Three-dimensional scene application method based on large language model
Through a method based on a large language model, using the propt UI interface and langchain framework, the problem of lack of scene editor resources is solved, and the user's ability to generate and edit three-dimensional scene resources is realized in real time.
Patent Information
- Application Number
- CN202411186734.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing scene editor resources are scarce, unable to meet the needs of users’ three-dimensional scenes, and it is impossible to realize that users generate resources based on their ideas in real time.
Through a method based on the large language model, the target template is generated using the propt UI interface, and sent to the target large language model connected to the langchain framework, the target response is returned, and it is rendered to the UI resource display interface for changes and resource addition and deletion operations, from the preset Select the target resource from the model library and import it into the editing scene for editing.
It effectively solves the problem of lack of resources in the scene editor, allows users to generate resources in real time according to their own ideas, and realizes the combination of a large language model and a general scene editor.
Smart Images

Figure CN119065649A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of big language technology, and in particular to a three-dimensional scene application method based on a big language model. Background Art
[0002] With the development of artificial intelligence technology, Wensheng graphs and Wensheng models have achieved significant technological breakthroughs and have shown great potential in practical applications. Many Wensheng graph and Wensheng model algorithms have been able to generate high-quality resources and can be customized according to user input. It is possible to generate resources in real time through AI and use the generated resources in the editor for scene editing. However, the existing scene editors currently use existing resources, which are scarce and cannot meet the user's three-dimensional scene needs. Therefore, how to solve the problem of scarce resources in scene editors and how to allow users to generate resources in real time according to their own ideas have become the focus of current research.
[0003] Therefore, the present invention provides a three-dimensional scene application method based on a large language model. Summary of the invention
[0004] The present invention provides a three-dimensional scene application method based on a large language model, which is used for returning a target response by transmitting a target template generated by using a promptUI interface to a target large language model based on a langchain framework; rendering the target response to a UI resource display interface for modification and resource addition and deletion operations; selecting a target resource from a preset model library and importing it into an editing scene for editing, thereby realizing the combination of a large language model and a general scene editor, effectively solving the problem of resource shortage in the scene editor, and also allowing users to generate resources in real time according to their own ideas.
[0005] The present invention provides a three-dimensional scene application method based on a large language model, comprising: Step 1: Select a target large language model based on the current model application requirements, and connect to the target large language model based on the langchain framework; Step 2: Generate a target template by using the prompt UI interface, transmit it to the target large language model, and return the target response; Step 3: Render the target response to the resource display interface of the UI. After the target user changes the resource generation situation, the satisfactory resource is imported into the preset model library; Step 4: Select target resources from the preset model library according to current user needs and import them into the editing scene for editing, and export the edited scene.
[0006] Preferably, a target large language model is selected based on the current model application requirements, and the target large language model is docked based on the langchain framework, including: Get the largest language model list based on the current editor requirements and langchain support requirements; If there is only a single large language model in the first large language model list, use the current large language model as the target large language model; Otherwise, performing performance evaluation on all large language models in the first large language model list based on the key evaluation indicator to obtain a first evaluation coefficient; Select the large language model with the highest first evaluation coefficient as the target large language model; The target large language model is connected based on the langchain framework, and a target function is created to encapsulate the call to the target large language model.
[0007] Preferably, the key evaluation indicators include model effectiveness, domain specificity, data security and cost-effectiveness.
[0008] Preferably, the target template is generated by using the prompt UI interface, transmitted to the target large language model, and the target response is returned, including: Step 11: Use the set technology stack to build the prompt UI interface based on the UI interface design requirements; Step 12: The user enters custom text in the parameter input box of the prompt UI interface, and when an error input prompt appears, the user changes the current custom text according to the error prompt information and re-enters the text; Step 13: The user enters the number of custom resources to be generated in the quantity input box of the prompt UI interface; Step 14: The user selects a custom resource type from the resource type selection list in the prompt UI interface; Step 15: The user selects a custom resource form from the resource form selection list in the prompt UI interface; Step 16: After completing all custom inputs, the user submits the input and generates the target template; Step 17: The target function receives the target template, generates a target request, transmits it to the target large language model, and returns a target response.
[0009] Preferably, it also includes: Obtain the required intermediate result parameters required by the target large language model to process the target request; Check whether the required intermediate result parameters exist in the cache; If the required intermediate result parameter exists in the cache, the parameter result of the required intermediate result parameter stored in the cache is reused, and a result reuse record is generated; If the required intermediate result parameter does not exist in the cache, the parameter result of the required intermediate result parameter is calculated according to the preset calculation process and added to the cache according to the set cache logic.
[0010] Preferably, it also includes: Using a preset monitoring tool to regularly monitor cache performance to obtain first performance indicator data; Extracting the hit rate, access delay, and space difference between the first storage space and the total available space of the cache at the current moment from the first performance indicator data and comprehensively analyzing them to obtain a first performance evaluation coefficient; Acquire a first cache invalidation strategy that matches the first performance evaluation coefficient as a new cache invalidation strategy for the cache; Extracting a first storage space amount cached at a current moment from the first performance indicator data; Calculating a first ratio of the first storage space amount to the total available space; If the first ratio is not greater than the set ratio threshold, it is determined that the current cache storage space usage is normal; If the first ratio is greater than a set ratio threshold, extracting a first storage space amount within a preset time period from the first performance indicator data and constructing a storage space change curve diagram according to the time series; Extracting the curve characteristics of the storage space change curve diagram and performing weighted average calculation to obtain a change characteristic coefficient; The change characteristic coefficient is combined with the first ratio, the set ratio threshold and the new cache invalidation strategy to calculate the space adjustment coefficient; The calculation formula of the space adjustment coefficient is as follows: ; In the formula, k represents the spatial adjustment coefficient; It is represented as the adjustment weight of the cache storage space amount by the i-th new cache invalidation strategy, where Indicates the number of new cache invalidation strategies currently used by the cache; It is expressed as the ratio of the change in cache storage space per unit time caused by using the i-th new cache invalidation strategy to the total available space in the corresponding period; It is expressed as the weight of the change trend of cache storage space on the adjustment of total available space; It is expressed as the coefficient of variation characteristic; It is expressed as setting the ratio threshold; It is expressed as the first ratio; It is represented by the weight of the adjustment effect of the current cache storage space usage on the total available space; Expressed as calculated loss factor; The total available space of the cache is adjusted using the space adjustment coefficient.
[0011] Preferably, the target response is rendered to a resource display interface of a UI, and after the target user changes the resource generation situation, the satisfactory resource is imported into a preset model library, including: Parsing the received target response into a data format to obtain first data; If there is error data in the first data, the error is handled by matching a corresponding error handling strategy according to the error type of the error data; If there is no erroneous data in the first data, extract key information from the first data according to the current user application requirements, and obtain first key data after data cleaning; If the data format of the current first key data meets the application requirement format, outputting the first key data as the first key resource data; If the current data format of the first key data does not meet the application requirement format, the data format of the first key data is uniformly converted into the application requirement format and then output as the first key resource data; Rendering the first key resource data into a first resource, and displaying it on a resource display interface of a UI; The user uses the resource deletion button to delete the first resource that does not meet the user's needs in the resource display interface; The user uses the resource import button to import the first resource that meets the user's needs in the resource display interface into the preset model library.
[0012] Preferably, according to the current user demand, a target resource is selected from the preset model library and imported into the editing scene for editing, and the edited scene is exported, including: The user selects a target resource from the preset model library using a preset resource manager; Associating the target resource with a matching object in the editing scene, and importing the target resource into the editing scene for rendering and display; When the resource display effect or scene display effect does not meet the user's expectations, the user uses the tools and functions provided by the preset editing interface to edit and adjust the target resource or editing scene displayed in the current editing scene to obtain the target editing scene; Exporting the target editing scene and monitoring the export operation; The scene data of the target editing scene is collected and packaged using a setting technology to obtain a target scene file for storage.
[0013] Compared with the prior art, the present invention has the following beneficial effects: By transmitting the target template generated by the prompt UI interface to the target large language model based on the langchain framework, the target response is returned; the target response is rendered to the UI resource display interface for modification and resource addition and deletion operations; the target resource is selected from the preset model library and imported into the editing scene for editing, thereby realizing the combination of the large language model and the general scene editor, effectively solving the resource shortage of the scene editor, and also allowing users to generate resources in real time according to their own ideas.
[0014] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.
[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 The present invention is a flowchart of a three-dimensional scene application method based on a large language model in an embodiment of the present invention. DETAILED DESCRIPTION
[0017] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.
[0018] The embodiment of the present invention provides a three-dimensional scene application method based on a large language model, such as Figure 1 As shown, including: Step 1: Select a target large language model based on the current model application requirements, and connect to the target large language model based on the langchain framework; Step 2: Generate a target template by using the prompt UI interface, transmit it to the target large language model, and return the target response; Step 3: Render the target response to the resource display interface of the UI. After the target user changes the resource generation situation, the satisfactory resource is imported into the preset model library; Step 4: Select target resources from the preset model library according to current user needs and import them into the editing scene for editing, and export the edited scene.
[0019] In this embodiment, the model application requirements refer to the editor requirements and the langchain support requirements, wherein the editor requirements refer to the editor's compatibility requirements for the model; langchain is an open source framework specifically used to build applications based on natural language processing (NLP); a large language model refers to a data model that can process natural language and deep artificial neural network representations, generally using a transformer's decoder-only architecture, such as the GPT series models.
[0020] In this embodiment, the prompt UI interface is a user interface that provides interactive prompts to users, and is intended to enhance the interaction experience between users and applications or systems through intuitive interface design. It is built based on the UI interface design requirements using a set technology stack; the set technology stack is pre-set and includes a front-end and a back-end, wherein the front-end is used to build a user interface, and the back-end is used to process data logic and interaction with a large language model.
[0021] In this embodiment, the UI interface design requirements include prompt parameter input, resource generation quantity input, resource type input, resource form input, resource display, and resource import and deletion; the target response refers to the response generated by the target large language model according to the received target request, wherein the target request is a request constructed by the target function according to the received target template; the target function is used to encapsulate the call to the LLM; the target template is a template generated by the user based on the prompt UI interface; the resource display interface is used to ensure that the response returned by the LLM can be clearly displayed to the user, and the corresponding interface elements can be designed to display this data according to the type of response, such as text and table; the preset model library is established in advance.
[0022] The beneficial effects of the above technical solution are: by transmitting the target template generated by using the prompt UI interface to the target large language model based on the langchain framework, the target response is returned; the target response is rendered to the UI resource display interface for modification and resource addition and deletion operations; the target resource is selected from the preset model library and imported into the editing scene for editing, thereby realizing the combination of the large language model and the general scene editor, effectively solving the resource shortage of the scene editor, and also allowing users to generate resources in real time according to their own ideas.
[0023] The embodiment of the present invention provides a three-dimensional scene application method based on a large language model, which selects a target large language model based on the current model application requirements, and connects the target large language model based on a langchain framework, including: Get the largest language model list based on the current editor requirements and langchain support requirements; If there is only a single large language model in the first large language model list, use the current large language model as the target large language model; Otherwise, performing performance evaluation on all large language models in the first large language model list based on the key evaluation indicator to obtain a first evaluation coefficient; Select the large language model with the highest first evaluation coefficient as the target large language model; The target large language model is connected based on the langchain framework, and a target function is created to encapsulate the call to the target large language model.
[0024] In this embodiment, langchain is an open source framework specifically used to build applications based on natural language processing (NLP); a large language model refers to a data model that can process natural language and deep artificial neural network representation, generally using a transformer decoder-only architecture, such as the GPT series models; the first large language model list is a list constructed from large language models that support editor application requirements and langchain support requirements.
[0025] In this embodiment, the key evaluation indicators include model effectiveness, domain specificity, data security, and cost-effectiveness; the first evaluation coefficient is used to quantify the performance of the large language model, and is obtained by weighted averaging the evaluation indicator values obtained by using the key evaluation indicators to evaluate the performance of the large language model.
[0026] The beneficial effect of the above technical solution is: by docking with a large language model pre-screened based on application requirements based on the langchain framework, the model performance can be ensured to be stable, which is conducive to improving development efficiency and optimizing user experience.
[0027] The embodiment of the present invention provides a three-dimensional scene application method based on a large language model, which generates a target template by using a promptUI interface, transmits it to the target large language model, and returns a target response, including: Step 11: Use the set technology stack to build the prompt UI interface based on the UI interface design requirements; Step 12: The user enters custom text in the parameter input box of the prompt UI interface, and when an error input prompt appears, the user changes the current custom text according to the error prompt information and re-enters the text; Step 13: The user enters the number of custom resources to be generated in the quantity input box of the prompt UI interface; Step 14: The user selects a custom resource type from the resource type selection list in the prompt UI interface; Step 15: The user selects a custom resource form from the resource form selection list in the prompt UI interface; Step 16: After completing all custom inputs, the user submits the input and generates the target template; Step 17: The target function receives the target template, generates a target request, transmits it to the target large language model, and returns a target response.
[0028] The technology stack set in this embodiment is pre-set, including the front-end and the back-end, wherein the front-end is used to build the user interface, and the back-end is used to process data logic and interaction with the large language model; the UI interface design requirements include prompt parameter input, resource generation quantity input, resource type input, resource form input, resource display, and resource import and deletion; resource types include articles, questions and answers, codes, etc.; resource forms include text, pictures, videos, etc.
[0029] In this embodiment, the error prompt information includes the error type, correction suggestions and error location, which is used to remind the user that the content entered in the parameter input box is incorrect; the custom resource generation quantity refers to the number of resources the user wants to generate; the custom resource type refers to the resource type the user wants to generate; the custom resource form refers to the resource form the user wants to generate; the target template refers to the prompt template constructed based on the user's input and selection.
[0030] In this embodiment, the target function is used to encapsulate the call to the LLM; the target request refers to the request generated by the target function according to the received target template; the target response refers to the response generated by the target large language model according to the received target request, which contains a large amount of resource data.
[0031] The beneficial effect of the above technical solution is: by using the prompt UI interface to generate a target template and transmitting it to the target large language model, and returning a target response, data support can be laid for subsequent resource rendering, which is conducive to allowing users to generate resources in real time according to their own ideas.
[0032] An embodiment of the present invention provides a three-dimensional scene application method based on a large language model, further comprising: Obtain the required intermediate result parameters required by the target large language model to process the target request; Check whether the required intermediate result parameters exist in the cache; If the required intermediate result parameter exists in the cache, the parameter result of the required intermediate result parameter stored in the cache is reused, and a result reuse record is generated; If the required intermediate result parameter does not exist in the cache, the parameter result of the required intermediate result parameter is calculated according to the preset calculation process and added to the cache according to the set cache logic.
[0033] In this embodiment, the demand intermediate result parameter refers to the basis of subsequent calculations generated in the calculation process when the large language model processes the request task, such as attention weight, hidden state, etc.; the result reuse record is composed of the reused demand intermediate result parameter, reuse time, number of successful reuses, and number of failed reuses; the preset calculation process is pre-set; setting the cache logic means adding the demand intermediate result parameter to the corresponding cache partition according to the parameter category, and setting the expiration time.
[0034] The beneficial effect of the above technical solution is: by combining a large language model with a cache, the intermediate result parameters can be effectively managed and reused, which can avoid repeated execution of computationally expensive operations, thereby saving computing resources and time, helping the model to respond to requests faster and improving the performance of the large language model.
[0035] An embodiment of the present invention provides a three-dimensional scene application method based on a large language model, further comprising: Using a preset monitoring tool to regularly monitor cache performance to obtain first performance indicator data; Extracting the hit rate, access delay, and space difference between the first storage space and the total available space of the cache at the current moment from the first performance indicator data and comprehensively analyzing them to obtain a first performance evaluation coefficient; Acquire a first cache invalidation strategy that matches the first performance evaluation coefficient as a new cache invalidation strategy for the cache; Extracting a first storage space amount cached at a current moment from the first performance indicator data; Calculating a first ratio of the first storage space amount to the total available space; If the first ratio is not greater than the set ratio threshold, it is determined that the current cache storage space usage is normal; If the first ratio is greater than a set ratio threshold, extracting a first storage space amount within a preset time period from the first performance indicator data and constructing a storage space change curve diagram according to the time series; Extracting the curve characteristics of the storage space change curve diagram and performing weighted average calculation to obtain a change characteristic coefficient; The change characteristic coefficient is combined with the first ratio, the set ratio threshold and the new cache invalidation strategy to calculate the space adjustment coefficient; The calculation formula of the space adjustment coefficient is as follows: ; In the formula, k represents the spatial adjustment coefficient; It is represented as the adjustment weight of the cache storage space amount by the i-th new cache invalidation strategy, where Indicates the number of new cache invalidation strategies currently used by the cache; It is expressed as the ratio of the change in cache storage space per unit time caused by using the i-th new cache invalidation strategy to the total available space in the corresponding period; It is expressed as the weight of the change trend of cache storage space on the adjustment of total available space; It is expressed as the coefficient of variation characteristic; It is expressed as setting the ratio threshold; It is expressed as the first ratio; It is represented by the weight of the adjustment effect of the current cache storage space usage on the total available space; Expressed as calculated loss factor; The total available space of the cache is adjusted using the space adjustment coefficient.
[0036] In this embodiment, the preset monitoring tool is pre-set; the first performance indicator includes a hit rate, an access delay, a first storage space amount, and a space difference between the first storage space amount and the total available space; the first performance evaluation coefficient is used to evaluate the performance of the cache at the current moment, and is obtained by weighted average calculation of the cache hit rate, access delay, and the space difference between the first storage space amount and the total available space; the first cache invalidation strategy is a cache invalidation strategy determined based on the first performance evaluation coefficient, wherein the cache invalidation strategy includes an LRU (least recently used) strategy, an LFU (least frequently used) strategy, and a FIFO (first in, first out) strategy, etc.
[0037] In this embodiment, the first storage space amount refers to the storage space amount currently used by the cache; the total available space refers to the total storage capacity of the cache; the first ratio refers to the ratio of the first storage space amount to the total available space; and the ratio threshold is set in advance.
[0038] In this embodiment, the preset time period is pre-set; the storage space change curve graph is used to visualize the change trend of the amount of cached storage space within the preset time period; the curve characteristics include slope, extreme value, average value and curvature; the change characteristic coefficient is obtained by weighted average calculation of the curve characteristics extracted from the storage space change curve graph, and is used to characterize the degree of change of the amount of cached storage space within the preset time period; the space adjustment coefficient is used to adjust the total available space of the cache.
[0039] The beneficial effect of the above technical solution is: by using monitoring tools to regularly cache performance indicator data and analyzing performance indicator data to promptly discover and solve problems, the cache performance can be effectively tuned, thereby improving the overall performance and response speed of the large language model.
[0040] The embodiment of the present invention provides a three-dimensional scene application method based on a large language model, which renders the target response to a resource display interface of a UI, and after the target user changes the resource generation situation, imports the satisfactory resource into a preset model library, including: Parsing the received target response into a data format to obtain first data; If there is error data in the first data, the error is handled by matching a corresponding error handling strategy according to the error type of the error data; If there is no erroneous data in the first data, extract key information from the first data according to the current user application requirements, and obtain first key data after data cleaning; If the data format of the current first key data meets the application requirement format, outputting the first key data as the first key resource data; If the current data format of the first key data does not meet the application requirement format, the data format of the first key data is uniformly converted into the application requirement format and then output as the first key resource data; Rendering the first key resource data into a first resource, and displaying it on a resource display interface of a UI; The user uses the resource deletion button to delete the first resource that does not meet the user's needs in the resource display interface; The user uses the resource import button to import the first resource that meets the user's needs in the resource display interface into the preset model library.
[0041] In this embodiment, the first data is data obtained by parsing the target response according to a data format that can be processed by the current application; error types include retry request, authentication error, request parameter error, response parsing error, etc.; the error handling strategy is pre-set; the first key data refers to the data obtained after data cleaning of the key information extracted from the first data according to user application requirements, wherein user application requirements include direct text output, best answer output, etc.
[0042] In this embodiment, the target response includes text, confidence, metadata, error code and message; the resource display interface is used to ensure that the response returned by LLM can be clearly displayed to the user, and corresponding interface elements can be designed for display according to the type of response, such as text, table; the application requirement format refers to the format suitable for the current application or service; data format conversion includes encoding conversion, data structure conversion, etc.
[0043] The beneficial effect of the above technical solution is: by rendering the target response to the resource display interface of the UI, the target user deletes the resources that do not meet the user's needs, and imports the resources that meet the user's needs into the preset model library, thereby providing effective resources for the establishment of subsequent editing scenes while improving user participation.
[0044] The embodiment of the present invention provides a 3D scene application method based on a large language model, which selects a target resource from the preset model library according to the current user needs and imports it into the editing scene for editing, and exports the edited scene, including: The user selects a target resource from the preset model library using a preset resource manager; Associating the target resource with a matching object in the editing scene, and importing the target resource into the editing scene for rendering and display; When the resource display effect or scene display effect does not meet the user's expectations, the user uses the tools and functions provided by the preset editing interface to edit and adjust the target resource or editing scene displayed in the current editing scene to obtain the target editing scene; Exporting the target editing scene and monitoring the export operation; The scene data of the target editing scene is collected and packaged using a setting technology to obtain a target scene file for storage.
[0045] In this embodiment, the preset resource manager is pre-established; the preset model library is established based on a comprehensive analysis of application requirements and resource quantities. For example, when the number of resources is huge and needs to be queried frequently, a relational database is used.
[0046] In this embodiment, the preset editing interface is pre-set, including a scene view, a toolbar, a property panel, etc.; the target editing scene is obtained by editing and adjusting the display resources and the editing scene in the editing scene using the preset editing interface; the setting technology is pre-set, and the scene data for the target editing scene is packaged into a file, such as a ZIP compressed package, an HTML file.
[0047] The beneficial effect of the above technical solution is that by allowing users to select target resources from a preset model library and import them into the editing scene for editing, centralized resource management can be achieved, facilitating resource reuse and expansion, thereby improving the construction efficiency and flexibility of the three-dimensional scene.
[0048] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
Claims
1. A three-dimensional scene application method based on a large language model, characterized in that: include: Step 1: Select a target large language model based on the current model application requirements, and connect to the target large language model based on the langchain framework; Step 2: Generate a target template by using the prompt UI interface, transmit it to the target large language model, and return the target response; Step 3: Render the target response to the resource display interface of the UI. After the target user changes the resource generation situation, the satisfactory resource is imported into the preset model library; Step 4: Select target resources from the preset model library according to current user needs and import them into the editing scene for editing, and export the edited scene.
2. A three-dimensional scene application method based on a large language model according to claim 1, characterized in that: Select a target large language model based on the current model application requirements, and connect to the target large language model based on the langchain framework, including: Get the largest language model list based on the current editor requirements and langchain support requirements; If there is only a single large language model in the first large language model list, use the current large language model as the target large language model; Otherwise, performing performance evaluation on all large language models in the first large language model list based on the key evaluation indicator to obtain a first evaluation coefficient; Select the large language model with the highest first evaluation coefficient as the target large language model; The target large language model is connected based on the langchain framework, and a target function is created to encapsulate the call to the target large language model.
3. The three-dimensional scene application method based on a large language model according to claim 1, characterized in that: The key evaluation indicators include model effectiveness, domain specificity, data security, and cost-effectiveness.
4. The three-dimensional scene application method based on a large language model according to claim 2, characterized in that: Generate a target template by using the prompt UI interface, transmit it to the target large language model, and return a target response, including: Step 11: Use the set technology stack to build the prompt UI interface based on the UI interface design requirements; Step 12: The user enters custom text in the parameter input box of the prompt UI interface, and when an error input prompt appears, the user changes the current custom text according to the error prompt information and re-enters the text; Step 13: The user enters the number of custom resources to be generated in the quantity input box of the prompt UI interface; Step 14: The user selects a custom resource type from the resource type selection list in the prompt UI interface; Step 15: The user selects a custom resource form from the resource form selection list in the prompt UI interface; Step 16: After completing all custom inputs, the user submits the input and generates the target template; Step 17: The target function receives the target template, generates a target request, transmits it to the target large language model, and returns a target response.
5. The three-dimensional scene application method based on a large language model according to claim 4, characterized in that: Also includes: Obtain the required intermediate result parameters required by the target large language model to process the target request; Check whether the required intermediate result parameters exist in the cache; If the required intermediate result parameter exists in the cache, the parameter result of the required intermediate result parameter stored in the cache is reused, and a result reuse record is generated; If the required intermediate result parameter does not exist in the cache, the parameter result of the required intermediate result parameter is calculated according to the preset calculation process and added to the cache according to the set cache logic.
6. The three-dimensional scene application method based on a large language model according to claim 5, characterized in that: Also includes: Using a preset monitoring tool to regularly monitor cache performance to obtain first performance indicator data; Extracting the hit rate, access delay, and space difference between the first storage space and the total available space of the cache at the current moment from the first performance indicator data and comprehensively analyzing them to obtain a first performance evaluation coefficient; Acquire a first cache invalidation strategy that matches the first performance evaluation coefficient as a new cache invalidation strategy for the cache; Extracting a first storage space amount cached at a current moment from the first performance indicator data; Calculating a first ratio of the first storage space amount to the total available space; If the first ratio is not greater than the set ratio threshold, it is determined that the current cache storage space usage is normal; If the first ratio is greater than a set ratio threshold, extracting a first storage space amount within a preset time period from the first performance indicator data and constructing a storage space change curve diagram according to the time series; Extracting the curve characteristics of the storage space change curve diagram and performing weighted average calculation to obtain a change characteristic coefficient; The change characteristic coefficient is combined with the first ratio, the set ratio threshold and the new cache invalidation strategy to calculate the space adjustment coefficient; The calculation formula of the space adjustment coefficient is as follows: ; In the formula, k represents the spatial adjustment coefficient; It is represented as the adjustment weight of the cache storage space amount by the i-th new cache invalidation strategy, where ; Indicates the number of new cache invalidation strategies currently used by the cache; It is expressed as the ratio of the change in cache storage space per unit time caused by using the i-th new cache invalidation strategy to the total available space in the corresponding period; It is expressed as the weight of the change trend of cache storage space on the adjustment of total available space; It is expressed as the coefficient of variation characteristic; It is expressed as setting the ratio threshold; It is expressed as the first ratio; It is represented by the weight of the adjustment effect of the current cache storage space usage on the total available space; Expressed as calculated loss factor; The total available space of the cache is adjusted using the space adjustment coefficient.
7. The three-dimensional scene application method based on a large language model according to claim 1, characterized in that: The target response is rendered to the resource display interface of the UI. After the target user changes the resource generation situation, the satisfactory resource is imported into the preset model library, including: Parsing the received target response into a data format to obtain first data; If there is error data in the first data, the error is handled by matching a corresponding error handling strategy according to the error type of the error data; If there is no erroneous data in the first data, extract key information from the first data according to the current user application requirements, and obtain first key data after data cleaning; If the data format of the current first key data meets the application requirement format, outputting the first key data as the first key resource data; If the current data format of the first key data does not meet the application requirement format, the data format of the first key data is uniformly converted into the application requirement format and then output as the first key resource data; Rendering the first key resource data into a first resource, and displaying it on a resource display interface of a UI; The user uses the resource deletion button to delete the first resource that does not meet the user's needs in the resource display interface; The user uses the resource import button to import the first resource that meets the user's needs in the resource display interface into the preset model library.
8. The three-dimensional scene application method based on a large language model according to claim 1, characterized in that: According to the current user needs, target resources are selected from the preset model library and imported into the editing scene for editing, and the edited scene is exported, including: The user selects a target resource from the preset model library using a preset resource manager; Associating the target resource with a matching object in the editing scene, and importing the target resource into the editing scene for rendering and display; When the resource display effect or scene display effect does not meet the user's expectations, the user uses the tools and functions provided by the preset editing interface to edit and adjust the target resource or editing scene displayed in the current editing scene to obtain the target editing scene; Exporting the target editing scene and monitoring the export operation; The scene data of the target editing scene is collected and packaged using a setting technology to obtain a target scene file for storage.
Citation Information
Patent Citations
Natural language analysis method and device, electronic equipment and storage medium
CN117391095A
Data cache node management method
CN117472279A
Workpiece three-dimensional model design generation method and system based on large language model
CN118012416A
Digital twin XR low-code development platform method
CN118093047A
Cited By
3D application generation method and device and electronic equipment
CN121116279A
