21 results about "Basic language" patented technology

The invention discloses a robot target navigation method, device and equipment and a medium. The method comprises the following steps: constructing a basic language instruction through a preset language model according to a fuzzy instruction input by a user and a preset family knowledge graph; performing information expansion through the preset language model according to the basic language instruction and preset reference information, and generating a target language instruction; generating a target navigation instruction from the target language instruction and a collected environment image through a preset visual language model; and controlling the robot to move according to the target navigation instruction. The method can be applied to robot navigation scenes in the business fields of medical treatment, health, old-age care and the like. By generating the target navigation instruction to control the robot to move, the problem that an indoor auxiliary robot with a visual language model cannot efficiently provide services for old people can be solved.

The invention discloses a multi-dimensional engineering paper abstract evaluation method, and belongs to the field of natural language processing. The implementation method comprises the following steps of: performing text correctness evaluation on a basic language specification in an abstract text, wherein a text correctness evaluation dimension comprises two sub-dimensions of grammar correctness and an expression specification degree; a pre-training language model GPT-2 is used as an evaluation base model, training fine tuning is carried out by using an abstract text in an RAAMove training set, and fluency dimension in abstract language paragraph fluency is evaluated; introducing a speech step theory, and constructing a speech step classification model based on comparative learning; performing double representation learning on sentence feature representation and speech step tag feature representation by adopting a supervised comparative learning method; performing coherence evaluation in paragraph fluency according to a speech step transfer similarity index; a ROUGE 1F1 value is calculated, and semantic similarity evaluation is carried out; and introducing a weighted summation strategy to carry out weighted summation on the scores of the dimensions to obtain a comprehensive evaluation score of the abstract text, namely realizing comprehensive evaluation on the abstract text.

The invention belongs to the field of operation inspection intelligent models, and discloses a power equipment operation inspection intelligent agent construction method and a related system.The power equipment operation inspection intelligent agent construction method comprises the steps that a power equipment related professional word list is introduced when a corpus is constructed, and the power equipment related professional word list is combined with an original word list; the understanding accuracy of the model on electric power terminologies, equipment operation scenes and details in an operation and inspection process can be obviously improved; in the multi-stage progressive pre-training and subsequent fine tuning process, the pre-training model continuously learns in the corpus which is richer and more accordant with the power operation inspection requirement, and professional knowledge and general language understanding ability supplement each other. By adopting multi-stage progressive pre-training, the method can progressively transition from basic language ability to knowledge mastering of a power equipment scene, and the logical reasoning ability of the model in a complex scene is gradually enhanced.

The invention discloses an English listening and speaking interaction intelligent training system based on a large language model, and relates to the technical field of English learning, and the training system comprises a user capability graph module, an interaction feedback module, a personalized learning path generation module, a scene simulation module, an interaction rhythm control module and a community cooperative training module. According to the method, the pragmatic ability, the cross-language communication ability and the linguistic ability are jointly incorporated into the training content, and the adaptive training content is generated in combination with a real cross-language communication scene, so that a user synchronously masters expression specifications and culture adaptation key points in different scenes in the training process; the effect that training content and real communication requirements are deeply matched is achieved, the user is helped to accurately adapt to a native English scene and real phrase habits of people, the requirements of an actual application scene are met, and substantial improvement from basic language knowledge mastering to real communication ability is achieved.

The invention discloses a method for accelerating reasoning of a multi-modal large language model based on speculative decoding, the input of the multi-modal large language model is composed of a system lexical element, a visual lexical element, an instruction lexical element and an output lexical element, and the method comprises the following steps: S1, extracting a predetermined number of samples from a ShareGPT data set which is finely adjusted by using an LLaMA model instruction, pre-training an Eagle small model, and obtaining a pre-trained Eagle small model; enabling the Eagle small model to establish a basic language generation capability in a plain text environment; s2, extracting a predetermined number of samples in a ShareGPT data set finely adjusted by an LLaVA model instruction, and migrating the capability of an Eagle small model by using a transition mode; s3, after the draft model is subjected to two-stage training, on the basis of the improved draft mechanism design, differential processing is carried out according to the characteristic difference between the text and the visual mode; according to the method, the performance of speculative decoding in a multi-modal task can be improved, the acceptance length of each verification of the target model is improved, and respective adaptation of the text lexical elements and the visual lexical elements in the draft model is realized.

The application provides a marketing copy automatic generation method and device based on reinforcement learning and a storage medium. The method comprises the following steps: performing semantic matching retrieval on public copy data to obtain a candidate copy; inputting a slot rewriting instruction into a pre-generated language model to generate a first marketing copy; performing supervised fine-tuning training on a preset basic language model to obtain a first training model; inputting new user product information and promotion requirements into the first training model to generate a second marketing copy, scoring the second marketing copy and generating evaluation data; constructing a partial order training sample according to the evaluation data, taking the partial order training sample as a reward signal, performing reinforcement learning training on the first training model to obtain a second training model; and calling the second training model in a copy generation system and outputting a target marketing copy based on user product information and promotion requirements. The application can realize batch generation of marketing copies with high compliance and consistent constraints.

The invention discloses an equipment entity identification method and system based on thinking chain enhancement and instruction optimization, and relates to the technical field of entity identification methods. The equipment entity recognition method based on thinking chain enhancement and instruction optimization comprises the steps of enhancing thinking chain data, constructing a cue word template containing reasoning requirements, processing an original production equipment text by utilizing a large-scale language model, generating enhanced training data with reasoning explanation, and explaining instance judgment reasons and category attribution. High-quality training data screening and model instruction fine tuning are carried out, a low-rank adaptive equal-parameter efficient fine tuning technology is adopted, a high-quality subset is used for fine tuning of a basic language model, and a domain adaptive model is obtained. According to the method, the injection logical reasoning is enhanced through the thinking chain, the training data quality is optimized through data screening, precise field adaptation is realized in combination with efficient fine tuning, and the accuracy, interpretability and generalization ability of equipment entity recognition are improved.

The invention discloses a large language model training method, device and equipment applied to a professional field and a storage medium, and the method comprises the steps: obtaining a related document of a current field, and carrying out the chapter splitting of the related document, and obtaining a plurality of text blocks; generating a plurality of structured question and answer pairs based on each text block, wherein each structured question and answer pair comprises questions and answers corresponding to the document chapters to which each text block belongs; and inputting each text block and each structured question and answer pair into a basic language model, and performing iterative training on the basic language model according to a preset progressive fine tuning strategy to obtain a target large language model. The domain document can be subdivided according to chapters and sections, structured question and answer training is performed on each text block, and then the model is trained by simulating a human learning process through the preset progressive fine tuning strategy, so that terminologies and concepts can be integrated step by step and iteratively; therefore, the knowledge adaptability of the model in the specific field is remarkably improved.

The application provides a language model reasoning method and device based on hierarchical state memory bank and combined initialization, a terminal, a medium and a product. The method comprises the following steps: constructing a hierarchical state memory bank; obtaining user query information, and performing hierarchical retrieval in the hierarchical state memory bank based on the user query information to obtain a best state set; performing combined initialization on the best state set to generate an initial state; and enabling a basic language model to start reasoning on the user query information in the initial state to generate a corresponding reasoning result. Through hierarchical retrieval, the application can provide matched context for queries of different granularities, solve the cold start problem of the existing linear attention language model in the form of RNN, and make the reasoning result of the model more accurate.

The invention relates to the technical field of intelligent customer service, and provides a slot position information extraction method and slot position information extraction equipment, which can reduce the dependence of sample data on manual annotation and reduce the labor cost. The method comprises the steps of obtaining dialogue data between a user and a customer service system; inputting the dialogue data into a pre-constructed slot information extraction model to obtain slot information associated with the dialogue data; the slot position information extraction model is obtained by training a first basic language model, and sample data used for training the first basic language model is obtained by labeling a second basic language model; the parameter quantity of the second basic language model is greater than that of the first basic language model; and obtaining a slot information integration result based on the slot information associated with the dialogue data.

The invention relates to a code internationalization processing method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: constructing an abstract syntax tree of a to-be-processed file, and extracting a to-be-translated character string set from the to-be-processed file based on the abstract syntax tree; semantic translation keys are generated, the character strings to be translated are associated with the corresponding semantic translation keys, and a key value pair set to be translated is formed; replacing the semantic translation keys in the translation key value pair set with general internationalization function calling through a target parser to obtain internationalization transformation codes; storing the to-be-translated key value pair set into a basic language package; performing multi-language translation on the basic language package, and respectively storing translation contents after the multi-language translation into a plurality of corresponding language packages; and when code internationalization processing needs to be carried out, reading translation content in a target language package corresponding to the target language through the internationalization transformation code. By adopting the method, contents which do not need to be translated can be effectively filtered.

The invention relates to the technical field of artificial intelligence and natural language processing, and particularly discloses a Chinese and English mixed children English dialogue data set construction method and teaching system.The method comprises the steps that English learning levels are classified according to the basic language level of a target user, and a question and answer prompt template is configured for each English learning level; dialogue data is generated; inputting the dialogue data into a large language model to obtain a learning feedback result of the target user; performing reinforcement learning on the learning feedback result of the target user and the English learning level of the target user; evaluating the dialogue data and the learning feedback result according to a large language model for data quality evaluation; and storing the dialogue data of which the dialogue data evaluation result is greater than a preset threshold value to obtain a Chinese and English mixed children English dialogue data set. The data set constructed by the Chinese and English mixed children English dialogue data set construction method provided by the invention is high in quality and can meet the real requirements of children English learning.

The invention provides a large language model training method and device, equipment, a storage medium and a program product. The method comprises the steps that a training set and a basic language model are obtained, the training set comprises at least one text data sample, the basic language model is a pre-trained model, and the basic language model comprises a preset number of decoding layers; copying the source decoding layer for a preset number of times to obtain a copied decoding layer; the source decoding layer is one or more continuous decoding layers in the preset number of decoding layers, and the copy decoding layer is correspondingly the same as the source decoding layer; the copy decoding layer and the source decoding layer are stacked to obtain an expanded language model, and the expanded language model comprises the copy decoding layer and the source decoding layer; and training the extended language model based on the training set, and quitting training when a training quitting condition is met to obtain a trained large language model. According to the method, the data volume and the training duration of large language model training can be reduced.

The invention discloses a language model reasoning ability optimization method for constructing preference data based on self-generation and reasoning evaluation, and the method comprises the following steps: (1), generating a plurality of different candidate response sequences through a to-be-trained basic language model according to a given input problem; (2) performing quality evaluation on a plurality of different candidate response sequences, extracting information representing reasoning process characteristics of the candidate response sequences, and constructing vectors representing candidate response sequence characteristics; (3) on the basis of the quality evaluation result and the feature vector, constructing a preference pair consisting of an optimal selection response and an inferior selection response for the input problem; and (4) training or optimizing the basic language model by using the preference pair, so that the probability of generating the preference response by the trained model is higher than the probability of generating the inferior selection response. According to the method, external high-quality annotation data or a large teacher model is not needed, and the performance, especially the reasoning ability, of the language model in the professional field can be efficiently improved at low cost.

This invention relates to the field of artificial intelligence application technology, specifically to a power knowledge question-answering system and method based on a large language model. The system includes a question-answering knowledge base for constructing a power knowledge graph and a dual index; a query rewriting module that parses the user's original query and outputs a standardized rewritten query; a knowledge retrieval module with dual recall and ranking that outputs Top-K results; and a content generation module that generates answers and performs closed-loop optimization after reflection and evaluation. The method includes constructing a power knowledge graph and dual index, rewriting the original query based on a large language model, obtaining Top-K results through dual recall and ranking, and reflectively generating and optimizing answers. This addresses the problems of weak semantic understanding in existing power knowledge question-answering technologies, insufficient professionalism and reasoning ability of the basic language model, and lack of a systematic knowledge organization structure.

The invention provides a few-sample intention detection method, medium, equipment and product, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the pre-training of a basic language model through employing an offline learning mode, and obtaining a prediction model; the prediction model receives user query data online in an online learning mode, a candidate intention list with sorting is generated, and whether the first intention in the candidate intention list meets a preset confidence threshold value or not is judged; if yes, outputting the first intention; if not, forming a reduced candidate set by the first K candidate intentions in the candidate intention list, and obtaining a large model prediction intention by using a large language model; if the large model prediction intention is consistent with the first intention, outputting the large model prediction intention; if not, the large language model outputs a final intention according to the user query data, the first intention, the large model prediction intention and comparative context information of example query. According to the invention, accurate few-sample intention detection is realized.

The invention discloses a large language model reasoning acceleration method based on speculative decoding, and relates to the technical field of natural language processing, and the method comprises the following steps: S100, freezing a basic language model; s200, extracting a hidden state of the key layer; s300, constructing a de-wharf network set; s400, training a de-wharf network; and S500, deploying the basic language model. According to the method, a multi-solution wharf network structure is adopted, and a parallel draft and verification mechanism is combined, so that the reasoning efficiency of the basic language model is improved, the resource consumption is reduced, and meanwhile, the consistency and the accuracy of a generated result are considered.

To obtain an UI verification device that can reduce the user's effort.SOLUTION: An operation recording part 11 generates a basic test script and basic UI component images for a test target system 200 in which a basic language is set, based on a series of testing processes performed by the user. A changed language script generation and image acquisition part 13 generates a changed language test script corresponding to the basic test script, based on the basic test script and the basic UI component images, when the used language of the test target system 200 is changed from the basic language to a changed language. An operation reproduction part 12 sequentially executes the processes defined by the multiple changed part scripts included in the changed language test script as a change reproduction process.SELECTED DRAWING: Figure 1

A low-computing-power language expert system based on a hierarchical architecture belongs to the field of artificial intelligence, and is formed by longitudinally coupling an expert generation layer, a reasoning optimization layer and a domain adaptation layer: the expert generation layer generates a batch of virtual experts with heterogeneous knowledge, and the virtual experts carry exclusive domain prompts and randomly select expert subsets; the reasoning optimization layer generates a prediction probability by using an expert subset randomly selected by the expert generation layer, executes abnormal truncation and fixed voting, rejects extreme prediction by using a mean value-standard deviation threshold, and weights and fuses confidence and expert orthogonality to determine a final output mark; and the domain adaptation layer pre-trains the plug-in memory component according to the final output mark obtained by the reasoning optimization layer, inserts the pre-trained plug-in memory component, and performs probability interpolation on the pre-trained plug-in memory component and the basic language model during reasoning to complete immediate domain migration. The method solves the problem of illusion of a small language model under the condition of low computing power.

The invention discloses an e-commerce customer service reply method, system and device and a medium. The method comprises the following steps: presetting a basic language model, a first language model and a second language model; inputting the first historical customer service reply context into a basic language model to obtain first data output by the basic language model; based on the first language model, performing distillation learning according to the basic language model, and performing low-rank adaptive fine tuning by using the first data to obtain a third language model; inputting the second historical customer service reply context into a third language model to obtain second data output by the third language model; based on the second language model, performing low-rank adaptive fine tuning by using the second data to obtain a target language model; the reply request is input into the third language model and the target language model, reply content corresponding to the reply request is obtained, and reply accuracy, logicality and interpretability can be improved through a model three-level architecture in combination with a hybrid reasoning training strategy.

The application discloses an e-commerce customer service reply method, system, device and medium. The method comprises the following steps: presetting a basic language model, a first language model and a second language model; inputting a first historical customer service reply context into the basic language model to obtain first data output by the basic language model; performing distillation learning based on the first language model according to the basic language model, and performing low-rank adaptive fine-tuning by using the first data to obtain a third language model; inputting a second historical customer service reply context into the third language model to obtain second data output by the third language model; performing low-rank adaptive fine-tuning by using the second data based on the second language model to obtain a target language model; and inputting a reply request into the third language model and the target language model to obtain reply content corresponding to the reply request. The three-level architecture of the model and the mixed inference training strategy can be combined to improve the reply accuracy, logic and interpretability.