13results about How to "Guaranteed Build Quality" patented technology

The invention provides a large language model text traceability method based on virtual cue word embedding, and belongs to the technical field of artificial intelligence natural language processing and the technical field of digital content security, the method comprises the following steps: (1) constructing a chat input instruction text with a role format; (2) obtaining embedded vector input; (3) obtaining system cue word embedding representation; (4) obtaining a generated text with a user identity information watermark; and (5) recovering user identity information to realize traceability of the generated text. According to the method, on the premise that main body parameters of the large language model are not modified and model text generation quality and response efficiency are not affected, instant embedding of traceable identity information in the large language model generation process is achieved, meanwhile, robustness of embedded watermark information in a text transformation scene is guaranteed, and the method is suitable for large language model generation. Verifiable and reproducible identity identification information can be stably extracted from the generated text, and reliable technical support is provided for source tracing of the generated text.

This invention discloses an attention acceleration method, apparatus, device, and medium based on a diffusion model. The method includes: acquiring the input feature set, current time step, and local window parameters corresponding to a target multi-head attention module in a target diffusion model; reshaping the shape of each input feature in the input feature set based on a preset tensor format to obtain a three-dimensional feature set; determining a feature recombination strategy corresponding to the three-dimensional feature set based on the current time step, and recombinating the three-dimensional feature set based on the feature recombination strategy to obtain a target feature result set corresponding to the target multi-head attention module; and performing local window attention calculation on the target feature result set based on the local window parameters and the feature recombination strategy to obtain the target attention result corresponding to the target multi-head attention module. Through the technical solution of this invention, efficient acceleration can be achieved while maintaining generation quality, improving the actual utilization rate of hardware.

This invention provides a method for tracing the source of large language model text based on virtual prompt word embedding, belonging to the fields of artificial intelligence natural language processing technology and digital content security technology. The method includes: (1) constructing chat input command text with role format; (2) obtaining embedded vector input; (3) obtaining system prompt word embedding representation; (4) obtaining generated text with user identity information watermark; (5) restoring user identity information to achieve source tracing of generated text. This invention achieves real-time embedding of traceable identity information in the large language model generation process without modifying the main parameters of the large language model or affecting the model text generation quality and response efficiency. At the same time, it ensures the robustness of the embedded watermark information in text transformation scenarios, and can stably extract verifiable and reproducible identity information from the generated text, providing reliable technical support for tracing the source of generated text.

This application provides a railway communication transmission wireless cabinet and communication base station, including: a cabinet, transmission equipment, and wireless equipment. The transmission equipment and wireless equipment are located inside the cabinet, with the transmission equipment located on the side of the wireless equipment facing away from the bottom of the cabinet. The transmission equipment includes: a transmission power distribution unit and a transmission subframe, with the transmission power distribution unit located on the side of the transmission subframe facing away from the wireless equipment. The wireless equipment includes: a remote radio frequency unit and a baseband processing unit, with the remote radio frequency unit located on the side of the baseband processing unit facing the transmission equipment. Thus, by integrating the transmission equipment and wireless equipment into a single cabinet, the installation location of the equipment is further optimized, the space utilization within the cabinet is improved, the construction period is shortened, construction quality is guaranteed, and the floor space is reduced.

The embodiment of the invention provides a content generation method and related equipment, and the method comprises the steps: indicating the execution of a calling reasoning model through business indication information related to a content generation business, when de-noising processing is carried out on corresponding input content coding characteristics of the noisy data content under a time step tj in the T time steps under the constraint of the first data content, obtaining a cache type for cache multiplexing according to content analysis parameters indicated by a cache multiplexing strategy related to the time step tj; and if the cache type is an inter-step cache type, multiplexing a predicted noise feature corresponding to a time step ti having an inter-step multiplexing relationship with the time step tj, performing denoising processing on an input content coding feature corresponding to the time step tj, and performing content generation based on an output content coding feature corresponding to the time step tj obtained by the denoising processing, the second data content is obtained. Therefore, the model reasoning speed can be effectively improved, and waste of computing resources is avoided.

This invention relates to a dynamic retrieval enhancement generation method based on critical path entropy increase trends and document selection. Addressing the issues of retrieval response lag and false triggering caused by single-point threshold triggering in existing technologies, as well as noise interference and resource waste caused by direct splicing of Top-K retrieval documents, this invention creatively proposes a critical path entropy increase trend triggering mechanism. It optimizes single-point entropy monitoring of all tokens to critical path token entropy increase trend monitoring, achieving on-demand retrieval triggering by dynamically determining uncertainty increments, thereby suppressing false triggering caused by noise fluctuations in non-critical locations and alleviating retrieval response lag. Furthermore, this invention proposes a document selection model based on direct preference optimization of DPO, optimizing the coarse-grained injection of Top-K documents into a fine-grained filtering and injection mechanism. By filtering out irrelevant and redundant documents, it reduces the invalid occupation of the context window, thereby solving the problems of reduced generation quality and wasted computational resources caused by noise interference.

This invention discloses a modular construction method for prefabricated utility tunnel structures, comprising the following steps: S1, prefabricating several segments of reinforced concrete integrated utility tunnel columns and beams; S2, excavating a foundation pit to construct a foundation platform, forming several spaced-apart sockets on the platform; S3, hoisting the integrated utility tunnel column and beam structure as a whole, positioning and fixing the bottom of the integrated structure into the sockets; S4, repeating step S3 to complete the installation of the next integrated utility tunnel column and beam structure; S5, pouring concrete at the joints between adjacent integrated utility tunnel columns and beams. This invention can further reduce construction costs, shorten the construction period, and ensure construction progress while maintaining construction efficiency and quality, and can effectively control problems such as construction waste, site dampness, and dust pollution during construction.

This application relates to the field of building construction technology, and discloses a precast cap beam assembly system, control method, and control device, including: a mounting frame, an assembly base, a detection module, and a controller assembly. The mounting frame is detachably mounted on the bridge pier; the assembly base is movably connected to the mounting frame via a lateral adjustment mechanism and a longitudinal adjustment mechanism, and is used to support the precast cap beam; the detection module is mounted on the mounting frame and is used to detect the relative position between the connecting rod and the connecting hole; the controller assembly is connected to the detection module, the lateral adjustment mechanism, and the longitudinal adjustment mechanism. In this application, the assembly system can automatically calibrate and adjust the position of the precast cap beam, reducing construction difficulty, facilitating efficient assembly of the precast cap beam, improving construction efficiency, and ensuring the overall construction quality and safety performance of the bridge.

The invention relates to the technical field of vehicle fault diagnosis, and discloses a wheel bolt looseness detection method and device, a vehicle and a medium, and the method comprises the steps: collecting an inertial measurement unit signal, a wheel speed signal and a steering wheel hand torque signal of the vehicle; determining target detection data based on the inertial measurement unit signal, the wheel speed signal and the steering wheel hand torque signal; detecting whether a wheel bolt of the vehicle is loosened by using the target detection data to obtain a detection result; and controlling the vehicle to execute corresponding actions based on the detection result. According to the method, the existing sensor of the vehicle is directly used for collecting the signals, then the signals are fused with the multi-dimensional vehicle running state signals, feature extraction and combination processing are combined to form target detection data, the target detection data are used for wheel bolt looseness detection, full-process automation from signal collection, data processing, state detection to vehicle control action execution is achieved, and the detection efficiency is improved. And the detection accuracy and the driving safety are improved, so that the vehicle fault diagnosis requirements of low cost, high real-time performance and easy deployment are met.

The present application relates to earthwork backfill engineering technical field, specifically relates to a kind of super large particle size wide gradation soil and stone mixture maximum dry density determination method, using large tonnage convex block roller compaction or dynamic compaction compaction, obtain the maximum dry density under different gradation conditions, and under different gradation conditions, different more than 20mm particle content, without technical treatment to the super-diameter particles in the original gradation of soil and stone mixture.In construction, with the measured gradation curve and more than 20mm particle content, compared with different gradation conditions, and under different gradation conditions, different more than 20mm particle content conditions, the larger of the two under the closer gradation condition and the closer gradation condition more than 20mm particle content is the maximum dry density.The maximum dry density determined by this method is closer to the actual construction condition, which can truly, objectively and accurately reflect the maximum dry density of soil and stone mixture, and ensure the construction quality of the project.

The present application relates to the cross field of artificial intelligence and computer architecture, in particular to an adaptive KV cache compression method, system and device for an audio-text multi-modal large model. The method comprises opening a full-precision hot table and a low-bit cold table in the video memory; writing the newly generated audio or text Token in the FP16 format into the hot table, and when the number of Tokens in the hot table reaches a preset threshold, gradually migrating the Tokens in the hot table to the cold table; identifying the modal type of the Token during the migration process, selecting a corresponding quantization grouping strategy according to the identified modal type, and performing INT2 compression storage to the cold table after online calculation of the quantization parameters of the Token to be migrated. Through the cooperation of the above steps, the present application realizes the improvement of the model inference throughput and the natural full-duplex human-computer interaction under the premise of ensuring the audio synthesis quality.

The application discloses an iterative chart code generation method and device based on a chart and a code error correction large model, and relates to the technical field of chart code generation. The method comprises the following steps: based on a code error classification system, using a multi-modal large model to perform data variation according to an original chart code pair, and constructing an original data set; based on the code error classification system and a line-level Levenshtein distance algorithm, adding correction prompt information according to the original data set, and constructing a training data set; based on an LLaMA-Factory fine-tuning framework and a GRPO training strategy, performing two-stage training on a code error correction large model according to the training data set, and obtaining an optimized code error correction large model; and based on the optimized code error correction large model and a chart code generation large model, performing iterative code generation according to a target reference chart, and obtaining target generated code. The application is a chart-based high-precision, interpretable and iterative chart code generation method.

The application relates to the technical field of battery early warning and discloses a battery alarm scene data enhancement and algorithm robustness verification method, which acquires a real alarm sample set; a pre-trained cascade generation model is used to generate a synthetic alarm sample; the synthetic alarm sample and the real alarm sample are mixed in proportion to construct an enhanced training set; a battery alarm related target recognition algorithm is trained using the enhanced training set, and the robustness improvement degree of the battery alarm related target recognition algorithm is quantitatively verified under a plurality of preset disturbance conditions; wherein the cascade generation model comprises the following steps: a denoising diffusion probability model is used to generate a basic sample, a generative adversarial network is used to perform detail enhancement and distribution correction on the generated result, a preset strategy is used to control a random latent variable to complete sample screening, and a high-quality and diversified alarm data set is constructed. Through the method, the robustness and stability of the alarm recognition algorithm can be significantly improved, and reliable prediction and accurate early warning of a battery system under actual complex working conditions can be realized.