Large language model processing method, system and medium based on private domain information dialogue
By constructing dialogue instructions and training data from private domain information, and using a general large language model to train a private domain large language model, the problem of high-cost training of large language models for small and medium-sized enterprises is solved, and low-cost, high-performance intelligent question answering capabilities are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGZHOU SAILINGLI TECH CO LTD
- Filing Date
- 2023-09-07
- Publication Date
- 2026-07-24
AI Technical Summary
Small and medium-sized enterprises (SMEs) face difficulties in applying the dialogue and question-answering capabilities of large-scale language models due to the high costs of training and deploying general-purpose large-scale language models and the difficulty in collecting suitable dialogue data for training.
By constructing dialogue instructions based on private domain information, a dialogue corpus database is generated using a general large language model, and historical dialogue data is extracted from it for training. The training of the private domain large language model has a computing power range of 6B-13B, reducing training costs and difficulties.
This technology enables small and medium-sized enterprises to train private domain large language models with intelligent question-answering capabilities at a lower cost, reducing model training and deployment costs, generating accurate and reliable dialogue data, and improving the performance of intelligent question answering.
Smart Images
Figure CN117370508B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent question-and-answer dialogue processing technology, and in particular to a method, system and medium for processing large language models based on private domain information dialogue. Background Technology
[0002] In related technologies, when a target object in a private domain large language model applies the dialogue and question-answering capabilities of a general large language model to its internal private information, the training and deployment of the general large language model involves up to 157 bytes of parameters, resulting in high training and deployment costs that the target object cannot afford. Furthermore, it is difficult to collect relevant dialogue data suitable for training the general large language model in the target object's private domain scenario, thus hindering the target object from applying the dialogue and question-answering capabilities of the large language model. Summary of the Invention
[0003] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a method, system, and medium for processing large language models based on private domain information dialogue, which can effectively reduce the training cost and application difficulty for small and medium-sized enterprises using large language models.
[0004] On one hand, embodiments of the present invention provide a method for processing large language models based on private domain information dialogue, including the following steps:
[0005] The dialogue instructions constructed based on private domain information are input into a general large language model to obtain a database of dialogue corpora to be selected. The dialogue instructions include questions and answers corresponding to the questions. The private domain information includes object information of the scene in which the private domain large language model is located.
[0006] Constructing training instructions based on private domain information;
[0007] Extract historical dialogue data corresponding to the training instruction from the database of dialogues to be selected;
[0008] The training instructions and the historical dialogue data are combined to form training data;
[0009] The private domain large language model is trained using the training data, and the computing power of the general large language model is less than that of the private domain large language model.
[0010] Intelligent question-answering processing for SMEs is achieved through the aforementioned private domain large language model.
[0011] In some embodiments, inputting the dialogue instructions constructed based on private domain information into a general large language model to obtain a database of dialogue corpora to be selected includes:
[0012] The first dialogue instruction is constructed based on user-defined information, private domain information text paragraphs, and dialogue history information; wherein, the user-defined information is used to represent the natural language description of the target object to the user group, the private domain information text paragraphs are used to represent the information fragments carried by the text data of the target object, and the dialogue history information is used to represent the database information of the historical dialogue.
[0013] The first dialogue instruction is input into a general large language model to generate a response, thus obtaining customer consultation dialogue data.
[0014] The database of dialogue data to be selected is updated based on customer consultation dialogue data.
[0015] In some embodiments, the step of inputting the dialogue instructions constructed based on private domain information into a general large language model to obtain a database of dialogue corpora to be selected further includes:
[0016] The second dialogue instruction is constructed based on robot definition information, private domain information text paragraphs, and dialogue history information; wherein, the robot definition information is used to characterize the natural language description of the robot that the target object is about to develop;
[0017] The second dialogue instruction is input into a general large language model to generate a response, thus obtaining robot response dialogue data.
[0018] The database of dialogue data to be selected is updated by updating the dialogue data of the robot's responses.
[0019] In some embodiments, after the step of updating the database of dialogue data to be selected via robot response dialogue data, the method further includes:
[0020] The dialogue data in the dialogue corpus database is input into the binary classification model to determine the current dialogue using a general large language model.
[0021] The current dialogue of the general large language model is determined to have ended, and the dialogue training of the general large language model is terminated.
[0022] In some embodiments, extracting historical dialogue data corresponding to the training instruction from the dialogue corpus database to be selected includes:
[0023] Historical dialogue data corresponding to the training instructions are extracted from the database of dialogues to be selected in a preset number of rounds.
[0024] In some embodiments, training the private domain large language model using the training data includes:
[0025] The training data is then input into the private domain large language model for supervised training.
[0026] In some embodiments, the computing power of the private domain large language model ranges from 6B to 13B, and the computing power of the general large language model is greater than or equal to 175B.
[0027] On the other hand, embodiments of the present invention provide a large language model processing system based on private domain information dialogue, including:
[0028] The first module is used to input dialogue instructions constructed based on private domain information into a general large language model to obtain a database of dialogue corpora to be selected. The dialogue instructions include questions and answers corresponding to the questions. The private domain information includes object information of the scene in which the private domain large language model is located.
[0029] The second module is used to construct training instructions based on private domain information;
[0030] The third module is used to extract historical dialogue data corresponding to the training instruction from the dialogue corpus database to be selected;
[0031] The fourth module is used to combine the training instructions and the historical dialogue data into training data;
[0032] The fifth module is used to train the private domain large language model using the training data, wherein the computing power of the general large language model is less than that of the private domain large language model.
[0033] The sixth module is used for intelligent question-answering processing for small and medium-sized enterprises through the private domain large language model.
[0034] On the other hand, embodiments of the present invention provide a large language model processing system based on private domain information dialogue, including:
[0035] At least one memory for storing programs;
[0036] At least one processor is used to load the program to execute the large language model processing method based on private domain information dialogue.
[0037] On the other hand, embodiments of the present invention provide a computer storage medium storing a computer-executable program, which, when executed by a processor, is used to implement the large language model processing method based on private domain information dialogue.
[0038] The embodiments of the present invention have the following beneficial effects:
[0039] This embodiment obtains a database of dialogue corpora by inputting dialogue instructions constructed based on private domain information into a general large language model. After constructing training instructions based on the private domain information of small businesses, historical dialogue data corresponding to the training instructions is extracted from the database of dialogue corpora. The training instructions and historical dialogue data are combined to form training data, which is used to train the private domain large language model. This allows the private domain large language model of the target object to be trained using the dialogue corpora obtained from the general large language model, thereby obtaining the intelligent dialogue capability of the general large language model. This effectively reduces the training cost and application difficulty of large language models for small and medium-sized enterprises.
[0040] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments, wherein:
[0042] Figure 1 This is a flowchart illustrating a large language model processing method based on private domain information dialogue according to an embodiment of the present invention;
[0043] Figure 2 This is a schematic diagram illustrating the generation of dialogue data according to an embodiment of the present invention;
[0044] Figure 3 This is a schematic diagram illustrating the training of a large language model based on private domain-generated dialogue data, according to an embodiment of the present invention. Detailed Implementation
[0045] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0046] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.
[0047] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.
[0048] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.
[0049] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0050] Reference Figure 1 This invention provides a method for processing large language models based on private domain information dialogue. This method can be applied to processors, servers, or cloud servers of data platforms for small and medium-sized enterprises, as well as processors, servers, or cloud servers of model training platforms. In application, the method of this embodiment includes, but is not limited to, the following steps:
[0051] Step S110: Input the dialogue instructions constructed based on private domain information into the general large language model to obtain the dialogue corpus database to be selected.
[0052] In this embodiment, the dialogue instruction includes a question and a corresponding answer. Private domain information includes object information of the scenario in which the private domain large language model resides. The computational power and data training cost of the large language model in the scenario are relatively low. Object information may include small and medium-sized enterprises, individual businesses, and other objects that can apply the private domain large language model. It is understood that step S110 may involve first constructing a first dialogue instruction based on user-defined information, private domain information text paragraphs, and dialogue history information, then inputting the first dialogue instruction into a general large language model to generate a response, thereby obtaining customer consultation dialogue data, and then updating the database of dialogue data to be selected using the customer consultation dialogue data. In this embodiment, user-defined information is used to represent the natural language description of the target object to the user group, private domain information text paragraphs are used to represent the information fragments carried by the text data of the target object, and dialogue history information is used to represent the database information of historical dialogues.
[0053] In this embodiment, after obtaining the customer consultation dialogue data, step S110 can further construct a second dialogue instruction based on the robot definition information, private domain information text paragraphs, and dialogue history information. Then, the second dialogue instruction is input into a general large language model to generate a response, thereby obtaining robot response dialogue data. Finally, the robot response dialogue data is used to update the candidate dialogue data database. It is understood that the robot definition information is used to characterize the natural language description of the robot that the target object is about to develop.
[0054] After obtaining the dialogue data of the robot's response for each step, this embodiment inputs the dialogue data in the dialogue data database into the binary classification model to determine the current dialogue of the general large language model, and ends the dialogue training of the general large language model when it is determined that the current dialogue of the general large language model has ended.
[0055] For example, with Figure 2 Taking the interactive diagram shown as an example, the process of generating a database of dialogue data to be selected, based on private domain information text segments, includes two stages. Stage 1 is used to construct dialogue data for "customer consultation," and Stage 2 is used to generate dialogue data for "bot response."
[0056] It is understandable that Phase 1 includes, but is not limited to, the following steps:
[0057] Step 1: Construct the first dialogue instruction. Specifically, in this embodiment, "user-defined information," "private domain information text paragraphs," and "dialogue history" are input into the "dialogue generation instruction construction" module. "User-defined information" is a natural language description of the user group by the SME, "private domain information text paragraphs" are information fragments carried by text data within the SME, and "dialogue history" is a database storing historical dialogues generated in this embodiment. Initially, the dialogue history is empty. In the "dialogue generation instruction construction" module, this embodiment uses a specific formatted text prompt template to construct the first dialogue instruction for dialogue generation. The template content of the first dialogue instruction is as follows:
[0058] Now let's assume you are: <User Defined>
[0059] The reference information is: <Private domain information text paragraph>
[0060] Please begin your conversation with the robot based on the reference information:
[0061] Robot: Hello
[0062] User: <Historical Conversation, First Customer Inquiry>
[0063] Robot: <Historical Dialogue, First Robot Response>
[0064] ...
[0065] User: <Historical Conversations, Nth Customer Inquiry>
[0066] Robot: <Historical Dialogue, Nth Robot Reply>
[0067] user:
[0068] Step 2: Generate answers using a general-purpose large language model. Specifically, the constructed first dialogue instruction is fed into a general-purpose large language model with a large number of parameters, such as OpenAI's GPT4 or Baidu's Wenxin Yiyan, to generate answers using the general-purpose large language model.
[0069] Step 3: Generate customer consultation dialogue data. Specifically, a general-purpose language model with a large number of parameters will generate dialogue data for "customer consultation". In this embodiment, the large language model has the ability to understand and reason, and can derive answers even for questions not seen in the training data by understanding the constructed "first dialogue instructions". This ability does not rely solely on whether the model has seen the exact same content during training, but is achieved through its internal "thinking ability".
[0070] Step 4: Update the dialogue corpus database. Specifically, save the "customer consultation" dialogue corpus data generated by the general large language model to the dialogue corpus database for the execution process of Phase 2.
[0071] For Phase 2, the following steps are included but are not limited to:
[0072] Step 1: Construct the second dialogue instruction. In this embodiment, the "robot definition information," "private domain information text paragraph," and "dialogue history information" are input into the "dialogue generation instruction construction" module. The "robot definition information" is a natural language description of the robot that a small or medium-sized enterprise is about to develop. In the "dialogue generation instruction construction" module, this embodiment uses a similar text prompt template to construct the second dialogue instruction. The template for the second dialogue instruction is as follows:
[0073] Now suppose you are: <Definition of a Robot>
[0074] The reference information is: <Private domain information text paragraph>
[0075] Please begin your conversation with the robot based on the reference information:
[0076] Robot: Hello
[0077] User: <Historical Conversation, First Customer Inquiry>
[0078] Robot: <Historical Dialogue, First Robot Response>
[0079] ...
[0080] User: <Historical Conversation, N-1th Customer Inquiry>
[0081] Robot: <Historical Dialogue, N-1st Robot Reply>
[0082] User: <Historical Conversations, Nth Customer Inquiry>
[0083] robot:
[0084] Step 2: Generate responses using a general-purpose large language model. Specifically, the second dialogue instruction is fed into a general-purpose large language model with a large number of parameters to generate dialogue data for the robot's response.
[0085] Step 3: Update the dialogue corpus database. Specifically, the dialogue data obtained from the robot's responses is saved to the dialogue corpus database to be selected, so as to update the data in the dialogue corpus database.
[0086] In this embodiment, for the dialogue end determination in stages 1 and 2, the dialogue data in the candidate dialogue corpus database can be input into a binary classification deep learning mini-model to determine whether the current dialogue has ended. If the current dialogue has not ended, the process returns to stage 1 and continues looping; if the current dialogue has ended, the entire dialogue training data generation process ends.
[0087] Step S120: Construct training instructions based on private domain information.
[0088] In this embodiment, the template for the training instructions can be referred to Figure 2 The template formats used in Phase 1 and Phase 2 will not be elaborated here.
[0089] Step S130: Extract historical dialogue data corresponding to the training instructions from the dialogue corpus database to be selected.
[0090] In this embodiment, historical dialogue data corresponding to the training instructions in a preset number of rounds can be extracted from a database of dialogue corpora to be selected. Specifically, from... Figure 2 The database of dialogue data to be selected is used to extract customer consultation dialogue data and robot response dialogue data from a specific round.
[0091] Step S140: Combine training instructions and historical dialogue data to form training data;
[0092] In the embodiments of this application, such as Figure 3 As shown, "robot definition information", "private domain information text paragraphs", "customer consultation information" and their associated "historical dialogues" can be input into the "model training instruction construction" module to obtain training data.
[0093] Step S150: Train the private domain large language model using the trained data.
[0094] In this embodiment, the computing power of the general-purpose large language model is less than that of the private-domain large language model. For example, the computing power of the private-domain large language model can be set to 6B-13B, while the computing power of the general-purpose large language model is greater than or equal to 175B. After determining the specific computing power of the general-purpose and private-domain large language models, training data is input into the private-domain large language model for supervised training. During training, the target output label of the private-domain large language model is the corresponding "bot response" dialogue data. In this embodiment, the parameter scale of the private-domain large language model is relatively small, which can be adapted to the computing power range (6B-13B parameter scale) that small and medium-sized enterprises can afford. The method of this embodiment can be likened to a knowledge distillation process: the high-performance, high-parameter general-purpose language model acts as the teacher, using the generated private-domain "dialogue history" data to teach the smaller-parameter private-domain large language model, allowing the private-domain large language model to also benefit from the private-domain data. This process is knowledge distillation based on data samples.
[0095] Step S160: Perform intelligent question-answering processing for SMEs using a private domain large language model.
[0096] In this embodiment of the application, after training the private domain large language model, the private domain large language model can be applied to the actual equipment or platform of small and medium-sized enterprises, so that the intelligent question answering function of the private domain large language model can be used for intelligent question answering processing.
[0097] In summary, the method of this embodiment has the following beneficial effects:
[0098] First, it reduces model training and deployment costs. This embodiment uses knowledge distillation to transfer knowledge from a general-purpose language model with a large number of parameters to a private-domain large-scale language model with a smaller parameter size. This allows the private-domain large-scale language model to benefit from the large-parameter model and achieve similar performance without incurring significant training and deployment costs. Small and medium-sized enterprises no longer need to invest heavily in training and deploying giant models, as the private-domain large-scale language model can achieve satisfactory performance with a relatively small parameter size.
[0099] Second, low-cost private domain training data generation. Traditionally, collecting private domain dialogue data suitable for training large language models is an expensive and difficult task. This embodiment utilizes existing private domain information text segments within the enterprise to construct dialogue instructions and generate dialogue history, thereby generating reliable dialogue training data and overcoming the difficulty of acquiring private domain data.
[0100] Third, accurate and reliable dialogue data generation. This embodiment ensures the consistency and authenticity of the generated dialogue data in terms of context by generating dialogue history in an orderly manner. This helps improve the model training effect, enabling the generated model to answer questions more accurately and generate appropriate responses in practical applications.
[0101] Fourth, performance improvement in the intelligent question-answering dialogue domain. This embodiment, through knowledge distillation, allows the private domain large language model to fully benefit from the extensive learning and reasoning capabilities of the ultra-large parameter general language model. The resulting private domain large language model exhibits higher accuracy and intelligence in the intelligent question-answering dialogue domain. It can better understand questions and generate more appropriate answers, thereby improving dialogue quality and meeting the expectations of enterprises and users for high-quality dialogue communication.
[0102] This invention provides a large language model processing system based on private domain information dialogue, comprising:
[0103] The first module is used to input dialogue instructions constructed based on private domain information into a general large language model to obtain a database of dialogue corpora to be selected. The dialogue instructions include questions and answers corresponding to the questions. The private domain information includes object information of the scene in which the private domain large language model is located.
[0104] The second module is used to construct training instructions based on private domain information;
[0105] The third module is used to extract historical dialogue data corresponding to the training instruction from the dialogue corpus database to be selected;
[0106] The fourth module is used to combine the training instructions and the historical dialogue data into training data;
[0107] The fifth module is used to train the private domain large language model using the training data, wherein the computing power of the general large language model is less than that of the private domain large language model.
[0108] The sixth module is used for intelligent question-answering processing for small and medium-sized enterprises through the private domain large language model.
[0109] The content of the method embodiments of the present invention is applicable to the system embodiments. The specific functions implemented in the system embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.
[0110] This invention provides a large language model processing system based on private domain information dialogue, comprising:
[0111] At least one memory for storing programs;
[0112] At least one processor is used to load the program for execution. Figure 1 The method shown is a large language model processing method based on private domain information dialogue.
[0113] The content of the method embodiments of the present invention is applicable to the system embodiments. The specific functions implemented in the system embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.
[0114] This invention provides a computer storage medium storing a computer-executable program, which, when executed by a processor, is used to implement... Figure 1 The method shown is a large language model processing method based on private domain information dialogue.
[0115] The content of the method embodiments of the present invention is applicable to the storage medium embodiments. The specific functions implemented by the storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.
[0116] This invention also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The method shown is a large language model processing method based on private domain information dialogue.
[0117] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention. Furthermore, the embodiments of the present invention and the features thereof can be combined with each other unless otherwise specified.
Claims
1. A method for processing large language models based on private domain information dialogue, characterized in that, Includes the following steps: A first dialogue instruction is constructed based on user-defined information, private domain information text segments, and dialogue history information; wherein, the user-defined information is used to represent the natural language description of the user group by the SME, the private domain information text segments are used to represent information fragments carried by text data within the SME, and the dialogue history information is used to represent database information of historical dialogues; the first dialogue instruction is input into a general large language model to generate a response, obtaining customer consultation dialogue data; the database of dialogue data to be selected is updated based on the customer consultation dialogue data; a second dialogue instruction is constructed based on robot definition information, private domain information text segments, and dialogue history information; wherein, the robot Define information to characterize the natural language description of the robot to be developed for the target object; input the second dialogue instruction into the general large language model to generate a response, and obtain robot response dialogue data; update the candidate dialogue data database using the robot response dialogue data; the dialogue instruction includes a question and the answer corresponding to the question; the private domain information includes object information of the scene in which the private domain large language model is located; after updating the candidate dialogue data database using the robot response dialogue data, input the dialogue data into a binary classification model to determine the current dialogue of the general large language model; when it is determined that the current dialogue has ended, the dialogue training of the general large language model ends. Constructing training instructions based on private domain information; Extract historical dialogue data corresponding to the training instruction from the database of dialogues to be selected; The training instructions and the historical dialogue data are combined to form training data; The private domain large language model is trained using the training data, and the parameter size of the general large language model is larger than that of the private domain large language model. Intelligent question-answering processing for SMEs is achieved through the aforementioned private domain large language model.
2. The method for processing a large language model based on private domain information dialogue according to claim 1, characterized in that, The step of extracting historical dialogue data corresponding to the training instruction from the candidate dialogue corpus database includes: Historical dialogue data corresponding to the training instructions are extracted from the database of dialogues to be selected in a preset number of rounds.
3. The method for processing a large language model based on private domain information dialogue according to claim 1, characterized in that, The step of training the private domain large language model using the training data includes: The training data is then input into the private domain large language model for supervised training.
4. A method for processing large language models based on private domain information dialogue according to any one of claims 1-3, characterized in that, The parameter size of the private domain large language model ranges from 6B to 13B, while the parameter size of the general large language model is greater than or equal to 175B.
5. A large language model processing system based on private domain information dialogue, characterized in that, include: The first module is used to construct a first dialogue instruction based on user-defined information, private domain information text segments, and dialogue history information; wherein, the user-defined information is used to represent the natural language description of the user group by the small and medium-sized enterprise, the private domain information text segments are used to represent information fragments carried by text data within the small and medium-sized enterprise, and the dialogue history information is used to represent database information of historical dialogues; the first dialogue instruction is input into a general large language model to generate a response, obtaining customer consultation dialogue data; the database of dialogue data to be selected is updated based on the customer consultation dialogue data; and a second dialogue instruction is constructed based on robot-defined information, private domain information text segments, and dialogue history information; wherein, the... The robot definition information is used to characterize the natural language description of the robot to be developed for the target object; the second dialogue instruction is input into the general large language model to generate a response, and the robot response dialogue data is obtained; the candidate dialogue data database is updated by the robot response dialogue data; the dialogue instruction includes a question and the answer corresponding to the question; the private domain information includes the object information of the scene in which the private domain large language model is located; after updating the candidate dialogue data database by the robot response dialogue data, the dialogue data is input into a binary classification model to determine the current dialogue of the general large language model; when it is determined that the current dialogue has ended, the dialogue training of the general large language model is terminated. The second module is used to construct training instructions based on private domain information; The third module is used to extract historical dialogue data corresponding to the training instruction from the dialogue corpus database to be selected; The fourth module is used to combine the training instructions and the historical dialogue data into training data; The fifth module is used to train the private domain large language model using the training data, wherein the parameter size of the general large language model is larger than the parameter size of the private domain large language model. The sixth module is used for intelligent question-answering processing for small and medium-sized enterprises through the private domain large language model.
6. A large language model processing system based on private domain information dialogue, characterized in that, include: At least one memory for storing programs; At least one processor is configured to load the program to execute the large language model processing method based on private domain information dialogue as described in any one of claims 1-4.
7. A computer storage medium, characterized in that, It stores a computer-executable program, which, when executed by a processor, is used to implement the large language model processing method based on private domain information dialogue as described in any one of claims 1-4.