Chip Selection Method, System and Circuit Design Method Based on Large Language Model

Abstract

The present invention provides a method for selecting a chip based on a large language model, a system, and a circuit design method. The method includes: SA1, obtaining the interaction information to be processed; SA2, extracting information from the interaction information to be processed to obtain initial requirement information; SA3, analyzing whether the initial requirement information includes abnormal requirements; SA4, if so, feedbacking the abnormal requirements and the reasons for the occurrence of the abnormal requirements to the user, and at the same time interacting with the user to generate new interaction information, so as to execute SA2 according to the newly generated interaction information; SA5, if not, performing requirement analysis and standardization processing according to the initial requirement information, so as to obtain chip product requirement information with data item names including at least product classification based on the initial requirement information and outputting it. The present invention optimizes the interaction function and improves the accuracy of the conclusion output of the large language model.

Description

Technical Field

[0001] The present invention relates to the field of computers, and particularly to a method for selecting a chip based on a large language model, a system, and a circuit design method. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models have begun to demonstrate their powerful potential in multiple fields. In the field of chip design, large language models can assist in multiple links of the automated chip design process through natural language processing and generation technologies, thereby improving design efficiency and quality.

[0003] In the automated chip design process, the first step is to use a large language model to interact and collect the chip product requirements of users. However, due to the lack of corresponding design, on the one hand, it is difficult for technical personnel to accurately express requirements in a standardized manner, so that no matter the length of the description, there are problems such as weak directivity or pertinence, and thus the accuracy of the conclusion output of the large language model is less than satisfactory; on the other hand, the chip product requirements provided by the front end usually have problems such as uncertainty, fuzzy presentation, and lack of key weight information, resulting in problems of standardization of information transmission and lack of key information when further processing the information, which also affects the accuracy of the conclusion output of the large language model. Summary of the Invention

[0004] Aiming at the deficiencies in the prior art, the present invention provides a method for selecting a chip based on a large language model, a chip selection system, and a circuit design method, which solve the problem that the existing large language model-driven automated chip design has defects in the interaction function and reduces the accuracy of the conclusion output of the large language model.

[0005] According to an embodiment of the present invention, in a first aspect, a method for selecting a chip based on a large language model is provided, including:

[0006] SA1. Obtain the interaction information to be processed, where the interaction information to be processed is all the interaction information generated during chip selection in a human-machine collaboration mode;

[0007] SA2. Extract information from the interaction information to be processed to obtain initial requirement information;

[0008] SA3. Analyze whether the initial requirement information includes abnormal requirements; the abnormal requirements indicate that the initial requirement information has at least one of unreasonable requirement expressions, fuzzy expressions, and irrelevant expressions;

[0009] SA4. If so, feedback the abnormal requirements and the reasons for the appearance of the abnormal requirements to the user, and at the same time interact with the user to generate new interaction information, so as to execute SA2 according to the newly generated interaction information;

[0010] SA5. If not, perform requirement analysis and standardization processing based on the initial requirement information to obtain chip product requirement information with data item names including at least product classification based on the initial requirement information and output it;

[0011] Among them, the first hybrid dataset including a proper noun dataset, a product classification dataset, a multi-round dialogue dataset, and a general dataset is used to supervise and fine-tune the training of the first basic intelligent agent to obtain an interactive intelligent agent to implement the SA1 to SA5.

[0012] Optionally, the proper noun dataset includes K noun data contents based on K proper nouns;

[0013] The k-th noun data content includes a proper noun QA pair composed of the correct explanation of the k-th proper noun and a question sentence based on the correct explanation of the k-th proper noun, and a system prompt word based on the proper noun QA pair;

[0014] Among them, k is a positive integer less than or equal to K, and K is a positive integer.

[0015] Optionally, the product classification dataset includes N category data contents based on M first-level chip categories;

[0016] If the n-th category data content belongs to the m-th first-level chip category, the n-th category data content includes a product classification QA pair composed of a second-level chip category subordinate to the m-th first-level chip category and a question sentence based on the second-level chip category subordinate to the m-th first-level chip category, and a system prompt word based on the product classification QA pair;

[0017] Among them, M and N are positive integers, n is a positive integer less than or equal to N, and m is a positive integer less than or equal to M.

[0018] Optionally, the multi-round dialogue dataset includes U multi-round dialogue data contents based on L multi-round dialogues;

[0019] If the u-th multi-round dialogue data content belongs to the l-th multi-round dialogue, the u-th multi-round dialogue data content includes a multi-round dialogue QA pair composed of a simulated question in the l-th multi-round dialogue and a reply content based on the simulated question in the l-th multi-round dialogue, a system prompt word based on the multi-round dialogue QA pair, and a historical dialogue set;

[0020] Among them, L and U are positive integers, l is a positive integer less than or equal to L, u is a positive integer less than or equal to U. In the l-th multi-round dialogue, the simulated question is generated by simulating the user's question about a randomly selected chip product manual; the reply content based on the simulated question in the l-th multi-round dialogue is generated by the randomly selected chip product manual.

[0021] Optionally, in the l-th multi-round conversation, the simulated questions are also obtained by randomly selecting from a preset question database. For the simulated questions obtained by randomly selecting from the preset question database, the reply content includes the content of the randomly selected chip product manual.

[0022] Wherein, the preset question database includes a plurality of preset questions, and each preset question includes at least one of unreasonable requirement expressions, vague expressions, and irrelevant expressions.

[0023] A second aspect provides a chip selection method based on a large language model, including:

[0024] SB1. Obtain chip product requirement information provided by an interaction agent, where the interaction agent executes the above-mentioned chip selection method based on a large language model;

[0025] SB2. The selection agent serves as a generation module, and based on the chip product requirement information, combines a pre-set retrieval module, a re-ranking module, and a local knowledge base to perform chip selection and output chip recommendation information;

[0026] The SB2 includes:

[0027] Compare the chip product requirement information with the chip product data stored in the local knowledge base through the retrieval module to obtain multiple relevant documents and send them to the re-ranking module;

[0028] Perform relevance ranking on the multiple relevant documents through the re-ranking module and send the ranking result to the selection agent; in the ranking result, the relevant documents with higher relevance have smaller ranking serial numbers, and the relevant documents with ranking serial numbers smaller than the preset serial number value are strongly relevant documents;

[0029] Analyze all strongly relevant documents according to the chip product requirement information and output chip recommendation information.

[0030] Optionally, construct a retrieval triple dataset to perform supervised contrastive learning training on the basic embedding model. After the training is completed, obtain the retrieval module;

[0031] Wherein, the constructed retrieval triple dataset includes a retrieval query element, a retrieval positive sample element, and a retrieval negative sample element; the retrieval query element is extracted from the total demand information data; the retrieval positive sample element is the only chip product data selected from the local knowledge base to reply to the retrieval query element; the retrieval negative sample element is multiple chip product data randomly selected from the local knowledge base.

[0032] Optionally, a rearranged triple dataset is constructed to perform supervised contrastive learning training on the basic rearrangement model. After the training is completed, the rearrangement module is obtained;

[0033] Among them, the constructed rearranged triple dataset includes rearranged query elements, rearranged positive sample elements, and rearranged negative sample elements; the rearranged query elements are simulated requirement descriptions, and the data size of the simulated requirement descriptions is less than a preset number of bytes; the rearranged positive sample elements are the only correct answers written based on the simulated requirement descriptions; the rearranged negative sample elements are multiple wrong answers written based on the simulated requirement descriptions, and the similarity between the wrong answers and the only correct answer is greater than a preset similarity.

[0034] Optionally, any chip product data stored in the local knowledge base includes document descriptions of product features, document descriptions of product application examples, document descriptions of key parameters, and document descriptions of complete information.

[0035] Optionally, the second basic intelligent agent is trained in two stages through continued pre-training and supervised fine-tuning training to obtain a selection intelligent agent to implement the SB1 to SB2;

[0036] In the continued pre-training, multi-dimensional chip text data is obtained, and the multi-dimensional chip text data is pre-processed to obtain multiple corpus chunks, and the second basic intelligent agent is trained through the multiple corpus chunks;

[0037] In the supervised fine-tuning training, the second basic intelligent agent after continued pre-training is trained through a second mixed dataset including a proper noun dataset, a product classification dataset, a chip selection dataset, and a general dataset to obtain a selection intelligent agent.

[0038] Optionally, the chip selection dataset includes P chip matching data;

[0039] The p-th chip matching data includes a chip matching QA pair composed of a simulated chip requirement description and recommended information that meets the simulated chip requirement description, and a system prompt word based on the chip matching QA pair;

[0040] Among them, the simulated chip requirement description is generated by randomly extracting a chip product manual;

[0041] The recommended information that meets the simulated chip requirement description is the chip product model and chip product introduction recorded in the randomly extracted chip product manual.

[0042] A third aspect provides a chip selection system based on a large language model, including an interaction agent driven by the large language model, a selection agent, and an evaluation agent; the interaction agent executes the chip selection method based on the large language model as described above; the selection agent executes the chip selection method based on the large language model as described above; the evaluation agent executes the following steps:

[0043] SC1. Obtain chip recommendation information;

[0044] Wherein, the chip recommendation information is the result of the selection agent as a generation module, combining a pre-set retrieval module, a rearrangement module, and a local knowledge base, and performing chip selection based on chip product requirement information;

[0045] SC2. Evaluate the confidence level of the chip recommendation information;

[0046] SC3. If the confidence level obtained according to SC2 is the highest level, send the chip recommendation information to the user. Optionally, in SC2, when evaluating the confidence level of the chip recommendation information, evaluation content is also generated.

[0047] Optionally, it further includes:

[0048] SC4. If the confidence level obtained according to SC2 is not the highest level, send the evaluation content corresponding to the confidence level obtained according to SC2 to the selection agent, so that the selection agent changes the chip recommendation information.

[0049] Optionally, the evaluation agent is designed through prompt engineering, and SC1 to SC3 are implemented through the evaluation agent.

[0050] A fourth aspect provides a circuit design method, including:

[0051] SD1. Obtain a circuit design request, where the circuit design request includes interaction information describing at least one circuit module; wherein, the interaction information of different circuit modules is different;

[0052] SD2. Decompose the circuit design request into multiple circuit module design tasks and the execution order of the multiple circuit module design tasks;

[0053] SD3. When executing the target circuit module design task, call the chip selection system as described above; the selection system is used to perform requirement analysis according to the circuit design request to obtain chip product requirement information, perform chip selection according to the chip product requirement information to obtain chip recommendation information, and evaluate the confidence level of the chip recommendation information;

[0054] SD4. When the confidence level of the chip recommendation information is the highest level, obtain the target chip based on the chip recommendation information, and according to the requirements of the target circuit module design task, use the target chip to build a simulation circuit or a test circuit for verification;

[0055] SD5. When the verification is passed, the target chip is the chip used to execute the target circuit module design task;

[0056] SD6. Execute each circuit module design task according to SD3 to SD5 to complete the circuit design.

[0057] The fifth aspect provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned chip selection method based on a large language model, or the above-mentioned chip selection method based on a large language model, or the above-mentioned circuit design method.

[0058] The sixth aspect provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the above-mentioned chip selection method based on a large language model, or the above-mentioned chip selection method based on a large language model, or the above-mentioned circuit design method.

[0059] The seventh aspect provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the above-mentioned chip selection method based on a large language model, or the above-mentioned chip selection method based on a large language model, or the above-mentioned circuit design method.

[0060] The chip selection method based on a large language model provided by the present invention interacts with the user continuously in a human-machine collaboration mode, guides the user to select a chip, extracts information from the to-be-processed interaction information, so as to collect the user's requirements for chip products, that is, the initial requirement information. It also analyzes the initial requirement information to avoid abnormal requirements such as unreasonable requirement expressions, fuzzy expressions, and irrelevant expressions from affecting subsequent information processing, thereby ensuring the availability of the output chip product requirement information. Moreover, it also performs standardization processing on the initial requirement information to improve the directivity of the output chip product requirement information, and finally improves the accuracy of the conclusion output of the large language model. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic flowchart of the implementation of the chip selection method based on a large language model according to an embodiment of the present invention;

[0062] Figure 2 It is a schematic flowchart of the implementation of another chip selection method based on a large language model according to an embodiment of the present invention;

[0063] Figure 3 Schematic diagram of the composition structure of the chip selection system based on large language model according to an embodiment of the present invention;

[0064] Figure 4 Execution steps to be achieved by the evaluation agent according to an embodiment of the present invention;

[0065] Figure 5 Schematic diagram of the implementation process of the circuit system design method according to an embodiment of the present invention;

[0066] Figure 6 Schematic diagram of the structure of the electronic device according to an embodiment of the present invention. Specific implementation manners

[0067] The technical solutions in the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0068] As Figure 1 shown, an embodiment of the present invention proposes a chip selection method based on a large language model, enabling the currently controlled agent, i.e., the interaction agent, to continuously interact with the user, obtain the user's requirements for chip products, and extract information. At the same time, a function for identifying abnormal requirements is provided to ensure the availability of the output chip product requirement information.

[0069] The chip selection method based on the large language model includes but is not limited to the following steps:

[0070] SA1. Obtain the interaction information to be processed, where the interaction information to be processed is all the interaction information generated during chip selection in the human-machine collaboration mode;

[0071] SA2. Extract information from the interaction information to be processed to obtain initial requirement information;

[0072] SA3. Analyze whether the initial requirement information includes abnormal requirements; the abnormal requirements indicate that the initial requirement information has at least one of unreasonable requirement expressions, ambiguous expressions, and irrelevant expressions;

[0073] SA4. If so, feedback the abnormal requirements and the reasons for the occurrence of the abnormal requirements to the user, and at the same time interact with the user to generate new interaction information, so as to execute SA2 according to the newly generated interaction information;

[0074] SA5. If not, perform requirement analysis and standardization processing according to the initial requirement information, so as to obtain chip product requirement information with data item names at least including product classification based on the initial requirement information and output it.

[0075] In the above step SA1, in the human-machine collaboration mode, the interaction agent of the embodiment of the present invention guides the user to select a chip. The detailed process may be as follows: The user interacts with the interaction agent in multiple rounds to provide information about chip selection in the form of voice, text, etc., such as chip signals, chip parameters, chip functions, etc. Based on this, the interaction information to be processed includes but is not limited to voice information, text information, etc. provided by the user.

[0076] For the above steps SA2, SA3, and SA5, the embodiment of the present invention provides a specific data set to train the interaction agent, enabling it to have the functions of information extraction from the interaction information to be processed, identifying abnormal requirements, performing requirement analysis on the initial requirement information that does not include abnormal requirements, and data standardization processing.

[0077] For the above step SA3, the abnormal requirements analyzed in step SA3 indicate situations such as unreasonable requirement expressions, ambiguous expressions, and irrelevant expressions in the initial requirement information. In specific applications, the abnormal requirements affect the chip product requirement information output in step SA5, ultimately resulting in the failure of the selection agent to select a chip or inaccurate chip selection results.

[0078] For the above step SA5, the chip product requirement information output in step SA5 is finally output to the selection agent. The chip product requirement information is standardized data, and the data item names include at least product classification. Exemplarily, the data item names include primary product category, secondary product category, product application, and product key parameters.

[0079] In the embodiment of the present invention, before the above step SA1, the basic agent is trained to obtain an interaction agent capable of implementing the functions shown in steps SA1 to SA5, which is: performing supervised fine-tuning training on the first basic agent through a first mixed data set including a proper noun data set, a product classification data set, a multi-round dialogue data set, and a general data set.

[0080] It should be noted that the proper noun data set and the product classification data set here are used to enable the interaction agent to better understand and solve problems in the field of electronic information, especially to have the ability to select appropriate chip products for each circuit module in board-level circuit design. The multi-round dialogue data set is used to enable the final interaction agent to continuously interact with the user without deviating from the field of electronic information.

[0081] For the above-mentioned proper noun dataset, in an embodiment of the present invention, the proper noun dataset includes K noun data contents based on K proper nouns, where the K proper nouns all belong to the field of electronic information. Exemplarily, the k-th noun data content includes a proper noun QA pair composed of the correct explanation of the k-th proper noun and a question sentence based on the correct explanation of the k-th proper noun, and a system prompt word based on the proper noun QA pair.

[0082] Among them, the system prompt word based on the proper noun QA pair is used to instruct the interactive agent to extract information from the to-be-processed interactive information according to the corresponding rule between the correct explanation of the proper noun exemplified by the proper noun dataset and the question sentence of the correct explanation of the proper noun.

[0083] In a specific application, use to represent the correct explanation of the k-th proper noun, and use to represent the correct explanation of any proper noun, that is the question sentence of, then and constitute a "question-answer" pair, that is, a proper noun QA pair, and then use to represent the system prompt word of the proper noun QA pair. Based on this, the proper noun dataset of the embodiment of the present invention can be expressed as:

[0084]

[0085] Among them, is the number of data in the proper noun dataset = K.

[0086] For the above-mentioned product classification dataset, in an embodiment of the present invention, the product classification dataset includes N category data contents based on M first-level chip categories; Exemplarily, if the n-th category data content belongs to the m-th first-level chip category, the n-th category data content includes a second-level chip category subordinate to the m-th first-level chip category and a product classification QA pair composed of a question sentence based on the second-level chip category subordinate to the m-th first-level chip category, and a system prompt word based on the product classification QA pair. Among them, the system prompt word based on the product classification QA pair is used to instruct the interactive agent to classify the first-level chip category and the second-level chip category based on a certain first-level chip category according to the corresponding rule between the second-level chip category of the m-th first-level chip category exemplified by the product classification dataset and the question sentence based on the second-level chip category of the m-th first-level chip category, based on the initial requirement information.

[0087] It should be noted that in the above product classification dataset, the N category data contents are divided into M first-level chip categories for storage. In one embodiment, 16 first-level chip categories are provided, namely amplifier, audio, battery management IC, clock and timing, data converter, DLP product, interface, isolator, logic and voltage conversion, microcontroller and processor, motor driver, power management, radio frequency and microwave, sensor, switch and multiplexer, wireless connector. Then, second-level chip categories are defined for the M first-level chip categories. Among them, the relationship between the second-level chip categories and the first-level chip categories is a category subordination relationship. Each first-level chip category includes multiple category data contents based on its second-level chip categories. For example, when the sensor is a first-level chip category, it includes photoelectric sensors, infrared sensors, etc. The photoelectric sensors and infrared sensors are the category data contents of the second-level chip categories. For a category data content in the product classification dataset, assuming that this category data content represents a photoelectric sensor, then this category data content includes the photoelectric sensor (second-level chip category) subordinate to the sensor (first-level chip category), as well as the product classification QA pairs composed of the question statements based on the sensor (first-level chip category) and the photoelectric sensor (second-level chip category), and also includes the system prompt words based on the product classification QA pairs.

[0088] In specific applications, use to represent the relationship between the first-level chip category and the second-level chip category, that is, the second-level chip category belonging to the m-th first-level chip category. For each a question statement is designed , then and constitute a "question-answer" pair, that is, the product classification QA pair. Then use to represent the system prompt word of the product classification QA pair. Based on this, the product classification dataset of the embodiment of the present invention can be expressed as:

[0089]

[0090] Among them, is the number of data in the product classification dataset , =N.

[0091] For the above multi-turn dialogue dataset, in the embodiment of the present invention, the multi-turn dialogue dataset includes U multi-turn dialogue data contents based on L multi-turn dialogues; exemplarily, if the u-th multi-turn dialogue data content belongs to the l-th multi-turn dialogue, the u-th multi-turn dialogue data content includes the multi-turn dialogue QA pairs composed of the simulated questions in the l-th multi-turn dialogue and the reply contents based on the simulated questions in the l-th multi-turn dialogue, the system prompt words based on the multi-turn dialogue QA pairs, and the historical dialogue set.

[0092] The above-mentioned simulated user simulates the questioning thinking of a natural person regarding the chip product manual. Exemplarily, in the l-th multi-round conversation, the simulated questions are generated by simulating the questions of the randomly selected chip product manual by the simulated user; based on the reply content of the simulated questions in the l-th multi-round conversation, the randomly selected chip product manual is used to generate. In addition, in the embodiments of the present invention, the interactive agent also realizes the analysis of abnormal requirements. Therefore, in the l-th multi-round conversation, the simulated questions are also randomly selected from a preset question database. For the simulated questions randomly selected from the preset question database, the reply content includes the content of the randomly selected chip product manual; wherein, the preset question database includes multiple preset questions, and each preset question includes at least one of unreasonable requirement expressions, ambiguous expressions, and irrelevant expressions.

[0093] In a specific application, use Q i | 多轮对话 to represent the simulated questions in the l-th multi-round conversation, and use A i | 多轮对话 to represent the reply content based on Q i | 多轮对话 . Then Q i | 多轮对话 and A i | 多轮对话 constitute a "question-answer" pair, that is, a multi-round conversation QA pair. Then use S| 多轮对话 to represent the system prompt word of the multi-round conversation QA pair. In addition, the historical conversation set H i | 多轮对话 is also used to assist in training. Based on this, the multi-round conversation dataset of the embodiments of the present invention is expressed as:

[0094]

[0095] Where is the number of data in the multi-round conversation dataset , =U.

[0096] It should be noted that in the embodiments of the present invention, the multi-round conversation dataset actually includes a dataset constructed based on the questioning thinking of a natural person regarding the chip product manual, and also includes a dataset constructed based on the questioning thinking of a natural person presenting abnormal requirements, that is, the preset question database. Then, based on the system prompt word of the multi-round conversation QA pair, it is used to instruct the interactive agent to analyze the abnormal requirements of the initial requirement information according to the corresponding rules between the simulated questions and the reply content based on the simulated questions exemplified by the multi-round conversation dataset, and at the same time guide the situation where the initial requirement information includes abnormal requirements, that is, interact with the user to re-generate interaction information, so as to supplement or correct the previous abnormal requirements according to the newly generated interaction information.

[0097] As Figure 2 shown, an embodiment of the present invention also proposes a chip selection method based on a large language model. After the SA5 in the above embodiment outputs chip product requirement information to the selection agent, the currently controlled agent, that is, the selection agent, performs chip selection according to the chip product requirement information and outputs chip recommendation information. At the same time, a retrieval enhancement technology is provided to improve the chip selection ability of the selection agent, which is different from the existing board-level circuit design, where automatic recommendation is made based on data such as market value and user selection rate. If there are multiple similar chip products, the user is prompted to make an independent choice. Figure 2 The chip selection method based on a large language model shown above includes but is not limited to the following steps:

[0098] SB1. Obtain chip product requirement information provided by the interaction agent, where the interaction agent executes the chip selection method of the above large language model;

[0099] SB2. The selection agent, as a generation module, combines a pre-set retrieval module, a re-rank module, and a local knowledge base, performs chip selection based on the chip product requirement information, and outputs chip recommendation information;

[0100] The SB2 includes:

[0101] Compare the chip product requirement information with the chip product data stored in the local knowledge base through the retrieval module to obtain multiple relevant documents and send them to the re-rank module;

[0102] Perform relevance ranking on multiple relevant documents through the re-rank module and send the ranking result to the selection agent; in the ranking result, the smaller the ranking serial number of the relevant document with higher relevance, and the relevant document with a ranking serial number less than the preset serial number value is a strongly relevant document;

[0103] Analyze all strongly relevant documents according to the chip product requirement information and output chip recommendation information.

[0104] The chip product requirement information in the above step SB1 is provided by the interaction agent. Specifically, the interaction agent will perform chip selection with the user according to the chip selection method based on a large language model as Figure 1 shown, and then generate to-be-processed interaction information during this continuous interaction process of chip selection. Finally, after steps such as information extraction and requirement analysis based on the to-be-processed interaction information, the above chip product requirement information is obtained.

[0105] The execution subjects of the above steps SB1 and SB2 are both the selection intelligent agent. According to the above step SB2, the embodiment of the present invention provides a chip selection method based on a large language model. It not only selects chips through the selection intelligent agent, but also designs a retrieval enhancement technology with the selection intelligent agent as the generation module, realizing the integration of the large language model and retrieval enhanced generation. In this way, the selection ability of chip selection is improved, and the optimal solution can be output even when there are multiple similar chip products.

[0106] The embodiments of the present invention will respectively elaborate on the retrieval module, the re-rank module, the local knowledge base, and the selection intelligent agent.

[0107] First, for the retrieval module, in one embodiment, a retrieval triple dataset is constructed to perform supervised contrastive learning training on the basic embedding model. After the training is completed, the retrieval module is obtained. In specific applications, the triple dataset contains three elements: query Q, positive sample P, and negative sample N. Where Q is the input question, P is the correct answer or the standard answer, and N is the negative and incorrect answer. In a better implementation, the retrieval triple dataset constructed by the embodiment of the present invention includes a retrieval query element, a retrieval positive sample element, and a retrieval negative sample element; the retrieval query element is extracted from the total demand information data; the retrieval positive sample element is the unique chip product data selected from the local knowledge base to reply to the retrieval query element; the retrieval negative sample element is multiple chip product data randomly extracted from the local knowledge base. It should be noted that the total demand information data refers to all the demand information output by the interaction intelligent agent within the historical time period, and here it is used as the data source of the retrieval query element in the retrieval unit dataset.

[0108] In specific applications, use to represent the retrieval query element, use to represent the retrieval positive sample element, use to represent the retrieval negative sample element. Based on this, the retrieval triple dataset of the embodiment of the present invention can be expressed as:

[0109]

[0110] Where is the number of data in this triple dataset in.

[0111] Secondly, for the rearrangement module, in one embodiment, a rearrangement triple dataset is constructed to perform supervised contrastive learning training on the basic rearrangement model. After the training is completed, the rearrangement module is obtained. In a specific application, the triple dataset contains three elements: query Q, positive sample P, and negative sample N. Among them, Q is the input question, P is the correct answer or the standard answer, and N is the negative and incorrect answer. In a better implementation, the rearrangement triple dataset constructed in the embodiment of the present invention is more concise and refined compared with the above-mentioned retrieval triple dataset, including rearrangement query elements, rearrangement positive sample elements, and rearrangement negative sample elements; the rearrangement query element is a simulated requirement description, and the data size of the simulated requirement description is less than a preset number of bytes; the rearrangement positive sample element is the only correct answer written based on the simulated requirement description; the rearrangement negative sample element is multiple incorrect answers written based on the simulated requirement description, and the similarity between the incorrect answer and the only correct answer is greater than a preset value. It should be noted that setting the constraint condition that the data size of the simulated requirement description is less than the preset number of bytes is used to indicate that the rearrangement triple dataset is more concise and refined than the retrieval triple dataset used to train the embedding model before; setting the constraint condition that the similarity between the incorrect answer and the only correct answer is greater than the preset value is used to enhance the performance of the rearrangement model, so that it can more accurately distinguish relevant documents with strong relevance and relevant documents with weak relevance, which is beneficial for the selection agent to find the optimal solution in the case of multiple similar chip products.

[0112] In a specific application, use to represent the rearrangement query element, use to represent the rearrangement positive sample element, and use to represent the rearrangement negative sample element. Based on this, the rearrangement triple dataset of the embodiment of the present invention can be expressed as:

[0113]

[0114] where is the number of data in this triple dataset in.

[0115] Then, for the local knowledge base, in one embodiment, any chip product data stored in the local knowledge base includes document descriptions of product features, document descriptions of product application instances, document descriptions of key parameters, and document descriptions of complete information. In a specific application, chip product data can be obtained by extracting information from multiple chip product manuals and then stored in the local knowledge base. It is also possible to regularly obtain new chip product manuals to extract new chip product data based on the new chip product manuals and then update them to the local knowledge base.

[0116] Finally, for the selection agent, in one embodiment, to train the second basic agent to obtain a chip selection method based on a large language model as shown in Figure 2 , especially the selection agent in step SB2 thereof. The detailed implementation steps include:

[0117] The second basic agent is trained in two stages through continued pre-training and supervised fine-tuning training to obtain a selection agent to analyze all strongly relevant documents according to the chip product requirement information and output chip recommendation information;

[0118] In the continued pre-training, multi-dimensional chip text data is obtained, and the multi-dimensional chip text data is pre-processed to obtain a plurality of corpus blocks, and the second basic agent is trained through the plurality of corpus blocks;

[0119] In the supervised fine-tuning training, the second basic agent after continued pre-training is trained through a second mixed data set including a proper noun data set, a product classification data set, a chip selection data set, and a general data set to obtain a selection agent.

[0120] It should be noted that the multi-dimensional chip text data includes, but is not limited to, materials such as books related to electronic science and technology, industry standard documents, research reports, web forums, and publicly available guides from well-known companies in the industry. The methods of text pre-processing include, but are not limited to, text content extraction, regular expression refinement, noise removal optimization, and manual refinement, etc. Among them, the continued pre-training based on the multi-dimensional chip text data aims to enable the selection agent to better understand and solve problems in the field of electronic information. The proper noun data set and the product classification data set used here are the same as the proper noun data set and the product classification data set shown in the above first mixed data set. The chip selection data set is used to train the chip selection ability of the selection agent, that is, to find chip products that match the target requirement information.

[0121] Different from the above first mixed data set, the chip selection data set in the second mixed data set includes P chip matching data; Exemplarily, the p-th chip matching data includes a chip matching QA pair composed of a preset chip requirement description and a recommended information that meets the preset chip requirement description, and a system prompt word based on the chip matching QA pair; wherein, the preset chip requirement description is generated by randomly extracting a chip product manual; the recommended information that meets the preset chip requirement description is the chip product model and chip product introduction recorded in the randomly extracted chip product manual.

[0122] In a specific application, use to represent the preset chip requirement description, and use to represent the recommended information that meets , then And constitute a chip matching QA pair, and then use to represent the system prompt words of the chip matching QA pair. Based on this, the chip selection data set of the embodiments of the present invention can be expressed as:

[0123]

[0124] Wherein is the number of data in the chip selection data set in, = P.

[0125] Among them, the system prompt words based on the chip matching QA pair are used to instruct the selection agent to use the corresponding rules between the preset chip requirement description and the recommended information that meets the preset chip requirement description demonstrated by the chip matching data of the multi-round dialogue data set, and perform chip selection based on the chip product requirement information to output chip recommendation information.

[0126] As Figure 3 shown, the embodiments of the present invention propose a chip selection box 30 based on a large language model, including an interaction agent 31 driven by the large language model, a selection agent 32, and an evaluation agent 33; wherein, the interaction agent 31 executes as Figure 1 shown in the chip selection method based on the large language model, and the selection agent 32 executes as Figure 2 shown in the chip selection method based on the large language model.

[0127] As Figure 4 shown, the evaluation agent 33 executes the following steps to compare the requirement information output by the interaction agent and the recommended information output by the selection agent, so as to judge whether the chip recommendation information output by the selection agent based on the target requirement information for chip selection and sent to the evaluation agent based on the target chip meets the requirements of the interaction agent, and give a confidence level evaluation and basis:

[0128] SC1. Obtain the chip recommendation information;

[0129] Among them, the chip recommendation information is the result of the selection agent as a generation module, combining the pre-set retrieval module, rearrangement module, and local knowledge base, and performing chip selection based on the chip product requirement information;

[0130] SC2. Perform a confidence evaluation on the chip recommendation information;

[0131] SC3. If the confidence level obtained according to SC2 is the highest level, send the chip recommendation information to the user.

[0132] For the above step SC2, when the embodiment of the present invention evaluates the confidence of the recommended information based on the target chip, evaluation content is also generated. Based on this, after the confidence of the chip recommendation information is evaluated in the above step SC2, a confidence level and evaluation content are generated. Exemplarily, the confidence levels are sorted from high to low, including:

[0133] Fully meet the requirements > Partially meet the requirements > Slightly meet the requirements > Do not meet the requirements.

[0134] Exemplarily, the evaluation content for the confidence level of fully meeting the requirements is: The recommended information based on the target chip fully meets the target requirement information. The evaluation content for the confidence level of partially meeting the requirements is: When the recommended information based on the target chip partially meets the target requirement information. The evaluation content for the confidence level of slightly meeting the requirements is: Only a small amount of the recommended information based on the target chip meets the user's requirements. The evaluation content for the confidence level of not meeting the requirements is: The recommended information based on the target chip completely fails to meet the user's requirements or the selection agent generates irrelevant content.

[0135] In another embodiment of the present invention, based on the steps SC1 to SC3 executed by the evaluation agent 33 as shown in Figure 4 , a processing method for the confidence level that is not the highest level is also proposed, including:

[0136] SC4. If the confidence level obtained according to SC2 is not the highest level, the evaluation content corresponding to the confidence level obtained according to SC2 is sent to the selection agent so that the selection agent changes the chip recommendation information.

[0137] It should be noted that for the selection agent, the chip selection method based on the large language model as shown in Figure 2 may further include:

[0138] SB3. If the evaluation content corresponding to the current confidence level generated when the received confidence level is not the highest level is received, the result of the chip selection in SB2 is changed, and the changed chip recommendation information is output.

[0139] Through the above steps, the selection agent re-performs chip selection according to the content fed back by the evaluation agent, that is, the evaluation content corresponding to the current confidence level, effectively optimizing the accuracy of chip selection.

[0140] For the above steps SC1 to SC3 and the above step SC4, in the embodiments of the present invention, a prompt engineering is designed for the evaluation agent to implement the above chip selection method based on the large language model, especially for implementing step SC2 therein. The prompt engineering is designed into three parts: task clarification prompt, evaluation reference, and evaluation precautions. The task clarification prompt aims to enable the evaluation agent to understand its own confidence evaluation task. The evaluation reference includes examples and descriptions of four-level evaluations. The evaluation precautions include the requirements for the answers of the evaluation agent so that the evaluation agent can give an accurate level evaluation and basis.

[0141] Based on the above detailed descriptions of the interaction agent 31, the selection agent 32, and the evaluation agent 33, the basic principle of the chip selection system 20 based on the large language model in the embodiments of the present invention is as follows:

[0142] The interaction agent continuously interacts with the user. After determining that the user's demand for the chip product is feasible, it extracts the user's chip product demand and sends the result to the selection agent. The selection agent comprehensively analyzes the chip product demand input by the interaction agent and finally inputs the proposed chip product model and related introduction as the result into the evaluation agent. The evaluation agent compares the user demand extracted by the interaction agent with the chip product proposed by the selection agent to determine whether the latter fully meets the requirements of the former, and gives a confidence level evaluation and basis. If the confidence level evaluation given by the evaluation agent is not the highest level, the basis given by the evaluation agent is re-input into the selection agent to change the proposed chip product. If the confidence level evaluation given by the evaluation agent is the highest level, the final recommended chip product, basis, and related product introduction parameters are input to the user.

[0143] Based on the above chip selection system 20 based on the large language model, in the embodiments of the present invention, all interaction information during the interaction between the interaction agent and the user and information extraction is obtained as the interaction information to be processed, and then the interaction information to be processed is subjected to demand analysis to output the data required for the selection agent to perform chip selection, that is, demand information. The selection agent, as a generation module, combines a pre-set retrieval module, a re-ranking module, and a local knowledge base to perform chip selection based on the demand information, realizing the integration of the large language model and retrieval-enhanced generation. In this way, the selection ability can be improved, and the optimal solution can also be output in the case of multiple similar chip products. In addition, the evaluation agent also performs a confidence evaluation on the result of the chip selection by the selection agent, that is, on the recommended information, to ensure the reliability of the recommended information and meet the user's expectations.

[0144] As Figure 5 shown, the embodiments of the present invention also provide a circuit design method, including:

[0145] SD1. Obtain a circuit design request, where the circuit design request includes interaction information describing at least one circuit module; wherein, the interaction information of different circuit modules is different;

[0146] SD2. Decompose the circuit design request into multiple circuit module design tasks and the execution order of the multiple circuit module design tasks;

[0147] SD3. When executing a target circuit module design task, call the chip selection system as described above; the selection system is used to perform requirement analysis based on the circuit design request to obtain chip product requirement information, perform chip selection based on the chip product requirement information to obtain chip recommendation information, and evaluate the confidence level of the chip recommendation information;

[0148] SD4. When the confidence level of the chip recommendation information is at the highest level, obtain a target chip based on the chip recommendation information, and according to the requirements of the target circuit module design task, use the target chip to build a simulation circuit or a test circuit for verification;

[0149] SD5. When the verification is passed, the target chip is the chip used to execute the target circuit module design task;

[0150] SD6. Execute each circuit module design task according to SD3 to SD5 to complete the circuit design.

[0151] Through the above steps SD1 to SD6, the embodiments of the present invention perform decomposition processing on the circuit design request, obtain all circuit modules required to complete the circuit design request, and the design order of each circuit module. Moreover, for each circuit module, chip selection is performed through the above chip selection system, simplifying the existing board-level circuit design process, realizing automatic chip selection for each module of the circuit system, and improving the design efficiency and accuracy of the board-level circuit.

[0152] Figure 6 FIG. shows a schematic structural diagram of an electronic device 60 that can be used to implement the embodiments of the present invention regarding the chip selection method and circuit design method based on a large language model. 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 assistants, 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 claimed herein.

[0153] As shown Figure 6 As shown, the electronic device 60 includes at least one processor 61 and a memory communicatively connected to the at least one processor 61, such as a read-only memory (ROM) 62, a random access memory (RAM) 63, etc. The memory stores a computer program executable by the at least one processor. The processor 61 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 62 or the computer program loaded from the storage unit 68 into the random access memory (RAM) 63. In the RAM 63, various programs and data required for the operation of the electronic device 60 can also be stored. The processor 61, the ROM 62, and the RAM 63 are connected to each other through a bus 64. An input / output (I / O) interface 65 is also connected to the bus 64.

[0154] Multiple components in the electronic device 60 are connected to the I / O interface 65, including: an input unit 66, such as a keyboard, a mouse, etc.; an output unit 67, such as various types of displays, speakers, etc.; a storage unit 68, such as a magnetic disk, an optical disc, etc.; and a communication unit 69, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 69 allows the electronic device 60 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0155] The processor 61 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 61 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 61 executes the various methods and processes described above, such as the optimization method based on the gun-ball linkage monitoring system.

[0156] In some embodiments, the data processing method can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as the storage unit 68. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 60 via the ROM 62 and / or the communication unit 69. When the computer program is loaded into the RAM 63 and executed by the processor 61, one or more steps of the data processing method described above can be executed. Alternatively, in other embodiments, the processor 61 can be configured to execute the data processing method by any other appropriate means (e.g., by means of firmware).

[0157] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0158] The 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, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The computer programs may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0159] In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0160] To provide interaction with a user, the systems and techniques described herein can 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 a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0161] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0162] The computing system can include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for selecting a chip based on a large language model, characterized in that, Including: SA1. Obtain the interaction information to be processed, where the interaction information to be processed is all the interaction information generated during chip selection in the human-machine collaboration mode; SA2. Extract information from the interaction information to be processed to obtain the initial requirement information; SA3. Analyze whether the initial requirement information includes abnormal requirements; the abnormal requirements indicate that the initial requirement information has at least one of unreasonable requirement expressions, ambiguous expressions, and irrelevant expressions; SA4. If so, feedback the abnormal requirements and the reasons for the appearance of the abnormal requirements to the user, and at the same time interact with the user to generate new interaction information, and execute SA2 according to the newly generated interaction information; SA5. If not, perform requirement analysis and standardization processing according to the initial requirement information to obtain chip product requirement information with data item names including at least product classification based on the initial requirement information and output it; Among them, the first hybrid dataset including a proper noun dataset, a product classification dataset, a multi-round dialogue dataset, and a general dataset is used to supervise and fine-tune the training of the first basic intelligent agent to obtain an interaction intelligent agent to implement SA1 to SA5; Among them, the product classification dataset includes N category data contents based on M first-level chip categories; If the nth category data content belongs to the mth first-level chip category, the nth category data content includes a product classification QA pair composed of a second-level chip category subordinate to the mth first-level chip category and a question statement based on the second-level chip category subordinate to the mth first-level chip category, and a system prompt word based on the product classification QA pair; Among them, M and N are positive integers, n is a positive integer less than or equal to N, and m is a positive integer less than or equal to M.

2. The method for chip selection based on large language models according to claim 1, wherein The proper noun dataset includes K noun data contents based on K proper nouns; The kth noun data content includes a proper noun QA pair composed of the correct explanation of the kth proper noun and a question statement based on the correct explanation of the kth proper noun, and a system prompt word based on the proper noun QA pair; Among them, k is a positive integer less than or equal to K, and K is a positive integer.

3. The method for selecting a chip based on a large language model according to claim 1, wherein, The multi-round dialogue dataset includes U multi-round dialogue data contents based on L multi-round dialogues; If the u-th multi-round dialogue data content belongs to the l-th multi-round dialogue, the u-th multi-round dialogue data content includes a multi-round dialogue QA pair composed of a simulated question in the l-th multi-round dialogue and a reply content based on the simulated question in the l-th multi-round dialogue, a system prompt word based on the multi-round dialogue QA pair, and a historical dialogue set; Among them, L and U are positive integers, l is a positive integer less than or equal to L, u is a positive integer less than or equal to U. In the l-th multi-round dialogue, the simulated question is generated by simulating the user's question about a randomly selected chip product manual; the reply content based on the simulated question in the l-th multi-round dialogue is generated by the randomly selected chip product manual.

4. The method for selecting a chip based on a large language model according to claim 3, wherein In the l-th multi-round conversation, the simulated questions are also obtained by randomly selecting from a preset question database. For the simulated questions obtained by randomly selecting from the preset question database, the reply content includes the content of the randomly selected chip product manual. Among them, the preset question database includes multiple preset questions, and each preset question includes at least one of unreasonable requirement expressions, ambiguous expressions, and irrelevant expressions.

5. A chip selection method based on a large language model, applied to an interactive agent, wherein the interactive agent executes the chip selection method based on a large language model according to any one of claims 1 to 4, characterized in that, Including: SB1. Obtain the chip product requirement information provided by the interaction agent; SB2. The selection agent, as a generation module, combines a pre-set retrieval module, a re-rank module, and a local knowledge base, and performs chip selection based on the chip product requirement information, and outputs chip recommendation information. The SB2 includes: Compare the chip product requirement information with the chip product data stored in the local knowledge base through the retrieval module to obtain multiple relevant documents and send them to the re-rank module; Perform relevance ranking on multiple relevant documents through the re-rank module, and send the ranking result to the selection agent; in the ranking result, the higher the relevance of the relevant document, the smaller the ranking serial number, and the relevant document with the ranking serial number less than the preset serial number value is a strongly relevant document; Analyze all strongly relevant documents according to the chip product requirement information, and output chip recommendation information.

6. The method for selecting a chip based on a large language model according to claim 5, wherein Construct a retrieval triple dataset to perform supervised contrastive learning training on the basic embedding model. After the training is completed, obtain the retrieval module; Among them, the constructed retrieval triple dataset includes a retrieval query element, a retrieval positive sample element, and a retrieval negative sample element; the retrieval query element is extracted from the total demand information data; the retrieval positive sample element is the only chip product data selected from the local knowledge base to reply to the retrieval query element; the retrieval negative sample element is multiple chip product data randomly selected from the local knowledge base.

7. The method for chip selection based on large language models according to claim 5, wherein Construct a re-rank triple dataset to perform supervised contrastive learning training on the basic re-rank model. After the training is completed, obtain the re-rank module; Among them, the constructed re-rank triple dataset includes a re-rank query element, a re-rank positive sample element, and a re-rank negative sample element; the re-rank query element is a simulated demand description, and the data size of the simulated demand description is less than the preset byte number; the re-rank positive sample element is the only correct answer written based on the simulated demand description; the re-rank negative sample element is multiple wrong answers written based on the simulated demand description, and the similarity between the wrong answer and the only correct answer is greater than the preset similarity.

8. The method for chip selection based on large language models according to claim 5, wherein Any chip product data stored in the local knowledge base includes a document description of product features, a document description of product application examples, a document description of key parameters, and a document description of complete information.

9. The method for selecting a chip based on a large language model according to claim 5, wherein Perform two-stage training on the second basic intelligent agent through continued pre-training and supervised fine-tuning training to obtain a selection agent to implement the SB1 to SB2; In the continued pre-training, obtain multi-dimensional chip text data, and perform text preprocessing on the multi-dimensional chip text data to obtain multiple corpus blocks, and train the second basic intelligent agent through the multiple corpus blocks. In the supervised fine-tuning training, the second base agent after continued pre-training is trained with a second mixed dataset including a proper noun dataset, a product classification dataset, a chip selection dataset, and a general dataset to obtain a selection agent.

10. The method for selecting a chip based on a large language model according to claim 9, wherein The chip selection dataset includes P chip matching data. The p-th chip matching data includes a chip matching QA pair composed of an analog chip requirement description and recommended information that meets the analog chip requirement description, and a system prompt word based on the chip matching QA pair. Where p is a positive integer less than or equal to P, and P is a positive integer. The analog chip requirement description is generated from a randomly selected chip product manual. The recommended information that meets the analog chip requirement description is the chip product model number and chip product introduction recorded in the randomly selected chip product manual.

11. A chip selection system based on a large language model, characterized in that, It includes an interaction agent driven by a large language model, a selection agent, and an evaluation agent; the interaction agent executes the chip selection method based on the large language model according to any one of claims 1 to 4; the selection agent executes the chip selection method based on the large language model according to any one of claims 5 to 10; the evaluation agent executes the following steps: SC1. Obtain chip recommendation information. Where the chip recommendation information is the result of the selection agent as a generation module, in combination with a pre-set retrieval module, a re-ranking module, and a local knowledge base, for chip selection based on chip product requirement information. SC2. Conduct a confidence evaluation on the chip recommendation information. SC3. If the confidence level obtained according to SC2 is the highest level, send the chip recommendation information to the user.

12. The chip selection system according to claim 11, wherein, In SC2, when conducting a confidence evaluation on the chip recommendation information, evaluation content is also generated.

13. The chip selection system according to claim 12, wherein It further includes: SC4. If the confidence level obtained according to SC2 is not the highest level, send the evaluation content corresponding to the confidence level obtained according to SC2 to the selection agent so that the selection agent changes the chip recommendation information.

14. The chip selection system according to claim 11, wherein The evaluation agent is designed through prompt engineering, and SC1 to SC3 are implemented through the evaluation agent.

15. A circuit design method, characterized in that, It includes: SD1. Obtain a circuit design request, where the circuit design request includes interaction information describing at least one circuit module; the interaction information of different circuit modules is different. SD2. Decompose the circuit design request into multiple circuit module design tasks and the execution order of the multiple circuit module design tasks. When executing the target circuit module design task, call the chip selection system according to any one of claims 11 to 14; the selection system is used to perform requirement analysis on the circuit design request to obtain chip product requirement information, perform chip selection according to the chip product requirement information to obtain chip recommendation information, and conduct a confidence evaluation on the chip recommendation information. SD4. When the confidence level of the chip recommendation information is the highest level, obtain a target chip based on the chip recommendation information, and build a simulation circuit or a test circuit using the target chip for verification according to the requirements of the target circuit module design task. SD5. When the verification is passed, the target chip is the chip used to execute the target circuit module design task; SD6. Execute each circuit module design task according to SD3 to SD5 to complete the circuit design.

16. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the large language model-based chip selection method according to any one of claims 1 to 4, or the large language model-based chip selection method according to any one of claims 5 to 10, or the circuit design method according to claim 15.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the large language model-based chip selection method according to any one of claims 1 to 4, or the large language model-based chip selection method according to any one of claims 5 to 10, or the circuit design method according to claim 15.

18. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the large language model-based chip selection method according to any one of claims 1 to 4, or the large language model-based chip selection method according to any one of claims 5 to 10, or the circuit design method according to claim 15.

Images