A geological text generation method, system and device based on a large language model

Abstract

The application relates to the technical field of information processing, and specifically discloses a geological text generation method, system and device based on a large language model, which comprises the steps of text data collection, data set construction, data preprocessing, large model construction and model fine-tuning, and realizes intelligent text generation of stratum lithology description based on a deep learning large language model. The application trains a ChatGLM large language model, fine-tunes based on P-Tuning v2 technology, and simultaneously incorporates geotechnical field knowledge, so that the model can understand and apply professional knowledge of geological lithology, thereby intelligently generating stratum lithology description text meeting the requirements of an engineering geology segmented specification according to known lithology description information, the stratum lithology description text being good in continuity and high in logic, the whole method being simple to operate, capable of efficiently improving the efficiency of artificial data collation and analysis, and convenient to popularize and apply.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and specifically to a method, system, and apparatus for generating geological text based on a large language model. Background Technology

[0002] Geology, as the study of the Earth, has a broad scope, ranging from the Earth's surface to its deep crust, from minerals and rocks to oil and gas resources, from paleoenvironmental changes to modern geological processes. With advancements in technology, the collection of geological data has become increasingly efficient, resulting in an explosive growth in data volume. This geological data includes not only text-based survey reports and research papers, but also various formats such as images, tables, and maps. Traditional data processing methods struggle to handle such massive amounts of data, and different types of data require different processing methods. How to extract useful information from massive amounts of data and transform it into scientific knowledge is a major challenge facing geological research.

[0003] The application of deep learning large language models in the field of geology is gradually developing. The development of a geological text generation method based on large language models with good coherence and logic is of great significance to geological research. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, and apparatus for producing geological text based on a large language model, addressing the problems existing in the prior art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A geological text generation method based on a large language model includes the following steps:

[0007] S1. Collect geological text data and extract the stratigraphic and lithological descriptions.

[0008] S2. The content of the stratigraphic lithology description is segmented into words to form a geotechnical description dataset of stratigraphic lithology.

[0009] S3. The soil and rock description dataset is deduplicated, cleaned, and serialized to obtain an optimized dataset.

[0010] S4. Use the optimized dataset to pre-train the language model ChatGLM, incorporate knowledge from the geological and geotechnical fields, and obtain a large language model for the geological field.

[0011] S5. Fine-tune the large language model of the geological domain based on P-Tuning v2 to obtain a lightweight large language model of the geological domain.

[0012] S6. Input the original soil and rock description data into the lightweight geological domain large language model to generate the expected geological text.

[0013] This invention provides a geological text generation method based on a large language model, including steps such as text data collection, dataset construction, data preprocessing, large model construction, and model fine-tuning. It achieves intelligent text generation of stratigraphic and lithological descriptions based on a deep learning large language model. This invention trains the ChatGLM large language model and fine-tunes it using P-Tuning v2 technology, while incorporating knowledge from the geotechnical field. This enables the model to understand and apply professional geological and lithological knowledge, thereby intelligently generating stratigraphic and lithological description text that meets the requirements of engineering geological segmentation specifications based on known lithological description information. The text exhibits good continuity and high logical consistency. The entire method is simple to operate, significantly improving the efficiency of manual data processing and analysis, and is easy to promote and apply.

[0014] Furthermore, in S1, the collected geological textual data includes at least one of the project geological description reports and scientific research papers, and / or geological textual data collected from databases.

[0015] The databases are geological information databases, geological spatial databases, geological research databases, geological public databases, or geological environment databases.

[0016] The geological text generation method based on a large language model provided by this invention uses geological text data collected from a database and geological segmentation instructions written by geological experts in accordance with standards to obtain relatively accurate geological text data, which contains professional knowledge in the geological field, and provides accuracy assistance for geological text knowledge processing and intelligent text generation.

[0017] Furthermore, S2 also includes searching and matching the database based on the geotechnical description dataset of the strata lithology; data that does not pass the matching are manually filtered and labeled, and then added to the geotechnical description dataset.

[0018] This application retrieves and matches the obtained stratigraphic lithology description dataset to the database, and then manually filters and annotates it to obtain a geotechnical description dataset. This data processing provides an accuracy guarantee for subsequent geological text generation.

[0019] Furthermore, in S2, based on the segmented information in the engineering geology database, relevant information such as the name, number, core layer description, weathering degree, and depth are retrieved and matched with the stratigraphic lithology description text of the corresponding segment in the engineering geology manual. If no match is found, it is manually marked according to geological specifications. The matched data is then cleaned and data serialization enhancement is performed to construct the final engineering geology stratigraphic lithology description dataset.

[0020] Furthermore, in S2, the geotechnical description dataset includes the names and characteristics of the lithology of the strata.

[0021] Furthermore, the specific deduplication operation in S3 is as follows: compare all segment reports in the geotechnical description dataset, find the segments with duplicate content due to close proximity, and then make a comprehensive judgment based on distance, time and content factors, and finally retain the data in one segment report and delete the remaining duplicate reports.

[0022] Preferably, the specific deduplication operation in S3 is as follows: compare all segment reports in the geotechnical description dataset, find the segments with duplicate content due to the close proximity of the predicted mileage segments, and filter them in turn according to the completeness of the fields contained in the content, the order of drilling time, and the breadth of the predicted mileage segments. Finally, retain the data from one segment report with a wider mileage, more keywords in the stratigraphic and lithological description, and a more recent drilling time, and delete the remaining duplicate reports; for example, the predicted mileage segment paradigm is D4K280+898.5~+946.1.

[0023] Furthermore, in S4, the specific pre-training process is as follows:

[0024] S41. Divide the optimized dataset into a training set and a test set in a ratio of 70%-80%:30%-20%, generate a JSON file containing the training corpus, and save it.

[0025] S42. Input the text with training corpus into the ChatGLM large-scale language model for learning and training, and normalize the model output.

[0026] Furthermore, in S42, the learning and training process is as follows:

[0027] The preset template for the original text input data is a text with n (n is a natural number greater than or equal to 3) entries (preferably, the entry text is text fragments at the semantic level, sentence level, and document level);

[0028] Two or more spans are randomly selected for masking (preferably, the length of the sampling span is determined by a Poisson distribution λ = 3), and the masked part is replaced by [M], where M represents the span of the mask (preferably, each span contains words or sentences at the semantic level and continuous text paragraphs at the document level in the original text input), and the original sentence input replaced by the mask is used as Part A;

[0029] The selected span tokens that are covered are randomly sorted, and S represents the start of each span in the covered part, which is Part B.

[0030] Add the corresponding 2D positional codes to the two separate parts and use them as model input;

[0031] The word prediction results at the mask position in the input vector are obtained through training and prediction using the Attention mechanism;

[0032] The loss value is calculated based on the word prediction results and the true label words, and the pre-trained language model is trained based on the loss value.

[0033] Furthermore, the specific operational process for fine-tuning the large language model in the geological field based on P-Tuning v2 is as follows:

[0034] First, set the dataset path and change prompt_column and response_column to the keys corresponding to the input and output text in the file;

[0035] And / or,

[0036] Increase max_source_length and max_target_length to match the maximum input / output length in the dataset;

[0037] And / or,

[0038] Dynamically adjust the soft prompt length `pre_seq_len` and the learning rate parameter `learning_rate`.

[0039] And / or,

[0040] Dynamically adjust quantization_bit to select the quantization level of the original model.

[0041] Another object of the present invention is to provide a geological text generation system.

[0042] A geological text generation system, comprising:

[0043] The first module is used to collect geological text data and extract the stratigraphic and lithological descriptions.

[0044] The second module performs word segmentation on the stratigraphic lithology description portion to form a geotechnical description dataset of stratigraphic lithology.

[0045] The third module is used to remove duplicates, clean the geotechnical description dataset, and perform serialization processing to obtain an optimized dataset.

[0046] The fourth module is used to pre-train the language model ChatGLM with the optimized dataset, incorporate knowledge from the geological and geotechnical fields, and obtain a large language model for the geological field.

[0047] The fifth module is used to fine-tune the large language model of the geological domain based on P-Tuning v2 to obtain a lightweight large language model of the geological domain.

[0048] The sixth module is used to input the original soil and rock description data into the lightweight geological domain's large language model to generate the expected geological text.

[0049] A computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the geological text generation method based on a large language model as described above.

[0050] A computer device includes a processor and a memory storing a computer program, characterized in that the processor is configured to execute the geological text generation method based on a large language model as described above when running the computer program.

[0051] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0052] 1. This invention provides a geological text generation method based on a large language model, including steps such as text data collection, dataset construction, data preprocessing, large model construction, and model fine-tuning. It achieves intelligent text generation of stratigraphic and lithological descriptions based on a deep learning large language model. This invention trains the ChatGLM large language model and fine-tunes it using P-Tuningv2 technology, while incorporating knowledge from the geotechnical field. This enables the model to understand and apply professional geological and lithological knowledge, thereby intelligently generating stratigraphic and lithological description text that meets the requirements of engineering geological segmentation instructions based on known lithological description information. The text exhibits good continuity and high logical consistency. The entire method is simple to operate, effectively improving the efficiency of manual data processing and analysis, and is easy to promote and apply.

[0053] 2. The geological text generation method based on a large language model provided by this invention uses geological text data collected from a database and geological segmentation instructions written by geological experts in accordance with standards to obtain relatively accurate geological text data, which contains professional knowledge in the geological field, and provides accurate assistance for geological text knowledge processing and intelligent text generation.

[0054] 3. This invention provides a geological text generation method based on a large language model. The method retrieves and matches the obtained geotechnical description dataset of stratigraphy and lithology to the database, and then manually filters and annotates it to obtain the geotechnical description dataset. This data processing provides an accuracy guarantee for subsequent geological text generation. Attached Figure Description

[0055] Figure 1 This is an example flowchart of the geological text generation method based on a large language model provided by the present invention.

[0056] Figure 2 This is an example structural diagram of the geological text generation system of the present invention. Detailed Implementation

[0057] The present invention will now be described in detail with reference to the accompanying drawings.

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] Example 1

[0060] like Figure 1 As shown, a geological text generation method based on a large language model includes the following steps:

[0061] S1. Collect geological text data and extract the stratigraphic and lithological descriptions; specifically, collect geological text data and databases, and extract the stratigraphic and lithological descriptions from them.

[0062] S2. The content of the stratigraphic lithology description is segmented into words to form a geotechnical description dataset of stratigraphic lithology.

[0063] The extracted content is segmented into words, specifically using the Word2Vec model, which breaks the text down into individual words or phrases. Finally, a knowledge base for stratigraphic lithology description is constructed based on deep learning large language model technology. This knowledge base can include common descriptions, names, and characteristics of stratigraphic lithology.

[0064] In some embodiments, the method further includes retrieving and matching the database based on the stratigraphic lithology description knowledge base. Data that passes the model matching can be constructed into a geotechnical description dataset, while data that fails the matching needs to be manually screened and processed. The main process involves manually selecting some data for annotation before adding them to the geotechnical description dataset.

[0065] S3. Data preprocessing: Remove duplicate data and perform serialization processing on the data to obtain an optimized dataset.

[0066] Furthermore, the specific implementation method of step S3 is as follows:

[0067] Compare all segmented reports to identify segments with duplicate content due to proximity. Then, make a comprehensive judgment based on factors such as distance, time, and content, and finally retain the data from one segmented report while deleting the remaining duplicate reports.

[0068] Preferably, all segmented reports are compared to identify segments with duplicate content due to the proximity of the predicted mileage segments. Then, based on factors such as distance, completeness of content fields, drilling time, and content, a comprehensive judgment is made on the order of priority and the breadth of the predicted mileage segments. Finally, the data from one segmented report with a wider mileage, more keywords in stratigraphic lithology description, and more recent drilling time is retained, while the remaining duplicate reports are deleted.

[0069] For the processed database, each segmented description is first processed, and the maximum and minimum depth values ​​are selected for concatenation; then the concatenated segmented descriptions are serialized to a format suitable for model input; finally, the descriptions in the engineering geological specification are used as reference text for model training.

[0070] Taking a dataset of fine sand as an example: Original data #<3-4> Fine sand: yellowish-brown, slightly damp, non-sticky, feels gritty to the touch, loose structure. <3-4> Fine sand: yellow, non-sticky, feels gritty to the touch, loose. <3-4> Fine sand: yellow, damp, non-sticky, feels gritty to the touch, loose structure, exhibits a shaking response. *Maximum thickness #10.6*Maximum thickness #2.0*Average thickness #nan.

[0071] S4. The optimized dataset is used to pre-train the language model ChatGLM, incorporating knowledge from the geological and geotechnical fields to obtain a large language model for the geological field.

[0072] The collected stratigraphic and lithological description data are treated as small sample data, which are then proportionally divided into training and test sets. Based on the autoregressive space-filling approach, the dataset is input into the ChatGLM large-scale language model for learning and training. Simultaneously, relevant knowledge of stratigraphic and lithological description is integrated into the model to standardize its output. This combination of geological knowledge and a large language model better accomplishes downstream text generation tasks.

[0073] That is, given a piece of original text input, the model predicts possible words or phrases based on the trained neural network and completes the generation of the entire text; since the data volume is a small sample dataset and the large number of parameters of the large language model, this invention further fine-tunes the model based on P-Tuning v2; the language model ChatGLM is fine-tuned using the constructed dataset to remove unnecessary redundant parameters and reduce the total number of parameters in the model fine-tuning.

[0074] Furthermore, the specific implementation method of step S4 is as follows:

[0075] S41. Download the necessary dependencies for the model and deploy the ChatGLM-6B model locally via API and Web modes. (ChatGLM-6B is an open-source Chinese-English bilingual dialogue language model that improves the quality of generated responses through supervised fine-tuning, feedback self-help, and human feedback reinforcement learning techniques, making it perform exceptionally well in Chinese question-answering tasks. The GLM model uses a unified pre-trained model, suitable for various language generation and understanding tasks. Based on the autoregressive blank-filling idea, it processes the input text through a single Transformer encoder, randomly deleting continuous labeled spans and reconstructing these spans sequentially, thus performing well on both conditional and unconditional generation tasks.)

[0076] S42. Access geological reports and related databases, prepare engineering geological stratigraphic lithology description training and test sets, generate a JSON file containing training corpus, and save it to a local path.

[0077] S43. ChatGLM, a pre-trained language model based on the autoregressive space-filling concept:

[0078] Based on the dataset, the preset prompt templates, and the preset mapping relationship between tags and natural words, a given original text input is fed into the pre-trained language model; at the same time, the constructed geotechnical knowledge base is used as an external knowledge base to integrate geotechnical knowledge into the large language model.

[0079] The preset template for the input data uses a text input of 6 terms as an example. An original text input is: [x1, x2, x3, x4, x5, x6]. For example, the lithological description of soft soil strata is as follows: Term 1: Brownish-gray, yellowish-brown; Term 2: Soft plastic, relatively homogeneous soil texture, strong cohesion; Term 3: Thickness 1–6 m; Term 4: Layered distribution below the surface soft soil layer in trenches; Term 5: Class II ordinary soil; Term 6: Class D fill material.

[0080] Two spans [x3] and [x5, x6] are randomly selected for masking. The masked parts are replaced with [M], where M represents the span of the mask. The original input with the masked parts replaced by the mask is used as Part A: [x1, x2, M, x4, M]. The selected span tokens are randomly sorted, and S is used to represent the start of each span of the masked part, which is used as Part B: [S, x5, x6, S, x3]. The two separate parts are added with the corresponding 2D position codes as the model input: [x1, x2, M, x4, M, S, x5, x6, S, x3].

[0081] The word prediction results at the mask position in the input vector are obtained through training and prediction using the Attention mechanism. E is used to represent the end of a Span prediction result: [x5, x6, E, x3, E].

[0082] The loss value is calculated based on the word prediction results and the true labeled words. The language model is then trained based on the loss value. The model, based on the trained neural network and contextual information, predicts possible words or phrases, thereby completing the generation of the entire text.

[0083] S5. Fine-tune the large language model of the geological domain based on P-Tuning v2 to obtain a lightweight large language model of the geological domain.

[0084] P-Tuning v2 fine-tuning: Set the path to the local JSON format dataset, and change prompt_column and response_column to the keys corresponding to the input and output text in the JSON file; increase max_source_length and max_target_length to match the maximum input and output length in the dataset; dynamically adjust parameters such as soft prompt length pre_seq_len and learning rate; the P-Tuning-v2 method freezes all model parameters and dynamically adjusts quantization_bit to select the quantization level of the original model, choosing the quantization level according to the current hardware conditions.

[0085] P-Tuning v2 optimizes deep cue tuning by applying continuous cueing to each layer of the pre-trained model, resulting in better generation and knowledge exploration performance. The principle behind P-Tuning v2 is to adaptively prune a large language model, removing redundant parameters and employing a special compression method to generate a smaller and more efficient lightweight model, significantly reducing the number of parameters required for model fine-tuning.

[0086] The hyperparameters trained using the P-Tuning v2 technique in this invention are shown in Table 1. When testing the model, the checkpoint with the minimum loss on the validation set is selected.

[0087] Table 1

[0088]

[0089] Parameter Training: The model with fine-tuned parameters as described in S44 is trained and evaluated for inference. A JSON file is used to fine-tune the parameters and retrain the model. After execution, an evaluation file is generated, with evaluation metrics including Chinese Rouge score and BLEU-4. In the generated TXT file, labels are the predicted output from dev.json, and predict is the result generated by ChatGLM-6B. The predicted output and generated result are compared, and the model's performance is evaluated in conjunction with expert opinions. Parameter optimization and model training are continuously performed.

[0090] S6. Redeploy the P-Tuning v2-based automatic generation model for stratigraphic and lithological description texts, fine-tuned from the ChatGLM-6B large language model.

[0091] The optimized model, after fine-tuning, is used to generate text from the dataset, outputting segmented descriptions in the geological field. It also displays the specific values ​​of the BLEU-4 and Rouge indices, which can be used to evaluate the degree to which the generated text conforms to the stratigraphic and lithological description rules.

[0092] Specifically, the optimal model obtained after fine-tuning is used to generate a geological lithological description: The predicted output obtained by training the input dataset of fine sand with a large language model is as follows: Fine sand (Q4d+p1): yellowish-brown, slightly moist, non-sticky, feels gritty to the touch, loose structure, thickness 2-10m, locally up to 10m. Distributed in surface water ditches, it belongs to Class 1 ordinary soil and Group E filler.

[0093] like Figure 2 As shown, this embodiment also provides a geological text generation system 100, including...

[0094] Module 101 is used to collect geological text data and extract the stratigraphic and lithological descriptions.

[0095] The second module 102 performs word segmentation on the stratigraphic lithology description portion to form a geotechnical description dataset of stratigraphic lithology.

[0096] The third module 103 is used to deduplicate and clean the geotechnical description dataset and perform serialization processing to obtain an optimized dataset.

[0097] The fourth module 103 is used to pre-train the language model ChatGLM with the optimized dataset, incorporate knowledge from the geological and geotechnical fields, and obtain a large language model for the geological field.

[0098] The fifth module 103 is used to fine-tune the large language model of the geological domain based on P-Tuning v2 to obtain a lightweight large language model of the geological domain.

[0099] The sixth module 106 is used to input the original soil and rock description data into the large language model of the lightweight geological field to generate the expected geological text.

[0100] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the geological text generation method based on a large language model as described above.

[0101] This embodiment also provides a computer device, including: a processor and a memory storing a computer program, the processor being configured to execute the geological text generation method based on a large language model as described above when running the computer program.

[0102] This invention provides a geological text generation method based on a large language model, including steps such as text data collection, dataset construction, data preprocessing, large model construction, and model fine-tuning. It achieves intelligent text generation of stratigraphic and lithological descriptions based on a deep learning large language model. This invention trains the ChatGLM large language model and fine-tunes it using P-Tuning v2 technology, while incorporating knowledge from the geotechnical field. This enables the model to understand and apply professional geological and lithological knowledge, thereby intelligently generating stratigraphic and lithological description text that meets the requirements of engineering geological segmentation specifications based on known lithological description information. The text exhibits good continuity and high logical consistency. The entire method is simple to operate, significantly improving the efficiency of manual data processing and analysis, and is easy to promote and apply.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating geological text based on a large language model, characterized in that, Includes the following steps: S1. Collect geological text data and extract the stratigraphic and lithological descriptions. S2. The content of the stratigraphic lithology description is segmented into words to form a geotechnical description dataset of stratigraphic lithology; it also includes searching and matching the database based on the geotechnical description dataset of stratigraphic lithology; data that does not pass the matching are manually screened and labeled, and then added to the geotechnical description dataset; S3. The geotechnical description dataset is deduplicated, cleaned, and serialized to obtain an optimized dataset. The deduplication process is as follows: all segment reports in the geotechnical description dataset are compared to find segments with duplicate content due to proximity. Then, a comprehensive judgment is made based on distance, time, and content factors. Finally, the data in one segment report is retained, and the remaining duplicate reports are deleted. The paradigm of the segment report is the predicted mileage segment. S4. Use the optimized dataset to pre-train the language model ChatGLM, incorporate knowledge from the geological and geotechnical fields, and obtain a large language model for the geological field. The specific pre-training process is as follows: S41, Divide the optimized dataset into a training set and a test set according to a ratio of 70%~80%:30%~20%, generate a JSON file storing the training corpus, and save it; S42. Input the text with training corpus into the ChatGLM large-scale language model for learning and training, and normalize the model output; the learning and training process is as follows: The default template for the original text input data is a text of n words, where n is a natural number greater than or equal to 3; Two spans are randomly selected for masking. The masked part is replaced with [M], where M represents the span of the mask. The original input with the masked part replaced by the mask is used as Part A. The selected span tokens that are covered are randomly sorted, and S represents the start of each span in the covered part, which is then used as Part B. Add the corresponding 2D positional codes to the two separate parts and use them as model input; The word prediction results at the mask position in the input vector are obtained through training and prediction using the Attention mechanism; Calculate the loss value based on the word prediction results and the true label words, and train the pre-trained language model based on the loss value; S5, based on P Tuningv2 fine-tunes the large language model in the geological field to obtain a lightweight large language model in the geological field. S6. Input the original soil and rock description data into the lightweight geological domain large language model to generate the expected geological text.

2. The geological text generation method based on a large language model according to claim 1, characterized in that, In S1, the collected geological textual data includes at least one of the project geological description reports and scientific research papers, as well as geological textual data collected in the database.

3. The geological text generation method based on a large language model according to claim 1, characterized in that, The specific deduplication operation in S3 is as follows: compare all segment reports in the geotechnical description dataset, find the segments with duplicate content due to the close proximity of the predicted mileage segments, filter them in turn according to the completeness of the fields contained in the content, the order of drilling time, and the breadth of the predicted mileage segments, and finally retain the data from the segment report with the widest mileage, the most keywords in the stratigraphic lithology description, and the most recent drilling time, and delete the remaining duplicate reports.

4. The geological text generation method based on a large language model according to claim 1, characterized in that, In S42, during the learning and training process, the length of the sampling span is determined by a Poisson distribution λ=3. Each span contains words or sentences at the semantic level and continuous text paragraphs at the document level from the original text input.

5. According to claim 1 4. The geological text generation method based on a large language model as described in any one of the above methods is characterized in that, Based on P The specific process of fine-tuning the large language model in the geological field using Tuningv2: Set the dataset path and change prompt_column and response_column to the keys corresponding to the input and output text in the file; And / or, Increase max_source_length and max_target_length to match the maximum input and output lengths in the dataset; And / or, Dynamically adjust the softprompt length `pre_seq_len` and the learning rate parameter `learning_rate`. And / or, Dynamically adjust quantization_bit to select the quantization level of the original model.

6. A method employing claim 1 5. The system of any one of the geological text generation methods based on large language models The system is characterized by, include: The first module is used to collect geological text data and extract the stratigraphic and lithological descriptions. The second module performs word segmentation on the stratigraphic lithology description portion to form a geotechnical description dataset of stratigraphic lithology. The third module is used to remove duplicates, clean the geotechnical description dataset, and perform serialization processing to obtain an optimized dataset. The fourth module is used to pre-train the language model ChatGLM with the optimized dataset, incorporate knowledge from the geological and geotechnical fields, and obtain a large language model for the geological field. The fifth module is used for P-based... Tuningv2 fine-tunes the large language model in the geological field to obtain a lightweight large language model in the geological field. The sixth module is used to input the original soil and rock description data into the lightweight geological domain's large language model to generate the expected geological text.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements as described in claim 1.

5. Any of the geological text generation methods based on large language models described in any one of the above.

8. A computer device, comprising: A processor and a memory storing a computer program, characterized in that the processor is configured to perform, when running the computer program, as claimed in claim 1.

5. Any of the geological text generation methods based on large language models described in any one of the above.

Images