A computer organization and principle course auxiliary question and answer method, device and medium
By cleaning the dataset and training the ChatGLM3-6B model with LoRa, and combining it with a text vector knowledge base, the problem of insufficient accuracy of answers in computer organization courses using large language models is solved. This achieves efficient and low-cost domain adaptation and provides multi-turn question answering and knowledge base retrieval functions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YANSHAN UNIV
- Filing Date
- 2026-03-09
- Publication Date
- 2026-06-09
AI Technical Summary
Existing large language models are insufficient to guarantee the accuracy and relevance of answers in computer organization courses, and there is a lack of efficient and low-cost domain adaptation solutions.
Data from the Computer Organization Principles course was collected, cleaned, and formatted to form a question-and-answer dataset. The ChatGLM3-6B model was trained in a supervised manner using the LoRa method to generate incremental weights. A text vector knowledge base was built, and a WebUI interface was provided using the LangChain-Chatchat framework for question answering.
It ensures the accuracy and relevance of answers in the Computer Organization Principles course, reduces development costs, improves learning assistance efficiency, and supports multi-round question answering and knowledge base retrieval.
Smart Images
Figure CN122174995A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of natural language processing technology, and in particular to a method, device and medium for assisting in question answering in a computer organization course. Background Technology
[0002] With the rapid development of artificial intelligence technology, large language models (LMs) have gradually become efficient solutions in the field of natural language processing. These models, pre-trained on massive amounts of multi-domain text data, acquire rich linguistic knowledge and general reasoning capabilities, enabling them to adapt to the needs of various downstream tasks. For example, open-source models such as ChatGLM3-6B have demonstrated good performance in general question answering and text generation scenarios.
[0003] However, despite the strong generalization ability of large language models, they still have significant limitations in highly specialized, knowledge-intensive vertical fields—such as professional courses like Computer Organization and Architecture. The content of Computer Organization and Architecture is complex, encompassing both abstract theories such as CPU architecture, instruction sets, memory hierarchy, and bus control, and hands-on practice based on hardware description languages or experimental platforms, making it a high-barrier-to-entry learning experience. Current general-purpose large language models often struggle to guarantee the accuracy and relevance of answers when dealing with specific problems in this field, easily producing responses with missing information, vague expressions, or those inconsistent with the course's focus. This makes them ineffective in supporting students' pre-study, review, course design, and experimental debugging.
[0004] Currently, if we want to deeply integrate domain-specific expertise into a large language model, we usually need to pre-train it on large-scale domain data from scratch. However, this requires high computing power, a long training period, and a high-quality domain corpus, which is impractical for most educational institutions or research teams. Therefore, how to adapt existing general-purpose large language models to specific domain knowledge systems and task requirements in an efficient and low-cost manner has become an urgent technical problem to be solved.
[0005] Currently, there is a lack of large language model application solutions that specifically address the knowledge system of computer organization principles courses through supervised fine-tuning and integrate retrieval mechanisms. Existing technologies still have significant shortcomings in terms of deep adaptation, knowledge accuracy, and teaching relevance in this vertical field. Summary of the Invention
[0006] The purpose of this invention is to provide a method, device, and medium for assisting in answering questions in a computer organization principles course, aiming to solve or improve at least one of the aforementioned technical problems.
[0007] To achieve the above objectives, the present invention provides the following solution: A method, device, and medium for assisting in question-and-answer sessions in a computer organization principles course, comprising: Data related to the Computer Organization Principles course was collected, cleaned, converted, and formatted using multi-turn dialogue formats to form a question-and-answer dataset for training. The relevant data included online consultation data, professional textbooks, question banks, experimental manuals, and code examples. Based on the question-and-answer dataset, the ChatGLM3-6B model is trained in a supervised manner using the Lora method to generate incremental weights suitable for question-and-answer in the Computer Organization Principles course, and the trained language question-and-answer model is saved locally. A text vector knowledge base is built by vectorizing course-related documents and questions using a semantic vector model. The text vector knowledge base includes a vector database that supports user uploading and updating of documents, as well as a custom knowledge base that supports code retrieval and question answering. The LangChain-Chatchat framework provides a WebUI interface that allows users to select a question-answering mode based on the question type, including a dialogue mode based on fine-tuning a large model and a retrieval question-answering mode based on a knowledge base; the LangChain-Chatchat framework is used to load the trained language question-answering model.
[0008] Optionally, the collection of relevant data from the Computer Organization Principles course, followed by cleaning, format conversion, and multi-turn dialogue formatting to form a question-and-answer dataset for training, specifically includes: The web crawler code was used to collect online platform data on questions and inquiries about the Computer Organization Principles course, and to collect code examples of course-related e-books, exam questions, lab manuals, and lab projects. The pandas library was used to clean the data on the online platform, removing irrelevant characters, punctuation marks, and HTML tags. Missing data was filled in, stop words were removed, and the data was formatted and saved as a CSV or TXT dataset for the Computer Organization Principles course Q&A dataset.
[0009] Optionally, based on the question-answering dataset, the ChatGLM3-6B model is trained in a supervised manner using the LoRa method to generate incremental weights suitable for question-answering in the Computer Organization Principles course, and the trained language question-answering model is saved locally, specifically including: The question-and-answer dataset is loaded using the load function, and the dialogue content is read to construct a list of messages. The source dataset does not have the "system" role, so the "system" role is automatically added using code and must be added before the "user" role. The prompt words are modified to guide the large model to simulate the thinking mode of a computer science course instructor for training. The processed training dataset is then converted into a JSON file and saved locally. The relevant parameters are set and divided into two parts: pre-trained large model weights and fine-tuned incremental weights. The pre-trained large model weights are the weights of the ChatGlm3-6B large model, and the fine-tuned incremental weights are the weight updates generated during the training of the large model. During training, let the input dataset be x and the output be y, and let y = wx + ... wx, with pre-trained weights w, and parameter update variables constructed using low-rank separable matrices A and B. w, train the newly added A and B, and update the final weights of the model to the sum of the product of the pre-trained weights w and the low-rank matrices A and B: y=(w+AB)x; The large model with final weights is converted into a quantized model file format and saved to a local folder.
[0010] Optionally, after converting the large model with determined final weights into a quantized model file format, the process also includes: Visualize the model's loss and feature maps using TensorBoardX, and verify the model's performance based on the trend curve of the loss val_loss.
[0011] Optionally, the step of using a semantic vector model to vectorize course-related documents and questions to build a text vector knowledge base specifically includes: First, the user-uploaded long text data is loaded. Then, the long text data is read and segmented into paragraphs or sentences of a set size. Next, the segmented paragraphs or sentences are vectorized using the pre-trained semantic vector model bge-large-zh-v1.5. Finally, a Faiss vector database is built to store and retrieve the vectorized text fragments, and the Faiss vector database is saved locally. The semantic vector model bge-large-zh-v1.5 is used to vectorize user-submitted questions related to the Computer Organization Principles course, and a text vector knowledge base is built for custom agent question answering.
[0012] Optionally, the text vector knowledge base used by the custom Agent question answering is built based on ChatGLM3-6B or qwen-api, and has the ability to load and query the local experimental knowledge base and generate code answers.
[0013] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the computer organization principle course auxiliary question-and-answer method according to the above.
[0014] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the computer organization principles course auxiliary question-and-answer method as described above.
[0015] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This invention discloses an auxiliary question-answering method, device, and medium for a computer organization and architecture course. The method includes collecting relevant data for the computer organization and architecture course, cleaning, converting formats, and organizing multi-turn dialogue formats to form a question-answering dataset for training. The relevant data includes online consultation data, professional textbooks, question banks, experimental manuals, and code examples. Based on the question-answering dataset, the ChatGLM3-6B model is trained in a supervised manner using the LoRa method to generate incremental weights suitable for question-answering in the computer organization and architecture course, and the trained language question-answering model is saved locally. A semantic vector model is used to vectorize course-related documents and questions to build a text vector knowledge base. The text vector knowledge base includes a vector database that supports user uploads and updates of documents and a custom knowledge base that supports code retrieval and question-answering. A WebUI interface is provided based on the LangChain-Chatchat framework, allowing users to select a question-answering mode according to the question type, including a dialogue mode based on fine-tuning a large model and a retrieval question-answering mode based on the knowledge base. The LangChain-Chatchat framework is used to load the trained language question-answering model.
[0016] This invention successfully utilizes the tools and interfaces provided by the LangChain-Chatchat framework to rapidly establish a knowledge base for computer organization principles courses. Users can subsequently upload and update the knowledge base based on newly added computer organization principles course-related literature or supplementary materials, and the knowledge base provides responses through retrieval, thus significantly saving development time in this field. The finely tuned ChatGlm3-6B large model is loaded into LangChain-Chatchat. This model, as an open-source, lightweight pre-trained model, has a low deployment threshold and can be easily run on devices with limited computing resources. Therefore, this invention not only fills the gap in intelligent teaching research for computer organization principles courses but also has broad practical application value.
[0017] This invention is based on the construction of a multi-turn dialogue data to fine-tune the ChatGlm3-6B large model, so that the method proposed in this invention provides answers that are more in line with the application scenario and answer questions more concisely and accurately compared with the basic large language model.
[0018] This invention utilizes the webUI technology within the LangChain-Chatchat framework, enabling not only direct large-scale model inference and multi-round question answering, but also the integration of a pre-built knowledge base to further enhance model performance. It selects the appropriate answer method based on the user's question, thus providing great convenience for users to solve practical problems. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a flowchart illustrating the auxiliary question-and-answer method for the computer organization principles course of this invention. Figure 2 This is a schematic diagram of the process of vectorizing and storing the knowledge base in this embodiment; Figure 3 This is a schematic diagram of the knowledge base retrieval process in this embodiment. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] The purpose of this invention is to provide a method, device, and medium for assisting in answering questions in a computer organization principles course, aiming to solve or improve at least one of the aforementioned technical problems.
[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0024] like Figure 1 As shown, this invention provides an auxiliary question-and-answer method for a computer organization principles course, including: Step 1: Collect relevant data from the Computer Organization Principles course, and clean, convert, and organize the multi-turn dialogue format to form a question-and-answer dataset for training; the relevant data includes online consultation data, professional textbooks, question banks, experimental manuals, and code examples.
[0025] As one specific implementation method, the detailed processing procedure of this step includes: We collected consultation data from websites, blogs, and online forums, as well as supplementary course datasets from professional course textbooks, lecture notes, lab guides, lab manuals, and exam question banks. The question-and-answer data was preprocessed using Python code and converted into multi-turn dialogue-style text, then divided into training and test sets at an 8:2 ratio. Other documents were converted into formats that can be uploaded to the knowledge base, including PDF, TXT, and DOC.
[0026] Step 2: Based on the question-and-answer dataset, the ChatGLM3-6B model is trained in a supervised manner using the LoRa method to generate incremental weights suitable for question-and-answer in the Computer Organization Principles course, and the trained language question-and-answer model is saved locally.
[0027] As one specific implementation method, the detailed processing procedure of this step is as follows: Step 1: Load the base model ChatGlm3-6B large model. Because it is better at dialogue functions and can be deployed quickly, it is a large language foundation model with powerful language understanding capabilities, making it easier to implement normal multi-turn dialogue functions.
[0028] Step Two: Fine-tuning the ChatGlm3-6B large-scale model with multi-turn question-answering data. Multi-turn dialogue data was constructed using Python code. The dialogue data recording the interaction between the model and the user was reorganized. Some dialogue data scraped from the webpage lacked system roles and dialogue content, which needed to be automatically filled in using Python code. Finally, the dataset was divided into training and test sets and saved as the corpus for fine-tuning the large-scale model.
[0029] The dataset preparation process specifically includes: The `load` function loads a CSV-formatted question-and-answer dataset, reads the dialogue content, and constructs a list of messages. Since the original dataset lacks the "system" role, code automatically adds the "system" role, which must precede the "user" role. The prompt words are modified to guide the large model in simulating the thought process of a computer science course instructor. After training the large model, more professional and in-depth answers can be obtained. The dataset is divided into training and validation sets in an 8:2 ratio and converted to JSON format for local storage.
[0030] Step 3: Set relevant parameters and begin fine-tuning the large model. Lora's large model fine-tuning involves two parts: one part keeps the original pre-trained weight parameters unchanged, and the other part adjusts the incremental weight parameters. This invention mainly adjusts the incremental weight parameters according to specific tasks, such as the dimension r of the low-rank matrix, setting the scaling ratio lora_alpha of the low-rank matrix relative to the original weight parameters, dynamically adjusting the two parameters based on the number of samples, adjusting the ratio of lora_alpha to r based on the loss curve of the validation set, reducing the value of lora_alpha to prevent overfitting, and increasing the value of the lora_dropout parameter.
[0031] The LoRa fine-tuning of large model parameters is divided into two parts: pre-training the large model weights (i.e., downloading the ChatGlm3-6B large model and saving it to a folder), and fine-tuning the incremental weights (i.e., saving the weight update output generated during the large model training process to a folder). Let the input dataset be x and the output be y, y = wx + During training, the pre-trained weights w are fixed, and the parameter update amounts are approximated using two low-rank separable matrices A and B. During training, new A and B are added to the model to obtain the final model weights y = (w + AB)x.
[0032] During the aforementioned LoRa fine-tuning process w is decomposed into two low-rank separable matrices, namely two initialized low-rank matrices A and B. A is initialized using Gaussian initialization, and B is initialized to a zero matrix. For example, The matrix w of size (d, k) can be decomposed into two matrices of size Ad. r, Br k, then the low-rank matrix mentioned last is actually the product of A and B. According to the core formula proposed by the LoRa fine-tuning principle, the weights after model fine-tuning should be updated to the sum of the original pre-trained weights w and the product of the low-rank matrices A and B, that is, y=(w+AB)x.
[0033] Detailed methods for setting fine-tuning parameters include: "finetuning_type": lora, "lora_rank( r)": 16, "lora_alpha": 32, "lora_dropout": 0.1 Detailed methods for setting training parameters include: "n_epochs": 5, "batch_size": 8, "learning_rate": le-4, "max_tokens": 4096 Step 4: Begin fine-tuning the large model. Save the quantized, fine-tuned large model to a local folder for later model inference. Use TensorBoardX to visualize the loss of the large model and observe the loss trend curve (val_loss) on the validation set to evaluate the effectiveness of the fine-tuning. Select the fine-tuned large model with the lowest loss as the completed model.
[0034] The formula for calculating the loss trend curve val_loss is shown below: in, For the t-th target word, For context sequence, This represents the model parameters. During training, only the Lora matrices A and B are optimized, while the original model weights remain frozen. The Lora parameters are updated via backpropagation to minimize this loss.
[0035] Step 3: Vectorize course-related documents and questions using a semantic vector model to build a text vector knowledge base; the text vector knowledge base includes a vector database that supports user uploading and updating of documents, and a custom knowledge base that supports code retrieval and question answering.
[0036] As one specific implementation method, the detailed processing procedure of this step is as follows: First, users upload datasets, including course-related textbooks, exam question banks, lab manuals, and Q&A data, to the pre-built "Computer Organization" knowledge base. The uploaded documents are loaded, preprocessed, and split. A semantic vector model is used to vectorize the documents; adding custom keywords improves the accuracy of vectorization. The vectorized documents are then stored in the database. When a user submits a question, the semantic vector model (BGE) vectorizes the question and then uses the Faiss library to retrieve the document vectors most similar to the question in the knowledge base. The k most similar text segments are selected and submitted to the fine-tuned ChatGlm3-6B main model. Next, a knowledge base for a custom agent question-answering mechanism is built. Course lab project files are split into multiple smaller files and uploaded to the constructed "Computer Organization Experiment" knowledge base. When a user inquires about lab code-related questions, the custom agent can use a tool to query the "Computer Organization Experiment" knowledge base to answer the code question. The custom agent question-answering model can be based on the ChatGlm3-6B main model or on the qwen-api online answering platform. Through these methods, users obtain more professional and relevant answers.
[0037] As a more specific approach: The process of vectorizing and storing knowledge bases is as follows Figure 2 As shown, the process includes: first, loading long texts such as course materials and exam questions uploaded by users; document_loaders provides functions for loading various types of documents, including unstructured long texts; second, reading the long text and segmenting it into paragraphs or sentences of appropriate size; then, vectorizing the segmented paragraphs or sentences using a pre-trained semantic vector model bge-large-zh-v1.5; and finally, establishing a Faiss vector database to store and retrieve the vectorized text fragments. For ease of use, the Faiss vector database is saved locally.
[0038] The process of knowledge base retrieval is as follows Figure 3 As shown, this includes: users submitting questions related to the Computer Organization Principles course via a webpage. Similarly, the questions are vectorized using the semantic vector model bge-large-zh-v1.5. A pre-built retrieval tool retrieves the k most similar text vector segments from the knowledge base, and the value of k is adjusted according to the actual application scenario. A nearest neighbor similarity algorithm is used to calculate the similarity score between the text vectors and the question vectors. The k text segments with the highest scores are submitted along with the question to prompt, which feeds into the fine-tuned chatglm3-6b model, ultimately obtaining a professional answer, which is then displayed to the user on the webpage.
[0039] The process of building the knowledge base used for custom agent question answering includes: constructing the knowledge base for custom agent question answering; organizing code files containing machine instructions and micro-instructions, including experiment names, mnemonic symbols, content, and instruction functions; splitting the code into multiple files and loading them into the custom computer organization experiment knowledge base; implementing a custom tool to query code based on the computer organization experiment knowledge base; writing code to call the custom tool; modifying the prompt (template) format for agent chat in the configuration file to be suitable for answering professional questions in the Computer Organization Principles course; and modifying the system so that custom agent question answering can answer questions related to experiment code based on queries to the computer organization experiment knowledge base.
[0040] Step 4: Provide a WebUI interface based on the LangChain-Chatchat framework, allowing users to select the question-answering mode according to the question type, including a dialogue mode based on fine-tuning a large model and a retrieval question-answering mode based on a knowledge base; the LangChain-Chatchat framework is used to load the trained language question-answering model.
[0041] As one specific implementation method, the detailed processing procedure of this step includes: Download and configure the Langchain-Chatchat project to use the fine-tuned ChatGLM3-6B model. Run the LangChain-Chatchat's built-in WebUI interface, enter a question on the page, select the answer method based on the question, and call the fine-tuned ChatGlm3-6B main model to answer complex theoretical questions or retrieve answers from the built knowledge base. Based on the ChatGlm3-6B main model or qwen-api, call a custom agent to query the experimental knowledge base to obtain more accurate and detailed code. Users can upload their accumulated professional course materials to the built knowledge base, thereby continuously updating and expanding the content of the knowledge base.
[0042] When using this method, users can choose the answering method based on the specific question. For example, for complex theoretical questions involving the course, they can choose to answer based on the finely tuned ChatGlm3-6B large model, referencing the pre-built "Computer Organization" knowledge base to retrieve questions from the question bank or textbook. For coding questions involving course experiments, they can choose to consult the machine instructions and micro-instructions needed based on a custom agent. This computer organization principles course auxiliary question-and-answer method proposed in this invention helps users understand course theoretical knowledge, improves their professional course design capabilities, and guides them to accurately complete experimental code writing.
[0043] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0044] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A question-and-answer auxiliary method for a computer organization principles course, characterized in that, include: Data related to the Computer Organization Principles course was collected, cleaned, converted, and formatted using multi-turn dialogue formats to form a question-and-answer dataset for training. The relevant data included online consultation data, professional textbooks, question banks, experimental manuals, and code examples. Based on the question-and-answer dataset, the ChatGLM3-6B model is trained in a supervised manner using the Lora method to generate incremental weights suitable for question-and-answer in the Computer Organization Principles course, and the trained language question-and-answer model is saved locally. A text vector knowledge base is built by vectorizing course-related documents and questions using a semantic vector model. The text vector knowledge base includes a vector database that supports user uploading and updating of documents, as well as a custom knowledge base that supports code retrieval and question answering. The LangChain-Chatchat framework provides a WebUI interface that allows users to select a question-answering mode based on the question type, including a dialogue mode based on fine-tuning a large model and a retrieval question-answering mode based on a knowledge base; the LangChain-Chatchat framework is used to load the trained language question-answering model.
2. The auxiliary question-and-answer method for computer organization principles course according to claim 1, characterized in that, The process involves collecting relevant data from the Computer Organization Principles course, cleaning, converting, and formatting the data according to multi-turn dialogue formats to form a question-and-answer dataset for training. Specifically, this includes: The web crawler code was used to collect online platform data on questions and inquiries about the Computer Organization Principles course, and to collect code examples of course-related e-books, exam questions, lab manuals, and lab projects. The pandas library was used to clean the data on the online platform, removing irrelevant characters, punctuation marks, and HTML tags. Missing data was filled in, stop words were removed, and the data was formatted and saved as a CSV or TXT dataset for the Computer Organization Principles course Q&A dataset.
3. The auxiliary question-and-answer method for computer organization principles course according to claim 1, characterized in that, Based on the question-answering dataset, the ChatGLM3-6B model is trained in a supervised manner using the LoRa method to generate incremental weights suitable for question-answering in the Computer Organization Principles course. The trained language question-answering model is then saved locally. Specifically, this includes: The question-and-answer dataset is loaded using the load function, and the dialogue content is read to construct a list of messages. The source dataset does not have the "system" role, so the "system" role is automatically added using code and must be added before the "user" role. The prompt words are modified to guide the large model to simulate the thinking mode of a computer science course instructor for training. The processed training dataset is then converted into a JSON file and saved locally. The relevant parameters are set and divided into two parts: pre-trained large model weights and fine-tuned incremental weights. The pre-trained large model weights are the weights of the ChatGlm3-6B large model, and the fine-tuned incremental weights are the weight updates generated during the training of the large model. During training, let the input dataset be x and the output be y, and let y = wx + ... wx, with pre-trained weights w, and parameter update variables constructed using low-rank separable matrices A and B. w, train the newly added A and B, and update the final weights of the model to the sum of the product of the pre-trained weights w and the low-rank matrices A and B: y=(w+AB)x; The large model with final weights is converted into a quantized model file format and saved to a local folder.
4. The auxiliary question-and-answer method for computer organization principles course according to claim 3, characterized in that, After converting the large model with determined final weights into a quantized model file format, the process also includes: Visualize the model's loss and feature maps using TensorBoardX, and verify the model's performance based on the trend curve of the loss val_loss.
5. The auxiliary question-and-answer method for computer organization principles course according to claim 1, characterized in that, The method of vectorizing course-related documents and questions using a semantic vector model to build a text vector knowledge base specifically includes: First, the user-uploaded long text data is loaded. Then, the long text data is read and segmented into paragraphs or sentences of a set size. Next, the segmented paragraphs or sentences are vectorized using the pre-trained semantic vector model bge-large-zh-v1.
5. Finally, a Faiss vector database is built to store and retrieve the vectorized text fragments, and the Faiss vector database is saved locally. The semantic vector model bge-large-zh-v1.5 is used to vectorize user-submitted questions related to the Computer Organization Principles course, and a text vector knowledge base is built for custom agent question answering.
6. The auxiliary question-and-answer method for computer organization principles course according to claim 5, characterized in that, The text vector knowledge base used for the custom Agent question answering is built on ChatGLM3-6B or qwen-api, and has the ability to load and query the local experimental knowledge base and generate code answers.
7. An electronic device, characterized in that, It includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the computer organization principle course auxiliary question and answer method according to any one of claims 1-6.
8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the computer organization principles course auxiliary question-and-answer method as described in any one of claims 1-6.