A pre-training method, device, storage medium, and electronic device for a large model

Abstract

This specification discloses a method, device, storage medium and electronic device for pre-training a large model, by pre-determining a number of preset geometric primitives and the first description texts corresponding to each geometric primitive, and determining the second description text of the constructed geometric structure according to the relationship between each geometric primitive and the corresponding first description text. Then, the property data of the geometric structure is determined, thereby determining the description text of the geometric structure, and the large model is pre-trained with the description text as a training sample. By determining the description text of a single geometric primitive and the relationship between the geometric primitives, the second description text of the constructed geometric structure is determined, and the geometric structure is efficiently converted into text that can be understood by the large model, which reduces the complexity of converting the geometric structure into text, provides clearer structured data for the large model, and reduces the difficulty of training the large model that can perform generative design.

Description

Technical Field

[0001] This specification relates to the field of computers, and particularly to a pre-training method, device, storage medium, and electronic device for large models. Background Art

[0002] Currently, with the development of computer technology, large language models (LLMs) have gradually become important tools in generative design due to their advantages in logical reasoning. Current generative design is widely used in fields such as architecture, engineering, manufacturing, and product design, greatly improving the efficiency of optimizing or generating complex geometric structures. However, current LLMs are more inclined to process natural language and cannot directly understand and generate geometric structures.

[0003] In the prior art, in order to achieve generative design through LLMs, multi-modal LLMs are usually designed and trained for the geometric structures to be generated to generate or understand data information such as text and images. However, multi-modal LLMs are often highly complex and difficult to train. Based on this, this application provides a pre-training method, device, storage medium, and electronic device for large models. Summary of the Invention

[0004] This specification provides a pre-training method, device, storage medium, and electronic device for large models to partially solve the above problems existing in the prior art.

[0005] This specification adopts the following technical solutions:

[0006] A pre-training method for a large model, comprising:

[0007] Determine a number of preset geometric primitives, and determine a first description text corresponding to the geometric primitive, where the first description text is used to describe the geometric shape and pose of the corresponding geometric primitive;

[0008] Construct a preset number of geometric structures according to at least two of the geometric primitives;

[0009] For each geometric structure, determine a second description text of the geometric structure according to the first description text corresponding to the geometric primitive that constructs the geometric structure;

[0010] Determine the property data of the geometric structure, splice the second description text and the property data, and determine the target description text of the geometric structure;

[0011] Use the preset number of target description texts as training samples to pre-train the large model to be trained until the large model reaches a preset pre-training end condition.

[0012] Optionally, determining the first description text corresponding to the geometric primitive specifically includes:

[0013] For each geometric primitive, determining the geometric shape of the geometric primitive and the pose of the geometric primitive in a preset space;

[0014] According to the geometric shape and the pose, representing the geometric primitive by an implicit function, and using the implicit function as the first description text of the geometric primitive.

[0015] Optionally, constructing a geometric structure according to at least two of the geometric primitives, specifically including:

[0016] Performing at least one Boolean operation on the at least two geometric primitives, and determining the constructed geometric structure according to the operation result.

[0017] Optionally, determining the second description text of the geometric structure according to the first description texts corresponding to the at least two geometric primitives, specifically including:

[0018] Regarding the at least two geometric primitives as leaf nodes, determining the relationship between the leaf nodes and their corresponding parent nodes according to the at least one Boolean operation performed, and constructing a binary tree corresponding to the geometric structure;

[0019] Determining the second description text of the geometric structure according to the binary tree and the first description texts corresponding to the at least two geometric primitives.

[0020] Optionally, the property data at least includes at least one of the mechanical property, optical property, and thermal property of the geometric structure, and the geometric property of the geometric structure.

[0021] Optionally, determining the property data of the geometric structure, specifically including:

[0022] Determining the second description text of the geometric structure, determining the geometric shape and geometric property of the geometric structure;

[0023] According to the geometric shape and the geometric property, simulating to obtain the property data of the geometric structure through a preset simulation software.

[0024] Optionally, the trained large model is used to generate a corresponding geometric structure according to the input property data; or,

[0025] Determining the property data of the geometric structure corresponding to the input second description text according to the input second description text.

[0026] This specification provides a pre-training device for a large model, including:

[0027] A first determination module, configured to determine a plurality of preset geometric primitives and determine a first description text corresponding to the geometric primitive, where the first description text is used to describe the geometric shape and pose of the corresponding geometric primitive;

[0028] A construction module, configured to construct a preset number of geometric structures according to at least two of the geometric primitives;

[0029] A second determination module, configured to, for each geometric structure, determine a second description text of the geometric structure according to the first description text corresponding to the geometric primitive that constructs the geometric structure;

[0030] A splicing module, configured to determine property data of the geometric structure, splice the second description text and the property data, and determine a target description text of the geometric structure;

[0031] A pre-training module, configured to use the preset number of description texts as training samples to perform self-supervised training on a large model to be trained until the large model reaches a preset pre-training end condition.

[0032] This specification provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above-mentioned pre-training method of the large model is implemented.

[0033] This specification provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above-mentioned pre-training method of the large model is implemented.

[0034] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0035] In the pre-training method of the large model provided in this specification, by pre-determining a plurality of preset geometric primitives and the first description texts corresponding to the geometric primitives, then constructing a preset number of geometric structures according to at least two geometric primitives, and determining the second description text of the geometric structure according to the first description text of each geometric primitive, then determining the property data of the geometric structure, and splicing the second description text and the determined property data, so as to determine the description text of the geometric structure, and using the description text as a training sample to pre-train the large model.

[0036] In the above method, by determining the description text of a single geometric primitive and the relationship between geometric primitives, the second description text of the constructed geometric structure is determined, realizing an efficient conversion of the geometric structure into text that can be understood by the large language model, reducing the complexity of converting the geometric structure into text, providing clearer structured data for the large language model, and reducing the training difficulty of the large language model that can perform generative design. Description of the Drawings

[0037] The drawings described herein are provided to further understand the present specification and form a part of the present specification. The illustrative embodiments of the present specification and their descriptions are used to explain the present specification and do not constitute an improper limitation of the present specification. In the drawings:

[0038] Figure 1 It is a schematic flow chart of a pre-training method for a large model provided by the present specification;

[0039] Figure 2 It is a schematic diagram of a geometric structure construction process provided by the present specification;

[0040] Figure 3 It is a schematic diagram of a binary tree corresponding to a geometric structure provided by the present specification;

[0041] Figure 4 It is a schematic diagram of a pre-training device for a large model provided by the present specification;

[0042] Figure 5 It is provided by the present specification corresponding to Figure 1 a schematic diagram of the structure of an electronic device. Detailed Embodiments

[0043] To make the purpose, technical solutions and advantages of the present specification clearer, the technical solutions of the present specification will be clearly and completely described below in conjunction with the specific embodiments of the present specification and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present specification, rather than all the embodiments. Based on the embodiments in the present specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0044] Geometric structures play extremely important roles in real life. They can not only help us understand and describe the world, but also find wide applications in various practical engineering, design, and scientific research fields, such as architectural design, mechanical engineering, electronics, biomedicine, etc. However, traditional design methods are often limited by designers' experience and personal perspectives. Especially in the field of 3D printing technology, designers still face great challenges when designing geometric structures. For example, in the field of 3D printing technology, designers usually design porous structures by combining multiple geometric structures. However, porous structures usually have complex geometric shapes, containing a large number of tiny holes or channels. Traditional tools often require a large amount of time and effort when creating and editing such complex geometric structures. Moreover, when simulating the behavior of porous materials, such as the flow of fluids through porous media, heat conduction, etc., highly accurate models and a large amount of computing resources are needed. Traditional finite element analysis (FEA) or other numerical simulation methods may be difficult to effectively handle such complex geometric details. Therefore, a method that uses algorithms and automated processes to explore various geometric structure design possibilities, namely generative design, has gradually become an important method for designing complex geometric structures.

[0045] With the development of computer technology, large language models (LLMs) have been widely applied in various fields. Moreover, due to the advantages of LLMs in logical reasoning, they have gradually become important tools in generative design. However, current LLMs are more inclined to process natural language and are difficult to directly understand and generate geometric structures. Therefore, currently, in order to achieve generative design through LLMs, multimodal LLMs are usually designed and trained for the geometric structures to be generated to generate or understand data information such as text and images. However, multimodal LLMs often have high complexity and high training difficulty.

[0046] In addition, currently, geometric structures can also be converted into mathematical representations and described by data tools such as geometric equations, matrices, and vectors to describe geometric shapes, positions, etc., thus realizing the conversion of geometric structures into text. However, as the geometric structures to be generated by LLMs or the geometric structures input into LLMs become more and more complex, on the premise of ensuring the conversion accuracy of geometric structures, the length of the converted text increases exponentially. And the input character length and output character length of LLMs are limited. When the length of the converted text is greater than the input character length and output character length of LLMs, it will affect the output of LLMs. Based on this, this specification provides a pre-training method, device, storage medium, and electronic device for large models. Before the start of model training, the geometric structure is converted into a one-dimensional descriptive text, and the large model is pre-trained through the descriptive text, enabling the LLM, which is good at natural language processing, to parse and generate geometric structures, that is, to learn the text description method of geometric structures and understand the properties of each geometric structure.

[0047] The technical solutions provided in each embodiment of this specification will be described in detail below with reference to the accompanying drawings.

[0048] Figure 1 The following is a schematic flowchart of a pre-training method for a large model provided in an embodiment of this specification, including the following steps:

[0049] S100: Determine a number of preset geometric primitives, and determine a first description text corresponding to the geometric primitive, where the first description text is used to describe the geometric shape and pose of the corresponding geometric primitive.

[0050] In one or more embodiments of this specification, there is no limitation on the specific device that executes the pre-training method of the large model. For example, a mobile terminal and a server, etc. However, since subsequent steps involve model pre-training, prompt word splicing, etc., these steps are generally executed by the server. Therefore, in the following of this specification, the server executing the geometric structure evaluation method is taken as an example for description. Among them, the server can be a single device, or composed of multiple devices. For example, a distributed server, a server for cloud services, etc. This specification does not limit this.

[0051] As technology develops, the required geometric structures are becoming more and more complex. However, no matter how complex the geometric structure is, the geometric structure can be regarded as composed of basic geometric primitives. Based on this, in order to facilitate the subsequent determination of the description text of complex geometric structures, the server can pre-determine geometric primitives and determine the first description text of each geometric primitive.

[0052] Specifically, the server determines a number of preset geometric primitives and determines a first description text for describing the geometric shape and pose of each geometric primitive. It should be noted that in one or more embodiments of this specification, there is no limitation on the specific shape of the determined geometric primitives, which can be determined according to the geometric structure that the large model to be trained needs to understand. For example, if the geometric structure is two-dimensional, each geometric primitive is a two-dimensional geometric primitive, such as a triangle, a square, a rectangle, a circle, etc. If the geometric structure to be described is three-dimensional, each geometric primitive is a three-dimensional geometric primitive, such as a sphere, a cylinder, etc. This specification does not limit this, and it can be set according to actual needs. Among them, the shape and / or pose of each geometric primitive is different from that of other geometric primitives. That is, if the geometric shapes of two geometric primitives are exactly the same, but their poses are different, these two geometric primitives are different geometric primitives. In addition, if the poses of two geometric primitives are exactly the same, but their geometric shapes are different, it also means that the two geometric primitives are different geometric primitives.

[0053] It should be noted that in one or more embodiments of this specification, there is no limitation on the method by which the server determines the preset geometric primitive. It can be to pre-generate a basic geometric primitive library and then extract geometric primitives from the basic geometric primitive library. It can also extract geometric primitives from the database of generative design software. Since there are many methods that can be adopted, this specification does not list them one by one, and it can be set according to actual needs.

[0054] In addition, in one or more embodiments of this specification, there is also no limitation on the specific method by which the server determines the first description text of each geometric primitive. For each geometric primitive, the geometric shape of the geometric primitive and the pose of the geometric primitive in the preset space can be determined, and then, based on the determined geometric shape and pose, the geometric primitive can be represented by an implicit function, and the implicit function can be used as the first description text of the geometric primitive.

[0055] The shape of the geometric primitive can also be represented by a preset identifier, and then, based on the determined pose of the geometric primitive, the identifier and the coordinates of the feature points used to represent the pose of the geometric primitive can be used as the first description text of the geometric primitive. Among them, the feature points of the geometric primitive are the points used to represent the main features of the geometric primitive, such as edge points, corner points, inflection points, etc. For example, if the geometric primitive is a sphere with a center coordinate of (0, 0, 0) and a radius of 1, the first description text of the sphere can be "sphere [center point (0, 0, 0)] [point on the surface (1, 1, 1)]". Of course, other methods can also be used to describe the geometric primitive, such as JavaScript Object Notation (JSON), etc. For example, if the geometric primitive is a sphere, the corresponding first description text can be "function cad_sphere(double center, double radius)", where "double" indicates that the data type of the parameters of the geometric primitive is of the "double" type, "center" indicates the center coordinate of the geometric primitive, and "radius" indicates the radius of the sphere.

[0056] The first description text can represent the position information, size information, and shape of the corresponding geometric structure. Since there are many methods that can be adopted, this specification does not list them one by one, and it can be set according to actual needs.

[0057] S102: Construct a preset number of geometric structures according to at least two of the geometric primitives.

[0058] In order to enable the large model to better understand the geometric structure, the server can construct a preset number of geometric structures according to the geometric primitives obtained in advance for the pre-training of the large model.

[0059] Specifically, the server can select at least two from each geometric primitive for combination to achieve the construction of a geometric structure. For example, Figure 2 as shown Figure 2 Figure 4 shows a geometric structure construction process provided by an embodiment of this specification. Among them, the server randomly selects at least geometric primitives from each geometric primitive, and then combines each geometric primitive according to the pose of the selected geometric primitive in space, so as to determine the constructed geometric structure. Among them, each geometric structure is composed of at least two geometric primitives.

[0060] It should be noted that in one or more embodiments of this specification, there is no limitation on the specific method used by the server to select geometric primitives. The geometric primitives can be determined by random sampling, or can be selected one by one through permutation and combination. Since there are many methods that can be adopted, this specification does not list them one by one and can be set according to actual needs.

[0061] In addition, in one or more embodiments of this specification, there is no limitation on the specific method used by the server to combine geometric primitives. It can be achieved by splicing each geometric primitive to combine each geometric primitive, or can be achieved by Boolean operations. It can be specifically set according to actual needs, and this specification does not limit this.

[0062] Furthermore, in one or more embodiments of this specification, there is no limitation on the specific method used by the server to construct a geometric structure. It can be constructed by combining several geometric primitives, or a geometric structure that has been constructed can be obtained, and then at least one geometric primitive can be selected and combined with the obtained geometric structure to construct a geometric structure. There are many selectable methods, and this specification does not limit this.

[0063] Of course, in one or more embodiments of this specification, there is no limitation on the number of geometric structures determined by the server, which can be set according to actual needs. The number of geometric structures is proportional to the accuracy of the large model.

[0064] S104: For each geometric structure, determine the second description text of the geometric structure according to the first description text corresponding to the geometric primitive that constructs the geometric structure.

[0065] In order to enable the large model to better understand the generated geometric structure, the server can determine the second description text of the geometric structure according to the first description text corresponding to the geometric primitive that constructs the geometric structure.

[0066] Specifically, the server determines the combination method of each geometric primitive when constructing the geometric structure, then determines the description text of the combination method, and determines the second description text of the geometric structure according to the description text of the combination method and the first description text of each geometric primitive. Taking the combination method selected by the server as the intersection in Boolean operations as an example, if the first description texts of each geometric primitive are "function cad_cube(Point&a, Point& b)" and "function cad_cylinder(double a, (a,a), (b,b))", the second description text of the geometric structure can be "functioncad_cube(Point&a, Point& b)+ function cad_cylinder(double a, (a,a), (b,b))", or it can also be "Intersection[function cad_cube(Point&a, Point& b), function cad_cylinder(double a, (a,a), (b,b))]". Of course, the above is only an embodiment provided in this specification. Since there are many available methods, this specification does not limit it and can be set according to actual needs.

[0067] In one or more embodiments of this specification, there is no limitation on the specific method used by the server to determine the description text of the Boolean manipulation descriptor, such as "+", "Intersection", "∩", "function cad_intersection(CadModel geo1, CadModel geo2)", etc., which can be set according to actual needs.

[0068] S106: Determine the property data of the geometric structure, splice the second description text and the property data, and determine the target description text of the geometric structure.

[0069] After the server determines the second description text of the geometric structure, in order to pre-train the large model, it can also first determine the property data of each geometric structure, and then splice the determined property data with the second description text to determine the target description text of each geometric structure, so that the large model can better understand and generate geometric structures.

[0070] Specifically, the server can determine the geometric shapes and geometric properties of each geometric structure according to the second description text of each geometric structure. Then, based on the geometric shapes and geometric properties, through a preset simulation software, the property data of each geometric structure can be simulated. Among them, the property data of each geometric structure includes at least one of the mechanical properties, optical properties, and thermal properties of the geometric structure, as well as the geometric properties of the geometric structure. The geometric properties of the geometric structure include at least one of the volume, surface area, porosity, etc. of the geometric structure. It should be noted that the property data of the geometric structure can be set according to the actual use or user requirements of the geometric structure, and this specification does not limit this.

[0071] In addition, the server can also first generate the corresponding geometric model, and then calculate the corresponding property data according to each geometric model, which can be set according to actual needs, and this specification does not limit this.

[0072] It should be noted that in one or more embodiments of this specification, there is no limitation on the specific method adopted by the server to determine and calculate the property data corresponding to the geometric structure. In addition to obtaining the property data of the geometric structure by simulating through the preset simulation software as described above, other property data can also be obtained through theoretical calculation according to the geometric properties of the geometric structure. It can be set according to actual needs, and this specification does not limit this.

[0073] In addition, in one or more embodiments of this specification, there is also no limitation on the specific method adopted by the server to splice the second description text and the property data. The second description text and the property data can be directly spliced, or the property data can be used to generate prompt words through a trained large language model, and the property data can be converted into a natural language description. Then it is spliced with the second description text to facilitate the large model to understand the target description text. There are many methods that can be adopted, and this specification does not limit this and can be set according to actual needs. Of course, in one or more embodiments of this specification, there is also no limitation on the position of the second description text in the target description text, which can be set according to actual needs.

[0074] S108: Use the preset number of target description texts as training samples to perform self-supervised training on the large model to be trained until the large model reaches the preset pre-training end condition.

[0075] After determining the target description texts of each geometric structure, the pre-training of the large model can be started.

[0076] Specifically, the server inputs each geometric structure into the large model to be trained, enabling the large model to perform pre-training based on each target description text until the large model reaches the preset pre-training end condition, such as the training reaching a preset duration, the accuracy of the large model reaching a preset value, etc. This specification does not limit this and can be set according to actual needs.

[0077] It should be noted that the pre-trained large model is used to generate the corresponding geometric structure based on the input property data, or to determine the property data of the geometric structure corresponding to the input second description text.

[0078] Based on Figure 1 In the large model pre-training method shown, by pre-determining a number of preset geometric primitives and the first description text corresponding to each geometric primitive, then constructing a preset number of geometric structures according to at least two geometric primitives, and determining the second description text of the geometric structure according to the first description text of each geometric primitive, then determining the property data of the geometric structure, and splicing the second description text and the determined property data, so as to determine the description text of the geometric structure, and using this description text as a training sample to pre-train the large model.

[0079] In the above method, by determining the description text of a single geometric primitive and the relationship between geometric primitives, the second description text of the constructed geometric structure is determined, realizing the efficient conversion of geometric structures into text that can be understood by large language models, reducing the complexity of converting geometric structures into text, providing clearer structured data for the large model, and reducing the training difficulty of the large model that can perform generative design.

[0080] In addition, in step S102, the server can also construct a geometric structure by constructing a binary tree, that is, by using solid geometric modeling technology to construct a geometric structure, taking each geometric primitive as a leaf node, and determining the relationship between the leaf node and its corresponding parent node according to at least one Boolean operation performed, constructing a binary tree, and then determining the second description text of the geometric structure according to the binary tree and the first description text corresponding to each geometric primitive.

[0081] As Figure 3 shown, Figure 3 is a schematic diagram of a binary tree corresponding to a geometric structure provided by an embodiment of this specification. This geometric structure includes the union operation of three cylinders.

[0082] When determining the geometric structure, the first description text of each geometric primitive is determined. Each first description text contains the necessary parameters of the corresponding geometric primitive, such as radius, coordinates, etc. The binary tree is mapped to a nested JSON structure in a recursive manner, where each boolean operator represents its child nodes through the models array. Thus, the geometric modeling of the binary tree structure is transformed into structured JSON data, which simplifies the text description of the geometric structure while ensuring the accuracy of the geometric structure. In addition, it should be noted that in one or more embodiments of this specification, when pre-training the same large model, the methods for determining the first description text, the second description text, the description text of the combination method, and the target description text should be unified so that the large model can better understand and generate geometric structures. This ensures the consistency and efficiency of the geometric structure data during storage, transmission, and parsing.

[0083] such as Figure 3 For the binary tree shown, the corresponding second description text in JSON format can be: "function cad_intersection":{"models":[{"cad_cylinder":{"points":[[1,1,1],[2,2,2]],"radius":2.5}},{"cad_cylinder":{"points":[[0,0,0],[1,0,0]],"thickness":2.5}},{"cad_cylinder":{"points":[[1,0,0],[2,1,0]],"radius":2}}]}".

[0084] Based on the same idea of a pre-training method for a large model provided by one or more embodiments of this specification, this specification also provides a corresponding large model pre-training device, such as Figure 4 shown.

[0085] Figure 4 is a schematic diagram of a pre-training device for a large model provided by this specification, specifically including:

[0086] A first determination module 400, configured to determine a plurality of preset geometric primitives and determine the first description text corresponding to the geometric primitives. The first description text is used to describe the geometric shape and pose of the corresponding geometric primitive;

[0087] A construction module 401, configured to construct a preset number of geometric structures according to at least two of the geometric primitives;

[0088] A second determination module 402, configured to, for each geometric structure, determine the second description text of the geometric structure according to the first description text corresponding to the geometric primitives that make up the geometric structure;

[0089] The splicing module 403 is configured to determine the property data of the geometric structure, splice the second description text and the property data, and determine the target description text of the geometric structure;

[0090] The pre-training module 404 is configured to use the preset number of target description texts as training samples to pre-train the large model to be trained until the large model reaches the preset pre-training end condition.

[0091] Optionally, the first determination module 400 is configured to, for each geometric primitive, determine the geometric shape of the geometric primitive and the pose of the geometric primitive in the preset space; according to the geometric shape and the pose, represent the geometric primitive by an implicit function, and use the implicit function as the first description text of the geometric primitive.

[0092] Optionally, the construction module 401 is configured to perform at least one Boolean operation on the at least two geometric primitives, and determine the constructed geometric structure according to the operation result.

[0093] Optionally, the construction module 401 is configured to determine the description text of the Boolean operation operator; use the at least two geometric primitives as leaf nodes, determine the relationship between the leaf nodes and their corresponding parent nodes according to the at least one Boolean operation performed, and construct a binary tree corresponding to the geometric structure; according to the binary tree, the description text of the Boolean operation operator, and the first description texts corresponding to the at least two geometric primitives, determine the second description text of the geometric structure.

[0094] Optionally, the splicing module 403 is configured to determine the second description text of the geometric structure, determine the geometric shape and geometric properties of the geometric structure; according to the geometric shape and the geometric properties, simulate and obtain the property data of the geometric structure through a preset simulation software.

[0095] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above Figure 1 A large model pre-training method provided.

[0096] This specification also provides Figure 5 A schematic structural diagram of an electronic device corresponding to Figure 1 as shown. As Figure 5 shown, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above Figure 1The described large model pre-training method.

[0097] Of course, in addition to the software implementation method, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, and can also be hardware or a logic device.

[0098] In the 1990s, it was obvious to distinguish whether an improvement in a technology was an improvement in hardware (e.g., improvement in circuit structures such as diodes, transistors, switches, etc.) or an improvement in software (improvement in method flows). However, with the development of technology, many improvements in method flows today can be regarded as direct improvements in hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structures by programming the improved method flows into the hardware circuits. Therefore, it cannot be said that an improvement in a method flow cannot be implemented with a hardware entity module. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can program by themselves to "integrate" a digital system on a piece of PLD without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a Hardware Description Language (HDL). And there is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. Currently, the most commonly used ones are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that by simply making a little logical programming of the method flow with the above-mentioned several hardware description languages and programming it into an integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0099] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that, in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to make the controller implement the same function in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0100] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0101] For the convenience of description, the above devices are described by dividing them into various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0102] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, this specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0103] This specification is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0104] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0106] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0107] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

[0108] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0109] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0110] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0111] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media including storage devices.

[0112] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiment.

[0113] The above description is only for the embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the scope of the claims of this specification.

Claims

1. A pre-training method for a large model, characterized in that: include: Determine a plurality of preset geometric primitives, and determine a first description text corresponding to the geometric primitives, wherein the first description text is used to describe the geometric shape and posture of the corresponding geometric primitives; Perform at least one Boolean operation on at least two geometric primitives, and determine to construct a preset number of geometric structures according to the operation result; For each geometric structure, the at least two geometric primitives are taken as leaf nodes, and the relationship between the leaf nodes and their corresponding parent nodes is determined according to the at least one Boolean operation performed, so as to construct a binary tree corresponding to the geometric structure; and a second description text of the geometric structure is determined according to the binary tree and the first description text corresponding to the at least two geometric primitives; Determine property data of the geometric structure, and combine the second description text with the property data to determine a target description text of the geometric structure, wherein the property data at least includes geometric properties and mechanical properties of the geometric structure; The preset number of target description texts are used as training samples to pre-train the large model to be trained until the large model reaches a preset pre-training end condition, wherein the trained large model is used to output corresponding geometric structures or corresponding property data according to user input.

2. The method according to claim 1, characterized in that Determining a first description text corresponding to the geometric primitive specifically includes: For each geometric primitive, determine the geometric shape of the geometric primitive and the position and posture of the geometric primitive in a preset space; According to the geometric shape and the posture, the geometric primitive is represented by an implicit function, and the implicit function is used as the first description text of the geometric primitive.

3. The method according to claim 1, characterized in that The property data also includes at least one of an optical property and a thermal property of the geometric structure.

4. The method according to claim 1, characterized in that Determine the property data of the geometric structure, including: Determine a second description text of the geometric structure, and determine the geometric shape and geometric properties of the geometric structure; According to the geometric shape and the geometric properties, the property data of the geometric structure is simulated by using preset simulation software.

5. The method according to claim 1, characterized in that The trained large model is used to generate a corresponding geometric structure according to the input property data; or, According to the input second description text, property data of the geometric structure corresponding to the second description text is determined.

6. A large model pre-training device, characterized in that: include: A first determination module, used to determine a plurality of preset geometric primitives, and determine a first description text corresponding to the geometric primitives, wherein the first description text is used to describe the geometric shape and posture of the corresponding geometric primitives; A construction module, used for performing at least one Boolean operation on at least two geometric primitives, and determining to construct a preset number of geometric structures according to the operation result; A second determination module is used for, for each geometric structure, taking the at least two geometric primitives as leaf nodes, determining the relationship between the leaf nodes and their corresponding parent nodes according to the at least one Boolean operation performed, and constructing a binary tree corresponding to the geometric structure; and determining a second description text of the geometric structure according to the binary tree and the first description text corresponding to the at least two geometric primitives; a splicing module, used to determine the property data of the geometric structure, splice the second description text and the property data to determine the target description text of the geometric structure, wherein the property data at least includes geometric properties and mechanical properties of the geometric structure; The pre-training module is used to use the preset number of descriptive texts as training samples to perform self-supervised training on the large model to be trained until the large model reaches the preset pre-training end condition, wherein the trained large model is used to output the corresponding geometric structure or corresponding property data according to the user's input.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of claims 1 to 5 is implemented.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 5 when executing the computer program.

Images