Method and device for training a large medical language model

Abstract

The present application discloses a method and device for training a large medical language model, the method comprising: obtaining a training medical data set, the training medical data set comprising a plurality of training prompt words, each training prompt word comprising a training medical text; extracting medical text features of each training medical text; fine-tuning a benchmark large language model using a low-rank adaptive method LoRA, determining the rank for a low-rank matrix according to the medical text features of each training medical text, thereby obtaining a first low-rank matrix and a second low-rank matrix, fusing the first low-rank matrix and the second low-rank matrix with the benchmark large language model, obtaining a fine-tuned large language model; training the fine-tuned large language model using training prompt words, thereby obtaining a trained medical large language model, wherein the trained medical large language model is used to classify medical texts. The present invention not only reduces the computational burden of fine-tuning a large language model, but also improves the generalization ability and adaptability of a large language model.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to methods, devices, media, electronic devices, and computer program products for training large medical language models. Background Art

[0002] In today's big data era, data processing and analysis of medical texts face huge challenges. Although traditional machine learning algorithms, including support vector machines (SVMs), decision trees, random forests, etc., have made contributions to medical text classification, they rely on manual feature extraction and are incapable of classification accuracy, generalization ability, and processing large data sets.

[0003] Currently, medical large language models are used to classify medical texts. However, for different scenarios, fine-tuning is usually required to make them learn prior knowledge in professional fields, thereby improving the accuracy of medical large language models in downstream tasks. However, traditional fine-tuning methods usually require updating all parameters of the medical large language model, which consumes a lot of computing resources and leads to high actual application costs. Summary of the invention

[0004] The embodiments of the present application provide a method for training a large medical language model, an apparatus, a device, a medium, an electronic device, and a computer program product for training a large medical language model.

[0005] In a first aspect, an embodiment of the present application provides a method for training a large medical language model for an electronic device, the method comprising:

[0006] An acquisition step, acquiring a medical training data set, wherein the training medical data set includes a plurality of training prompt words, and each training prompt word includes a training medical text;

[0007] An extraction step, extracting medical text features of the training medical text in each training prompt word;

[0008] A fine-tuning step, using a low-rank adaptive method LoRA to fine-tune the benchmark large language model, wherein the rank for the low-rank matrix is ​​determined according to the medical text features of each training medical text, thereby obtaining a first low-rank matrix and a second low-rank matrix, and the first low-rank matrix and the second low-rank matrix are fused with the benchmark large language model to obtain a fine-tuned large language model;

[0009] In the training step, each training prompt word is used to train the corresponding fine-tuned large language model, thereby obtaining a trained medical large language model.

[0010] Among them, the trained medical large language model is used to classify medical texts.

[0011] In a second aspect, an embodiment of the present application provides a device for training a large medical language model, the device comprising:

[0012] An acquisition unit, which acquires a training medical data set, wherein the training medical data set includes a plurality of training prompt words, and each training prompt word includes a training medical text;

[0013] An extraction unit extracts medical text features of the training medical text in each training prompt word;

[0014] A fine-tuning unit, which uses a low-rank adaptive method LoRA to fine-tune the benchmark large language model, wherein the rank for the low-rank matrix is ​​determined according to the medical text features of each training medical text, thereby obtaining a first low-rank matrix and a second low-rank matrix, and the first low-rank matrix and the second low-rank matrix are fused with the benchmark large language model to obtain a fine-tuned large language model;

[0015] The training unit uses each training prompt word to train the corresponding fine-tuned large language model, thereby obtaining a trained medical large language model.

[0016] Among them, the trained medical large language model is used to classify medical texts.

[0017] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, causes the computer to execute the method for training a large medical language model in the first aspect described above.

[0018] In a fourth aspect, an embodiment of the present application provides an electronic device, comprising: one or more processors; one or more memories; the one or more memories storing one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device executes the method for training a medical large language model in the first aspect mentioned above.

[0019] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising computer executable instructions, which are executed by a processor to implement the method for training a large medical language model in the first aspect.

[0020] In the present invention, LoRA is used to fine-tune the medical large language model, and the rank used for the low-rank matrix is ​​dynamically adjusted according to different medical text contents, so that the medical large language model can flexibly adapt when processing medical text contents of different complexities. For simple tasks, the medical large language model can maintain high efficiency, and when faced with complex tasks, the medical large language model will automatically increase the parameter dimension to improve classification performance. In this way, an efficient medical large language model fine-tuning process is achieved. The present invention not only reduces the computational burden during fine-tuning of the medical large language model, but also improves the generalization ability and adaptability of the medical large language model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 According to an embodiment of the present application, a flow chart of a method for training a large medical language model is shown;

[0022] Figure 2 According to an embodiment of the present application, a schematic diagram of a training prompt word is shown;

[0023] Figure 3 According to an embodiment of the present application, a structural diagram of an apparatus for training a large medical language model is shown;

[0024] Figure 4 According to an embodiment of the present application, a block diagram of an electronic device is shown. DETAILED DESCRIPTION

[0025] The illustrative embodiments of the present application include, but are not limited to, methods, apparatuses, media, electronic devices, and computer program products for training a large medical language model.

[0026] The embodiments of the present application will be described in further detail below in conjunction with the accompanying drawings.

[0027] Figure 1 The method for training a large medical language model in an embodiment of the present application is shown. Figure 4 Electronic equipment shown.

[0028] Medical large language models include GPT-4, Wenyan Yixin, DeepSeek-V2, etc. In this embodiment, DeepSeek-V2 is taken as an example for detailed description.

[0029] In the acquisition step S11, a training medical data set is acquired, the training medical data set includes a plurality of training prompt words, and each training prompt word includes a training medical text.

[0030] Figure 2A schematic diagram of training prompt words is shown, and it can be seen that the training prompt words contain contextual information describing the task, which helps the medical large language model understand the classification target. Specifically, the training prompt words contain training medical text, and the training medical text can be image text, such as medical text content recognized by OCR.

[0031] The categories of medical texts include medical records, reports, etc. After the training prompt words are input into the medical big language model, the categories of the training medical texts will be output.

[0032] In the extraction step S12, the medical text features of the training medical text in each training prompt word are extracted. After each training prompt word in the training medical data set is input into the benchmark large language model, the medical text features of the training medical text in each training prompt word are extracted using the feature extraction network of the large language model.

[0033] It is understandable that, whenever the benchmark large language model receives a training prompt word, it extracts the medical text features of the training medical text in the training prompt word. Figure 2 After the training prompt words are shown, the medical text features of the training medical text will be extracted. The specific method of extracting the medical text features is the same as that of the prior art and will not be described in detail here.

[0034] Medical text features include the complexity of the medical text, such as text length, number of key terms, etc.

[0035] In the fine-tuning step S13, the baseline large language model is fine-tuned using the low-rank adaptive method LoRA, wherein the rank for the low-rank matrix is ​​determined according to the medical text features of each training medical text, thereby obtaining a first low-rank matrix and a second low-rank matrix, and the first low-rank matrix and the second low-rank matrix are fused with the baseline large language model to obtain the fine-tuned large language model.

[0036] The basic idea of ​​LoRA is to reduce the number of parameter updates during fine-tuning by decomposing the weight matrix into a low-rank matrix. The LoRA update formula is:

[0037] in, is the original weight matrix, The first low-rank matrix, are the second lowest rank matrices, and their ranks are obtained according to the medical text features x. It is a learnable scalar used to control the update intensity of the low-rank matrices A(x) and B(x) to the original weight W during the LoRA fine-tuning process. It is a coefficient for adjusting the update amplitude of the low-rank matrix. For example, its value can be obtained through training.

[0038] The role of α is to control the influence of low-rank matrices A(x) and B(x) on the original weights. It can be understood as a regulator to balance the fine-tuning strength. When the value is large, the low-rank matrix has a greater impact on the original weights, and the fine-tuning strength of the model is higher. The role of α is to control the impact of the low-rank matrices A(x) and B(x) on the original weights. It can be understood as a regulator to balance the fine-tuning strength. When the value is large, the low-rank matrix has a greater impact on the original weights and the fine-tuning intensity of the model is higher.

[0039]

[0040] Among them, Rank(x) is the rank of the low-rank matrix obtained for the medical text feature x, represents the minimum dimension of a low-rank matrix, represents the maximum dimension of the low-rank matrix, is the scaling factor, represents the function that generates medical text features x, () represents the floor function, ensuring that the calculated rank is an integer.

[0041] Specifically, Rank(x) represents the rank of the low-rank matrices A(x) and B(x), that is, it determines the dimension size of the low-rank matrices A(x) and B(x). Rank(x) is the optimal rank for a medical text feature. Based on different medical text features (complexity), the number of fine-tuning parameters can be dynamically adjusted. Used to ensure that low-rank matrices are not over-compressed, It is used to prevent the low-rank matrix from becoming too large and increasing the computational overhead. k is used to adjust the influence of medical text features on the rank and determines the proportional relationship between medical text features and rank. It is a function for evaluating medical text features, which generates a numerical value based on specific medical text features.

[0042] As described above, after calculating the optimal rank Rank(x) for the medical text feature x, the first low-rank matrix A(x) and the second low-rank matrix B(x) can be obtained. The first low-rank matrix A(x) and the second low-rank matrix B(x) are fused with the benchmark large language model to obtain a fine-tuned large language model.

[0043] It can be understood that the first low-rank matrix A(x) and the second low-rank matrix B(x) are used to update the weights of the large language model to capture the input medical text features. The first low-rank matrix A(x) and the second low-rank matrix B(x) are used to replace the weights of the fully connected layer of the benchmark large language model. Through the product of the two low-rank matrices, the large language model can adjust the weights to adapt to the input medical text features without increasing a lot of computational overhead.

[0044] It can be understood that the rank for the low-rank matrix is ​​dynamically determined according to different medical text features, thereby obtaining a dynamic first low-rank matrix and a second low-rank matrix.

[0045] In the training step S14, each training cue word is used to train the corresponding fine-tuned large language model, thereby obtaining a trained medical large language model.

[0046] It is understandable that, as described above, a corresponding fine-tuned large language model can be obtained according to a training prompt word. Each training prompt word is used to train the corresponding fine-tuned large language model, and finally a trained medical large language model is obtained.

[0047] Specifically, when a training cue word is input into the fine-tuned large language model, a category is output, and the output category is compared with the true category of the training medical text in the training cue word, so as to adjust the parameters of the fine-tuned large language model. This is repeated until all training cue words have trained the fine-tuned large language model, and the final trained medical large language model is obtained.

[0048] It can be understood that for each training prompt word, the corresponding first low-rank matrix and second low-rank matrix can be dynamically determined, thereby obtaining a dynamically changing fine-tuned large language model.

[0049] The trained medical language model is used to classify medical texts. It is understandable that new prompt words containing medical texts to be classified are input into the trained medical language model to output the categories of the medical texts to be classified. For example, disease classification, symptom recognition, diagnosis suggestion generation, etc.

[0050] It can be understood that in each training cycle, the training data is loaded, forward propagation and back propagation are performed, and the parameters of the model are adjusted using the LoRA update rule.

[0051] Loss calculation and optimization: Calculate the loss of the model on the training medical dataset and optimize the parameters based on the loss backpropagation. Use a dynamic learning rate adjustment strategy to flexibly adjust according to the performance changes of the validation set.

[0052] Real-time monitoring: Monitor key indicators during training (such as loss value, accuracy, etc.) to detect and solve potential problems in a timely manner.

[0053] Test set performance evaluation: After training is completed, the test set is used to perform a final evaluation of the model and calculate the classification accuracy.

[0054] Error Analysis: Step by step analyze the model’s misclassification on specific categories to identify possible areas of improvement and improve the overall performance of the model.

[0055] In the classification process, the DeepSeek-V2 model trained as described above will first understand the medical text content through deep semantic analysis, and then output the category of the document based on the prompt word matching.

[0056] In the present invention, LoRA is used to fine-tune the medical large language model, and the rank used for the low-rank matrix is ​​dynamically adjusted according to different medical text contents, so that the medical large language model can flexibly adapt when processing medical text contents of different complexities. For simple tasks, the medical large language model can maintain high efficiency, and when faced with complex tasks, the medical large language model will automatically increase the parameter dimension to improve classification performance. In this way, an efficient medical large language model fine-tuning process is achieved. The present invention not only reduces the computational burden during fine-tuning of the medical large language model, but also improves the generalization ability and adaptability of the medical large language model.

[0057] The present invention provides a device for training a large medical language model. Figure 3 As shown, the device 30 includes: an acquisition unit 301, which acquires a training medical data set, wherein the training medical data set includes multiple training prompt words, and each training prompt word includes a training medical text; an extraction unit 302, which extracts the medical text features of the training medical text in each training prompt word; a fine-tuning unit 303, which uses a low-rank adaptive method LoRA to fine-tune the benchmark large language model, wherein the rank for the low-rank matrix is ​​determined according to the medical text features of each training medical text, thereby obtaining a first low-rank matrix and a second low-rank matrix, and the first low-rank matrix and the second low-rank matrix are fused with the benchmark large language model to obtain a fine-tuned large language model; a training unit 304, which uses each training prompt word to train the corresponding fine-tuned large language model to obtain a trained medical large language model, wherein the trained medical large language model is used to classify medical texts.

[0058] It is understandable that the acquisition unit 301, the extraction unit 302, the fine-tuning unit 303, and the training unit 304 can be Figure 4 The processor 1404 in the electronic device 1400 having the functions of these modules or units is implemented.

[0059] The present invention also provides a computer-readable storage medium, wherein the storage medium stores instructions, which, when executed on a computer, cause the computer to execute Figure 1 The method shown in .

[0060] The present invention also provides a computer program product, including computer executable instructions, which are executed by the processor 1404 to implement Figure 1 The method shown in .

[0061] Reference now Figure 4 , Figure 4 An example electronic device 1400 according to an embodiment of the present invention is schematically shown. In one embodiment, the electronic device 1400 may include one or more processors 1404, a system control logic 1408 connected to at least one of the processors 1404, a system memory 1412 connected to the system control logic 1408, a non-volatile memory (NVM) 1416 connected to the system control logic 1408, and a network interface 1420 connected to the system control logic 1408.

[0062] In some embodiments, processor 1404 may include one or more single-core or multi-core processors. In some embodiments, processor 1404 may include any combination of general-purpose processors and special-purpose processors (eg, graphics processors, application processors, baseband processors, etc.).

[0063] In some embodiments, system control logic 1408 may include any suitable interface controller to provide any suitable interface to at least one of processors 1404 and / or any suitable device or component in communication with system control logic 1408 .

[0064] In some embodiments, the system control logic 1408 may include one or more memory controllers to provide an interface to the system memory 1412. The system memory 1412 may be used to load and store data and / or instructions. In some embodiments, the memory 1412 of the electronic device 1400 may include any suitable volatile memory, such as a suitable dynamic random access memory (DRAM).

[0065] NVM / memory 1416 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. In some embodiments, NVM / memory 1416 may include any suitable non-volatile memory such as flash memory and / or any suitable non-volatile storage device, such as at least one of a HDD (Hard Disk Drive), a CD (Compact Disc) drive, and a DVD (Digital Versatile Disc) drive.

[0066] NVM / storage 1416 may include a portion of storage resources on a device on which electronic device 1400 is installed, or it may be accessible by the device but not necessarily a portion of the device. For example, NVM / storage 1416 may be accessed over a network via network interface 1420.

[0067] In particular, system memory 1412 and NVM / storage 1416 may include, respectively, a temporary copy and a permanent copy of instructions 1424. Instructions 1424 may include instructions that, when executed by at least one of processors 1404, cause electronic device 1400 to implement the following: Figure 1 In some embodiments, instructions 1424, hardware, firmware, and / or software components thereof may additionally / alternatively be placed in system control logic 1408, network interface 1420, and / or processor 1404.

[0068] The network interface 1420 may include a transceiver for providing a radio interface for the electronic device 1400, thereby communicating with any other suitable device (such as a front-end module, an antenna, etc.) through one or more networks. In some embodiments, the network interface 1420 may be integrated with other components of the electronic device 1400. For example, the network interface 1420 may be integrated with at least one of the processor 1404, the system memory 1412, the NVM / storage 1416, and a firmware device (not shown) having instructions. When at least one of the processors 1404 executes the instructions, the electronic device 1400 implements the following. Figure 1 The method shown.

[0069] The network interface 1420 may further include any suitable hardware and / or firmware to provide a multiple-input multiple-output radio interface. For example, the network interface 1420 may be a network adapter, a wireless network adapter, a telephone modem and / or a wireless modem.

[0070] In one embodiment, at least one of the processors 1404 may be packaged together with logic for one or more controllers of the system control logic 1408 to form a system in package (SiP). In one embodiment, at least one of the processors 1404 may be integrated on the same die with logic for one or more controllers of the system control logic 1408 to form a system on a chip (SoC).

[0071] The electronic device 1400 may further include an input / output (I / O) device 1432. The I / O device 1432 may include a user interface to enable a user to interact with the electronic device 1400; the design of the peripheral component interface enables the peripheral components to interact with the electronic device 1400. In some embodiments, the electronic device 1400 further includes a sensor for determining at least one of an environmental condition and location information related to the electronic device 1400.

[0072] In some embodiments, the user interface may include, but is not limited to, a display (e.g., an LCD display, a touch screen display, etc.), a speaker, a microphone, one or more cameras (e.g., a still image camera and / or a video camera), a flashlight (e.g., an LED flash), and a keyboard.

[0073] The various embodiments of the mechanism disclosed in the present application can be implemented in hardware, software, firmware or a combination of these implementation methods. The embodiments of the present application can be implemented as a computer program or program code executed on a programmable system, which includes at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device and at least one output device.

[0074] Program code can be applied to input instructions to perform each function described in the present application and generate output information. Output information can be applied to one or more output devices in a known manner. For the purpose of the present application, the processing system includes any system with a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC) or a microprocessor.

[0075] Program code can be implemented with high-level programming language or object-oriented programming language to communicate with the processing system. When necessary, program code can also be implemented with assembly language or machine language. In fact, the mechanism described in this application is not limited to the scope of any specific programming language. In either case, the language can be a compiled language or an interpreted language.

[0076] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, the instructions may be distributed over a network or through other computer-readable media. Therefore, a machine-readable medium may include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computer), including, but not limited to, a floppy disk, an optical disk, an optical disc, a read-only memory (CD-ROM), a magneto-optical disk, a read-only memory (ROM), a random access memory (RAM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic card or an optical card, a flash memory, or a tangible machine-readable memory for transmitting information (e.g., a carrier wave, an infrared signal, a digital signal, etc.) using the Internet in an electrical, optical, acoustic, or other form of propagation signal. Therefore, a machine-readable medium includes any type of machine-readable medium suitable for storing or transmitting electronic instructions or information in a form readable by a machine (e.g., a computer).

[0077] In the accompanying drawings, some structural or method features may be shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be required. Instead, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. In addition, the inclusion of structural or method features in a particular figure does not mean that such features are required in all embodiments, and in some embodiments, these features may not be included or may be combined with other features.

[0078] It should be noted that the units / modules mentioned in the various device embodiments of the present application are all logical units / modules. Physically, a logical unit / module can be a physical unit / module, or a part of a physical unit / module, or can be implemented as a combination of multiple physical units / modules. The physical implementation method of these logical units / modules themselves is not the most important. The combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed by the present application. In addition, in order to highlight the innovative part of the present application, the above-mentioned device embodiments of the present application do not introduce units / modules that are not closely related to solving the technical problems proposed by the present application, which does not mean that there are no other units / modules in the above-mentioned device embodiments.

[0079] It should be noted that in the examples and description of this patent, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article or device. In the absence of further restrictions, the elements defined by the sentence "including one" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0080] Although the present application has been illustrated and described with reference to certain preferred embodiments thereof, it will be apparent to those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present application.

Claims

1. A method for training a large medical language model for electronic equipment, characterized in that: The method comprises: An acquisition step of acquiring a training medical data set, wherein the training medical data set includes a plurality of training prompt words, and each training prompt word includes a training medical text; An extraction step, extracting medical text features of the training medical text in each training prompt word; A fine-tuning step, using a low-rank adaptive method LoRA to fine-tune the benchmark large language model, wherein the rank for the low-rank matrix is ​​determined according to the medical text features of each training medical text, thereby obtaining a first low-rank matrix and a second low-rank matrix, and the first low-rank matrix and the second low-rank matrix are fused with the benchmark large language model to obtain a fine-tuned large language model; In the training step, each training prompt word is used to train the corresponding fine-tuned large language model, thereby obtaining a trained medical large language model. Among them, the trained medical language model is used to classify medical texts. According to different medical text features, the rank for the low-rank matrix is ​​dynamically determined, thereby obtaining the first low-rank matrix and the second low-rank matrix dynamically. Wherein, the rank is the best rank for the medical text feature, The medical text features include text complexity, where the rank for the low-rank matrix is ​​determined according to the following function, ， Among them, Rank(x) is the rank of the low-rank matrix obtained for the medical text feature x, r min represents the minimum dimension of a low-rank matrix, r max represents the maximum dimension of the low-rank matrix, k is the scaling factor, f(x) represents the function for generating the medical text feature x, and floor() represents the floor function.

2. The method according to claim 1, characterized in that The new prompt words containing the medical text to be classified are input into the trained medical large language model, so as to output the category of the medical text to be classified.

3. A device for training a large medical language model, characterized in that: The device comprises: An acquisition unit, which acquires a training medical data set, wherein the training medical data set includes a plurality of training prompt words, and each training prompt word includes a training medical text; An extraction unit extracts medical text features of the training medical text in each training prompt word; A fine-tuning unit, which uses a low-rank adaptive method LoRA to fine-tune the benchmark large language model, wherein the rank for the low-rank matrix is ​​determined according to the medical text features of each training medical text, thereby obtaining a first low-rank matrix and a second low-rank matrix, and the first low-rank matrix and the second low-rank matrix are fused with the benchmark large language model to obtain a fine-tuned large language model; The training unit uses each training prompt word to train the corresponding fine-tuned large language model, thereby obtaining a trained medical large language model. Among them, the trained medical language model is used to classify medical texts. According to different medical text features, the rank for the low-rank matrix is ​​dynamically determined, thereby obtaining the first low-rank matrix and the second low-rank matrix dynamically. Wherein, the rank is the best rank for the medical text feature, The medical text features include text complexity, where the rank for the low-rank matrix is ​​determined according to the following function, ， Among them, Rank(x) is the rank of the low-rank matrix obtained for the medical text feature x, r min represents the minimum dimension of a low-rank matrix, r max represents the maximum dimension of the low-rank matrix, k is the scaling factor, f(x) represents the function for generating the medical text feature x, and floor() represents the floor function.

4. A computer-readable storage medium, characterized in that: The storage medium stores instructions, which, when executed on a computer, cause the computer to execute the method for training a large medical language model according to claim 1.

5. An electronic device, characterized in that: include: one or more processors; One or more memories; the one or more memories store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device executes the method for training a large medical language model as described in claim 1.

6. A computer program product comprising computer executable instructions, characterized in that: The instructions are executed by the processor to implement the method for training a large medical language model according to claim 1.

Images