8 results about "Hash list" patented technology

The invention provides a key value cache scheduling method and system for language model inference, and is applied to the technical field of video memory management and key value cache multiplexing of a large-scale language model inference engine. The method comprises the following steps: acquiring an input sequence, extracting a candidate multiplexing fragment, calculating a fragment hash value of the candidate multiplexing fragment, querying a corresponding historical logic block hash list in a fragment index table, and generating a current logic block hash list for the fragment in a pre-filling process, mapping the Hash of the current logic block to the physical address of the historical physical key value cache block through a prefix Hash table to realize logic Hash alias, and scheduling key value cache data in an underlying video memory to participate in autoregressive decoding according to the logic Hash alias; by means of the scheme, prefix tree matching limitation can be broken through, zero-copy cross-context multiplexing of semantic fragments at any position is achieved, pre-filling calculation redundancy and first word delay are remarkably reduced, and GPU video memory utilization efficiency and reasoning throughput under a multi-agent concurrent scene are improved.

The invention belongs to the technical field of geological exploration, and discloses a fast access method and device for super-large first arrival data, electronic equipment and a medium, and the method comprises a file adaptive segmentation process, a multi-core multi-thread reading process, a dual hash list establishment process and a data access process. Reading of first arrival data is completed in a mode of combining file self-adaptive segmentation, multi-thread reading and writing and a hash list; according to the data access method, the index corresponding to the keyword is positioned through the keyword, so that the first-arrival information of the keyword is quickly accessed and read to obtain the first-arrival data, and the quick access of the super-large first-arrival data is realized. According to the method, the problems of low speed, low retrieval efficiency, memory overflow, software jamming and the like in conventional software access are solved, and the access and quality control efficiency of the super-large first arrival data is improved. The method is suitable for quick access of the super-large first arrival data.

The invention relates to the technical field of virtualization frameworks, in particular to a customized component sharing method and electronic equipment. According to the scheme, the method comprises the steps that a virtual machine is started in an Android system, and a small Android system is loaded in the virtual machine; calculating a hash value of each file of the customized component to generate a hash list, and performing digital signature operation on the hash list; starting a server application in the Android system, and enabling the server application to monitor a connection request from the small Android system; starting an authentication file system in the small Android system, sending a connection request to a server application of the Android system through the authentication file system to establish connection, and after the connection is successful, obtaining the hash list and each file of the customized component through the server application; and verifying the digital signature of the obtained Hash list through an authentication file system, and installing the Hash list and the customized component in the small Android system after the verification is passed. According to the method, the complexity and the cost of system maintenance are remarkably reduced.

Disclosed are a consensus method and device, and a blockchain system. A primary node directly broadcasts a transaction hash list including a hash value of at least one unverified transaction in the system, so that a backup node and the primary node simultaneously verify the at least one unverified transaction, and the backup node stores a first verification result obtained by verification and a first hash value of the at least one unverified transaction. The primary node broadcasts an obtained second verification result and a second hash value of the at least one unverified transaction in a pre-preparation message after completing the verification, so that the backup node searches for the first verification result based on the second hash value, and broadcast a preparation message when the second verification result is the same as the searched first verification result, to enter a subsequent consensus stage, and complete the consensus processing.

The application provides a dangerous source information management method, and belongs to the field of safety management. The method sets a global dangerous source HASH list, and saves and manages dangerous source information based on a published state machine and a block chain; each publication generates a state of a target dangerous source, the data of the target dangerous source in the state is saved into a current block, then the HASH of the target dangerous source is recalculated and the global dangerous source HASH list is updated, and the state conversion of the target dangerous source is completed; each publication is recorded in the current block, and when the number of publications exceeds a threshold value, a new block is created. The application can truly realize a user anonymous mechanism, while ensuring that an anonymous user can track the dangerous source state filled by the user in real time, and ensuring a full-cycle and tamper-proof risk evaluation record. In addition, when a dangerous source under continuous monitoring reinitiates dangerous source identification, the problem of data conflict caused by dangerous source version upgrade is solved.

The invention discloses a physical question solution visual generation method and system based on natural language processing, and aims to solve the problems of abstraction and poor interactivity of traditional physical learning. The method comprises the following steps: 1, performing multi-modal analysis on a topic text / formula / screenshot, verifying and correcting an entity position by REL, and extracting physical elements by NLP; 2, constructing a parameterized model through field mapping, attribute standardization and constraint management; 3, performing scene difference processing on multiple question surfaces, and performing physical engine and time step combined stable simulation; 4, rendering a dynamic scene and supporting parameter adjustment; and 5, generating an offline package and a hash list. The system comprises a multi-modal analysis module, a scene modeling module, a physical engine module and the like. According to the method, abstract questions are converted into interactive scenes, the understanding threshold is reduced, the learning initiative and the question solving accuracy are improved, teaching and autonomous learning scenes are adapted, and the computing power deployment cost is reduced.

This invention discloses a method for pairing quantum keys and quantum ciphertext data packets. A first terminal assembles identification information into the front end of each quantum ciphertext data packet, and the front end of the quantum key corresponding to all quantum ciphertext data packets is assembled with the same identification information. A second terminal establishes a hash list, storing the assembled quantum ciphertext data packets and quantum keys in the hash list using the identification information, and then performs pairing and decryption. This invention improves the efficiency of pairing and decryption by adding an entry code (RID) and a message sequence number (MsgID) before transmitting the quantum ciphertext data packets and quantum keys, and then classifying and storing them in the hash list according to the message sequence number (MsgID). By finding quantum ciphertext data packets and quantum keys with the same entry code (RID) and message sequence number (MsgID) at the first terminal, pairing and quantum decryption are performed, saving a significant amount of reading time.

This invention relates to the field of artificial intelligence technology and can be applied to business system platforms in fields such as healthcare and fintech. It discloses a data anomaly identification method, apparatus, computer equipment, and storage medium. The method involves acquiring data to be identified input from a client, calculating the hash value of each data unit in the data, and generating a hash list of the data based on these hash values. A Merkle tree of the data is constructed based on the hash list, and the Merkle root is calculated from the Merkle tree and written to a blockchain. A first anomaly identification result is generated based on an on-chain anomaly identification strategy. A second anomaly identification result is generated based on a pre-trained data anomaly identification model. The first and second anomaly identification results are then fused to generate a comprehensive anomaly identification result, which is returned to the client. This enables efficient and accurate data anomaly identification.