses a design method and device of a satellite control software general architecture. The method comprises the following steps: establishing a satellite control software layered architecture, the layered architecture comprising a computer interface abstraction layer, an operating system interface abstraction layer, a component interface abstraction layer, a core service layer and a model application layer arranged in sequence from bottom to top; constructing the computer interface abstraction layer to uniformly encapsulate the differences between different processor architectures; constructing the operating system interface abstraction layer to uniformly encapsulate the differences between different operating systems; constructing the component interface abstraction layer to uniformly encapsulate the differences between different devices; constructing the core service layer to form general functions on the basis of the computer interface abstraction layer, the operating system interface abstraction layer and the component interface abstraction layer; and constructing the model application layer to select the general functions in the core service layer for the target satellite model, to parameterize the configuration items, and to extend the model-specific functions. The application can effectively improve the cross-platform reuse capability, development efficiency and quality consistency of the software.

Systems and methods are described for scheduling and / or resource provisioning for foundation model applications. The resource provisioner may involve determining a slack from a performance target and an amount of consumed resources for a workflow request, the workflow request referencing a workflow, the workflow comprising at least one machine learning model; determining a remaining resource to complete the workflow request; tracking a slack violation amount responsive to the remaining resource exceeding the slack; and deploying at least one new replica of the machine learning model to at least one computational node based on a slack violation amount.

The invention relates to a method for determining the recruitment process of pilots using artificial intelligence-based fuzzy multi-criteria decision-making models, which creates an artificial intelligence-based decision-making model by considering numerous criteria such as socio-cognitive abilities, performance competencies, communication skills, visual perspective perception, and spatial working range in the selection process of airline pilots; processes and evaluates uncertain, vague, and subjective data with fuzzy logic techniques (fuzzy TOPSIS, fuzzy VIKOR, and fuzzy PROMETHEE) and enables the ranking of candidates; transforms non-numerical linguistic terms into fuzzy mathematical models; ensures the integration of qualitative and quantitative data and the transformation of uncertain data into reliable results in pilot selections; facilitates the successful ranking of candidate pilots and the selection of competent personnel for safe operations by companies; improves the recruitment process of pilots and utilizes an artificial intelligence-based decision-making model application in their selection

Systems and methods are described for scheduling and / or resource provisioning for foundation model applications. The scheduler may involve determining a slack from a performance target and an amount of consumed resources for a workflow request; determining an available resource for at least one replica; selecting a replica, from the at least one replica, associated with a maximum amount of the available resource; and routing at least one machine learning model to a task queue associated with the selected replica for execution. The resource provisioner may involve determining a remaining slack from an associated performance target and an amount of consumed resources for a workflow request; determining a remaining resource to complete the workflow request; tracking a slack violation amount when the remaining resource exceeds the remaining slack; and increasing a number of replicas processing the workflow request based at least on the slack violation amount.

The application discloses a kind of deep neural network model application development systems based on deep learning, it is related to artificial intelligence technical field, the present application aims at solving the problems of gradient distortion caused by sample shunt in training and algorithm logic and bottom hardware load disconnection.System includes: saliency probe module, for generating sample saliency score;Bottom hardware direct connection controller, real-time monitoring hardware state and triggering computing power rescheduling instruction;Dynamic calculation path engine, according to the score, sample is distributed to full or sparse calculation path;Gradient alignment compensation module, scale compensation is carried out to sparse path gradient.Through the above-mentioned module cooperation, the present application constructs the real-time feedback link between algorithm and hardware, eliminates gradient magnitude fluctuation, improves hardware throughput while ensuring the stability and precision of model training.

This invention relates to the technical fields of artificial intelligence and distributed privacy computing, specifically to a federated learning global model correction method based on adaptive filtering of generated data. The method is applied to a server, which, along with multiple clients, constitutes a federated learning system. The method includes: performing federated learning iterations with the multiple clients to obtain a target classification model; wherein the t-th iteration in the federated learning iteration is as follows: broadcasting the (t-1)-th round main classification model and target generator to the clients; the clients freeze the generator, train the classification model and discriminator, and then report the results; the server aggregates the classification model parameters and trains the generator using the multi-discriminator parameters reported by the clients; then, the generator outputs a pseudo-sample set; and weighted knowledge distillation is performed using the aggregated model as the teacher and the previous round's main model as the student to obtain the current round's main classification model. This invention can improve the model application effect of the final global model output by the federated learning system.

The present application relates to the field of artificial intelligence, and particularly relates to a visual artificial intelligence algorithm arrangement method and device, equipment and medium. The method comprises the following steps: developing an operator to solidify various inputs, outputs, training, reasoning and evaluation in an artificial intelligence process into an operator component; loading the operator component to a visual arrangement page built based on a react framework, and receiving, by the visual arrangement page, combination and configuration of the operator component by a user to generate a target model; receiving, by the visual arrangement page, training parameter configuration of the target model by the user, and calling a container to train the target model based on the training parameter configuration; and publishing the trained target model into a model application to provide model service externally. The scheme of the present application enables the user to realize visual operation in the arrangement of training, evaluation, reasoning and the like, reduces understanding and application cost, and improves model production efficiency.

The invention discloses a method for predicting vehicle dynamic response under settlement deformation driven by physical data, and belongs to the technical field of rail transit engineering.The method comprises the following steps that S1, a coupling dynamics database is constructed; s2, data preprocessing; s3, designing a model architecture; s4, performing model training; and S5, model application. According to the method for predicting the dynamic response of the vehicle under the settlement deformation driven by the physical data, the dynamic response of a plurality of parts of the vehicle can be rapidly predicted, the method fits the actual working condition, it is ensured that the prediction result conforms to the physical law, the model interpretability is improved, the prediction precision is improved, and meanwhile, the method is expanded to a composite settlement deformation scene and meets the engineering requirements.

A service processing method executed by a computer device includes acquiring prompt information of a target item recalled for a service user; calling an item processing model under a model application inference framework to predict a degree of association of the service user with the target item based on the prompt information of the target item, the item processing model being obtained by training under a model training inference framework and reused in the model application inference framework, and a processing speed of the item processing model in the model application inference framework being higher than a processing speed of the item processing model in the model training inference framework; and performing service processing on the target item based on the degree of association of the service user with the target item.

A method, device and program product for trusted verification of a model application deployed on a public cloud, the method comprising: obtaining trusted benchmark information of the model application; the model application being deployed on the public cloud, the model application comprising a plurality of model services, the plurality of model services comprising at least two first model services deployed on at least one trusted node of the public cloud, the trusted benchmark information comprising service baseline values of the at least two first model services respectively and a first association relationship between the plurality of model services; obtaining, from the trusted node, a service measurement value of each first model service, and obtaining, from nodes corresponding to respective model services in the plurality of model services, a second association relationship between the plurality of model services; and performing trusted verification of the model application based on the trusted benchmark information, all service measurement values and the second association relationship. Thus, comprehensive and complete trusted verification of the plurality of model services in the whole process of the model application deployed on the public cloud is achieved.

The application discloses a kind of multi-modal fusion's power equipment defect knowledge graph construction method and system, it is related to digital model application technical field, the method of the present application includes multi-modal data fusion, knowledge graph construction and graph reasoning early warning, first pass through multidimensional feature vector, calculate cross-modal correlation strength index, filter out relevant unique code, extract effective dimension, construct power multidimensional cross-modal fusion feature vector, secondly, extract entity and attribute from fusion feature vector, construct initial knowledge graph;Based on motor equipment similarity law, supplement initial graph, form strengthened knowledge graph, finally according to current equipment inspection data matching to staff prompt relevant information, defect early warning, the present application improves feature representation ability by multi-modal data deep fusion, relies on similarity law to supplement and optimize the integrity of knowledge graph, finally through graph reasoning early warning, improve the effect of power equipment defect management.

This invention relates to a method and system for automated review of water conservancy models based on a rule engine, belonging to the field of model application and automated review technology. The invention constructs a basic review rule library for water conservancy models containing general rules and generates a complete set of dedicated review rules through rule instantiation; it utilizes a large water conservancy model to perform multi-format semantic parsing of the original submitted materials, and automatically generates standardized model operation strategies based on the large model parsing results and verification indicators in the dedicated review rule set, driving the automated trial operation of the water conservancy model; the rule engine collects real-time data on the trial operation status and verification indicators of the water conservancy model, calls the dedicated review rule set bound to the model, and executes rule judgments in parallel by review dimension; if the review is successful, a standardized review report is generated; if the review fails, anomalies are automatically marked and a manual review process is triggered. This invention significantly improves the efficiency and accuracy of water conservancy model review.