Large language model development method and device, equipment and storage medium

Abstract

The invention discloses a large language model development method and device, equipment and a storage medium, and the method comprises the steps: obtaining an original model selected by a user and an original data set, carrying out the preprocessing of the original data set, obtaining a target data set, and building a large language model according to the original model; searching a training strategy matched with the large language model according to the target data set through an automatic parallel system deployed in the development platform; training the large language model according to the training strategy, and evaluating a training result of the large language model; and if it is determined that the large language model meets a preset performance condition according to the evaluation result, taking the large language model as a target model. According to the technical scheme, non-professional developers can also participate in development of the large language model, the use threshold is lowered, and the model development and iteration speed is increased.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, equipment and storage medium for developing a large language model. Background Art

[0002] In the field of artificial intelligence, especially in the fields of natural language processing (NLP) and computer vision, with the improvement of hardware performance and the popularization of big data, the scale and complexity of AI models are increasing. In particular, the emergence of large language models marks that AI technology has entered a new stage. With its powerful understanding and generation capabilities, large language models have demonstrated excellent performance in text generation, language translation, dialogue systems, etc. However, as the scale of models increases, the development, training, and deployment of these large models also face unprecedented challenges.

[0003] The development of large models involves complex algorithms and a lot of code writing, which is difficult for ordinary developers to master. From data preprocessing, model training to final deployment, the entire process takes a long time and it is difficult to respond quickly to market demand. In the prior art, some platforms for training and deploying large language models are provided, but these platforms usually require users to have a certain technical background to fully utilize their functions. They may not be friendly enough for completely non-technical users, and the computing efficiency is low while the resource consumption cost is also high. Summary of the invention

[0004] The present invention provides a method, apparatus, device and storage medium for developing a large language model, which can enable non-professional developers to participate in the development of a large language model, lower the usage threshold, greatly reduce manual operations and repetitive work, and speed up model development and iteration.

[0005] According to one aspect of the present invention, a method for developing a large language model is provided, which is applied to a large language model development platform, and the method comprises:

[0006] Acquire an original model and an original data set selected by a user, preprocess the original data set to obtain a target data set, and construct a large language model based on the original model;

[0007] Searching for a training strategy that matches the large language model according to the target data set through an automatic parallel system deployed in the development platform;

[0008] According to the training strategy, the large language model is trained, and the training result of the large language model is evaluated;

[0009] An evaluation result corresponding to the large language model is obtained, and if it is determined according to the evaluation result that the large language model meets a preset performance condition, the large language model is used as a target model.

[0010] Optionally, searching for a training strategy matching the large language model according to the target data set through an automatic parallel system deployed in the development platform includes:

[0011] Searching for a plurality of initial strategies matching the large language model according to the target data set through an automatic parallel system deployed in the development platform;

[0012] The multiple initial strategies are pre-pruned and pre-sorted, and then a greedy search is performed according to the processing results and the corresponding priorities to obtain a training strategy that matches the large language model.

[0013] Optionally, evaluating the training result of the large language model includes:

[0014] Obtaining batch data with the same format as the original data set as an evaluation data set;

[0015] The evaluation data set is input into the trained large language model, and the evaluation result corresponding to the large language model is determined according to the processing result of the evaluation data set by the large language model.

[0016] Optionally, determining the evaluation result corresponding to the large language model according to the processing result of the large language model on the evaluation data set includes:

[0017] During the processing of the evaluation data set by the large language model, determining in real time the computational cost and the communication cost consumed by the processing process;

[0018] A scoring result corresponding to the large language model is determined according to the computational cost and the communication cost.

[0019] Optionally, while training the large language model according to the training strategy, the method further includes:

[0020] The indicator monitoring results and training logs corresponding to the large language model training process are displayed to the user through a visual interface;

[0021] The indicator monitoring results include a loss curve and a hardware usage curve corresponding to the large language model training process.

[0022] Optionally, after taking the large language model as the target model, the method further includes:

[0023] Obtaining a prompt word sequence input by a user for the target model;

[0024] The target model is used to infer and output the generation result of the prompt word sequence.

[0025] According to another aspect of the present invention, there is provided a large language model development device, which is applied to a large language model development platform, and the device comprises:

[0026] A data acquisition module is used to acquire the original model and the original data set selected by the user, preprocess the original data set to obtain the target data set, and construct a large language model based on the original model;

[0027] A strategy search module, used to search for a training strategy matching the large language model according to the target data set through an automatic parallel system deployed in the development platform;

[0028] A model training module, used to train the large language model according to the training strategy and evaluate the training result of the large language model;

[0029] The model evaluation module is used to obtain an evaluation result corresponding to the large language model. If it is determined according to the evaluation result that the large language model meets the preset performance conditions, the large language model is used as the target model.

[0030] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0031] at least one processor; and

[0032] a memory communicatively connected to the at least one processor; wherein,

[0033] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for developing a large language model described in any embodiment of the present invention.

[0034] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for developing a large language model described in any embodiment of the present invention when executed.

[0035] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the method for developing a large language model according to any one of the embodiments of the present invention is implemented.

[0036] The technical solution provided by the embodiment of the present invention obtains the original model and the original data set selected by the user, pre-processes the original data set to obtain the target data set, builds a large language model according to the original model, searches for a training strategy matching the large language model according to the target data set through an automatic parallel system deployed in the development platform, trains the large language model according to the training strategy, and evaluates the training result of the large language model. If it is determined according to the evaluation result that the large language model meets the preset performance conditions, the large language model is used as a technical means of the target model. An efficient, convenient, low-code full-process AI large language model development platform is provided, which covers the entire life cycle of the large language model, enables non-professional developers to participate in the development of the large language model, lowers the usage threshold, greatly reduces manual operations and repetitive work, speeds up model development and iteration, significantly shortens the training time of the large language model, reduces the development cost of the large language model, and can dynamically expand computing resources to meet the needs of models of different scales.

[0037] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1a is a flow chart of a method for developing a large language model according to an embodiment of the present invention;

[0040] Figure 1b is a functional schematic diagram corresponding to a large language model development platform provided according to an embodiment of the present invention;

[0041] Figure 2 is a flow chart of another method for developing a large language model according to an embodiment of the present invention;

[0042] Figure 3 is a structural schematic diagram of a large language model development device provided according to an embodiment of the present invention;

[0043] Figure 4 It is a schematic diagram of the structure of an electronic device for implementing the method for developing a large language model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0045] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0046] Figure 1a This is a flow chart of a method for developing a large language model provided by an embodiment of the present invention. This embodiment is applicable to the case where a user independently develops and deploys a large language model through a large language model development platform. The method can be executed by a large language model development device, which can be implemented in the form of hardware and / or software, and can be configured in a large language model development platform. Figure 1a As shown, the method includes:

[0047] Step 110: obtain the original model and the original data set selected by the user, preprocess the original data set to obtain the target data set, and build a large language model based on the original model.

[0048] In this embodiment, specifically, Figure 1b is a functional schematic diagram corresponding to the large language model development platform, such as Figure 1b As shown, the large language model development platform integrates large language model (LLM) pre-training, fine-tuning, human feedback reinforcement learning (RLHF) alignment, model evaluation, and model reasoning functions.

[0049] In this step, the user can independently select the original model (such as the Transformer model, etc.) and the original data set through the large language model development platform. The large language model development platform can automatically process the original model and the original data set, including preprocessing the original data set (such as denoising, normalization, etc.), and constructing a large language model to be trained based on the original model.

[0050] Step 120: Searching for a training strategy that matches the large language model according to the target data set through the automatic parallel system deployed in the development platform.

[0051] In this embodiment, in order to minimize the user usage threshold, an automatic parallel system is pre-integrated in the large language model development platform. The automatic parallel system is used to search for the optimal large language model training strategy under current hardware conditions based on the target data set.

[0052] In a specific embodiment, optionally, the automatic parallel system can search for the optimal large language model training strategy under current hardware conditions based on the computing cost and communication cost consumed by the large language model training process.

[0053] Step 130: Train the large language model according to the training strategy, and evaluate the training result of the large language model.

[0054] In this step, the training strategy searched above can be applied to the large language model training process (such as pre-training, fine-tuning, RLHF alignment, etc.), and after the large language model training is completed, the training results of the large language model are evaluated.

[0055] In a specific embodiment, a preset evaluation data set may be input into a trained large language model, and a user may determine whether the large language model meets the training requirements based on the performance scoring result of the large language model for the evaluation data set.

[0056] Step 140: Obtain an evaluation result corresponding to the large language model. If it is determined according to the evaluation result that the large language model meets a preset performance condition, use the large language model as a target model.

[0057] In this embodiment, if it is determined according to the evaluation result that the large language model meets the preset performance conditions, it can be determined that the development of the large language model is completed, and the large language model is deployed as the final model (ie, the target model).

[0058] The technical solution provided by the embodiment of the present invention obtains the original model and the original data set selected by the user, pre-processes the original data set to obtain the target data set, builds a large language model according to the original model, searches for a training strategy matching the large language model according to the target data set through an automatic parallel system deployed in the development platform, trains the large language model according to the training strategy, and evaluates the training result of the large language model. If it is determined according to the evaluation result that the large language model meets the preset performance conditions, the large language model is used as a technical means of the target model. An efficient, convenient, low-code full-process AI large language model development platform is provided, which covers the entire life cycle of the large language model, enables non-professional developers to participate in the development of the large language model, lowers the usage threshold, greatly reduces manual operations and repetitive work, speeds up model development and iteration, significantly shortens the training time of the large language model, reduces the development cost of the large language model, and can dynamically expand computing resources to meet the needs of models of different scales.

[0059] Figure 2 A flowchart of another method for developing a large language model provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the method includes:

[0060] Step 210: obtain the original model and the original data set selected by the user, preprocess the original data set to obtain the target data set, and construct a large language model based on the original model.

[0061] Step 220: Search for multiple initial strategies that match the large language model according to the target data set through the automatic parallel system deployed in the development platform.

[0062] Step 230: pre-prune and pre-sort the multiple initial strategies, and then perform a greedy search based on the processing results and the corresponding priorities to obtain a training strategy that matches the large language model.

[0063] In this embodiment, after the automatic parallel system searches for multiple initial strategies that support a large language model based on the target data set, it can use prior knowledge to pre-prune and pre-sort all initial strategies, and then perform greedy search according to priority to obtain the final training strategy. Optionally, the automatic parallel system can integrate high-performance operators, low-rank decomposition, data parallelism, tensor parallelism, pipeline parallelism, sequence parallelism, context parallelism, recalculation, gradient accumulation and other functions.

[0064] Pre-pruning is mainly used before model training or in the early stages of training, with the aim of reducing the complexity and computational requirements of the model by removing parts of the model that are expected to contribute less to the final performance. The advantage is that it can reduce the computational burden before model training and help avoid overfitting.

[0065] Pre-sorting can be understood as a pre-processing method for data or features before the model processes the data to optimize the subsequent calculation process. When processing large-scale data sets, pre-sorting the data can significantly improve the efficiency of data processing.

[0066] The advantage of this setting is that it can greatly shorten the strategy search time, improve the training efficiency of large language models and the effectiveness of training results.

[0067] Step 240: Train the large language model according to the training strategy.

[0068] In one implementation of this embodiment, while training the large language model according to the training strategy, it also includes: displaying the indicator monitoring results and training logs corresponding to the large language model training process to the user through a visual interface.

[0069] The indicator monitoring results include a loss curve and a hardware usage curve corresponding to the large language model training process.

[0070] The advantage of this setting is that it allows users to understand the training results of the large language model in real time, thereby lowering the threshold for users to use the large language model development platform.

[0071] Step 250: Obtain batch data in the same format as the original data set as an evaluation data set, input the evaluation data set into the trained large language model, and determine the evaluation result corresponding to the large language model based on the processing result of the evaluation data set by the large language model.

[0072] In one implementation of the present embodiment, the evaluation result corresponding to the large language model is determined according to the processing result of the large language model on the evaluation data set, including: during the processing of the evaluation data set by the large language model, the computing cost and the communication cost consumed by the processing process are determined in real time; based on the computing cost and the communication cost, the scoring result corresponding to the large language model is determined.

[0073] In a specific embodiment, the estimation of computing cost and communication cost should be carried out in two modes, because the computing cost is affected by the implementation of the computing operator, and the implementation of different high-performance operators varies greatly, which is difficult to estimate using a mathematical model, while the communication mode is relatively fixed and can be estimated more conveniently using a mathematical model. Optionally, batch data with the same format as the original data set can be input into the large language model, and the computing cost can be determined by the computing performance monitoring module in the development platform. As for the communication cost, mathematical modeling can be directly performed on all collective communications and point-to-point communications, and then the final communication cost can be estimated.

[0074] In one implementation of this embodiment, if the large language model structure is composed of multiple layers (such as Transformer layers), when evaluating the computational cost, only one layer needs to be selected, thereby improving the development efficiency of the model.

[0075] Step 260: Obtain an evaluation result corresponding to the large language model. If it is determined according to the evaluation result that the large language model meets a preset performance condition, use the large language model as a target model.

[0076] In one implementation of this embodiment, after the large language model is used as the target model, the large language model development platform can also support model reasoning. Specifically, the prompt word sequence input by the user for the target model can be obtained; through the target model, the generation result of the prompt word sequence is inferred and output.

[0077] The technical solution provided by the embodiment of the present invention obtains the original model and the original data set selected by the user, pre-processes the original data set to obtain the target data set, builds a large language model according to the original model, searches for multiple initial strategies matching the large language model according to the target data set through the automatic parallel system deployed in the development platform, pre-prunes and pre-sorts the multiple initial strategies, and then performs greedy search according to the processing results and corresponding priorities to obtain a training strategy matching the large language model, trains the large language model according to the training strategy, obtains batch data with the same format as the original data set as an evaluation data set, inputs the evaluation data set into the trained large language model, determines the evaluation result corresponding to the large language model according to the processing result of the evaluation data set by the large language model, and if it is determined according to the evaluation result that the large language model meets the preset performance conditions, then the large language model is used as a technical means of the target model, so that non-professional developers can also participate in the development of the large language model, lower the use threshold, greatly reduce manual operations and repetitive work, and speed up model development and iteration.

[0078] Figure 3A schematic diagram of the structure of a large language model development device provided by an embodiment of the present invention, wherein the device is applied to a large language model development platform, such as Figure 3 As shown, the device includes: a data acquisition module 310, a strategy search module 320, a model training module 330 and a model evaluation module 340.

[0079] The data acquisition module 310 is used to acquire the original model and the original data set selected by the user, preprocess the original data set to obtain the target data set, and construct a large language model according to the original model;

[0080] A strategy search module 320, configured to search for a training strategy matching a large language model according to the target data set through an automatic parallel system deployed in the development platform;

[0081] A model training module 330, configured to train the large language model according to the training strategy and evaluate the training result of the large language model;

[0082] The model evaluation module 340 is used to obtain an evaluation result corresponding to the large language model. If it is determined according to the evaluation result that the large language model meets the preset performance condition, the large language model is used as the target model.

[0083] The technical solution provided by the embodiment of the present invention obtains the original model and the original data set selected by the user, pre-processes the original data set to obtain the target data set, builds a large language model according to the original model, searches for a training strategy matching the large language model according to the target data set through an automatic parallel system deployed in the development platform, trains the large language model according to the training strategy, and evaluates the training result of the large language model. If it is determined according to the evaluation result that the large language model meets the preset performance conditions, the large language model is used as a technical means of the target model. An efficient, convenient, low-code full-process AI large language model development platform is provided, which covers the entire life cycle of the large language model, enables non-professional developers to participate in the development of the large language model, lowers the usage threshold, greatly reduces manual operations and repetitive work, speeds up model development and iteration, significantly shortens the training time of the large language model, reduces the development cost of the large language model, and can dynamically expand computing resources to meet the needs of models of different scales.

[0084] Based on the above embodiment, the policy search module 320 includes:

[0085] An initial strategy search unit, configured to search for a plurality of initial strategies matching the large language model according to the target data set through an automatic parallel system deployed in the development platform;

[0086] The strategy processing unit is used to pre-prune and pre-sort the multiple initial strategies, and then perform greedy search according to the processing results and corresponding priorities to obtain a training strategy that matches the large language model.

[0087] The model training module 330 includes:

[0088] An evaluation data set acquisition unit, used to acquire batch data in the same format as the original data set as the evaluation data set;

[0089] An evaluation data set input unit, used to input the evaluation data set into the trained large language model, and determine the evaluation result corresponding to the large language model according to the processing result of the evaluation data set by the large language model;

[0090] A cost determination unit, used for determining in real time the computing cost and the communication cost consumed by the processing process of the evaluation data set by the large language model;

[0091] A scoring determination unit, configured to determine a scoring result corresponding to the large language model according to the computational cost and the communication cost;

[0092] A log display unit, used to display the indicator monitoring results and training logs corresponding to the large language model training process to the user through a visual interface;

[0093] The indicator monitoring results include a loss curve and a hardware usage curve corresponding to the large language model training process.

[0094] The device also includes:

[0095] The model inference module is used to obtain a prompt word sequence input by a user for the target model, and infer and output a generation result of the prompt word sequence through the target model.

[0096] The above device can execute the methods provided by all the above embodiments of the present invention, and has the corresponding functional modules and beneficial effects of executing the above methods. For technical details not fully described in the embodiments of the present invention, please refer to the methods provided by all the above embodiments of the present invention.

[0097] Figure 4A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0098] like Figure 4 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0099] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0100] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as a method for developing a large language model.

[0101] In some embodiments, the method for developing a large language model may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for developing a large language model described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the method for developing a large language model in any other appropriate manner (e.g., by means of firmware).

[0102] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0103] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0104] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0105] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0106] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0107] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0108] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0109] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for developing a large language model, characterized in that: Applied to a large language model development platform, the method includes: Acquire an original model and an original data set selected by a user, preprocess the original data set to obtain a target data set, and construct a large language model based on the original model; Searching for a training strategy that matches the large language model according to the target data set through an automatic parallel system deployed in the development platform; According to the training strategy, the large language model is trained, and the training result of the large language model is evaluated; An evaluation result corresponding to the large language model is obtained, and if it is determined according to the evaluation result that the large language model meets a preset performance condition, the large language model is used as a target model.

2. The method according to claim 1, characterized in that By using the automatic parallel system deployed in the development platform, searching for a training strategy matching the large language model according to the target data set includes: Searching for a plurality of initial strategies matching the large language model according to the target data set through an automatic parallel system deployed in the development platform; The multiple initial strategies are pre-pruned and pre-sorted, and then greedy search is performed according to the processing results and corresponding priorities to obtain a training strategy that matches the large language model.

3. The method according to claim 1, characterized in that The training result of the large language model is evaluated, including: Obtaining batch data with the same format as the original data set as an evaluation data set; The evaluation data set is input into the trained large language model, and the evaluation result corresponding to the large language model is determined according to the processing result of the evaluation data set by the large language model.

4. The method according to claim 3, characterized in that Determining an evaluation result corresponding to the large language model according to a processing result of the large language model on the evaluation data set includes: During the processing of the evaluation data set by the large language model, determining in real time the computational cost and the communication cost consumed by the processing process; A scoring result corresponding to the large language model is determined according to the computational cost and the communication cost.

5. The method according to claim 1, characterized in that While training the large language model according to the training strategy, the method further includes: The indicator monitoring results and training logs corresponding to the large language model training process are displayed to the user through a visual interface; The indicator monitoring results include a loss curve and a hardware usage curve corresponding to the large language model training process.

6. The method according to claim 1, characterized in that After the large language model is used as the target model, the method further includes: Obtaining a prompt word sequence input by a user for the target model; The target model is used to infer and output the generation result of the prompt word sequence.

7. A device for developing a large language model, characterized in that: Applied to a large language model development platform, the device comprises: A data acquisition module is used to acquire the original model and the original data set selected by the user, preprocess the original data set to obtain the target data set, and construct a large language model based on the original model; A strategy search module, used to search for a training strategy matching the large language model according to the target data set through an automatic parallel system deployed in the development platform; A model training module, used to train the large language model according to the training strategy and evaluate the training result of the large language model; The model evaluation module is used to obtain an evaluation result corresponding to the large language model. If it is determined according to the evaluation result that the large language model meets the preset performance condition, the large language model is used as the target model.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the method for developing a large language model according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for developing a large language model according to any one of claims 1 to 6 when executed.

10. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method for developing a large language model according to any one of claims 1 to 6.

Images