4results about How to "Avoid long-term occupation" patented technology

The application discloses a kind of based on end side's big model key value cache dynamic optimization method and system, method includes: S1, the context resource of multiple inference sequences is centrally controlled, and corresponding context space is pre-allocated;S2, whether the currently available context space satisfies the allocation demand of inference sequence is judged, if it satisfies, then the intelligent context allocation is carried out to inference sequence, if it does not satisfy, then whether the remaining memory space of end side equipment is enough is judged, according to the judgment result, step S3 or S4 is executed;S3, if space is enough, then the available context space is increased by dynamic expansion, and the context space allocation and context update of inference sequence are re-performed;S4, if space is insufficient, first carry out context recovery, then discard and retain context by sequence perception sliding window mechanism, and execute context update after allocation is completed.The application can realize the efficient inference of big model under the condition of end side limited resource.

The application discloses a large file cross-unit transmission method based on a business cooperation scene, relates to the technical field of cross-organization data transmission and resource scheduling, and takes a business context as a core to map an event type, an emergency level and an SLA deadline time into a calculable priority, and is linked with real-time network measurement results, so that scheduling no longer stays in static file attributes, but can instantly improve the service share of a key request according to events such as out-of-stock early warning and promotion start, thereby reducing the probability that a high-emergency task is blocked by regular transmission. By introducing unit-level quotas, admission and interruptible preemption on the side of a bottleneck link, long-term occupation caused by first-come-first-served is avoided, basic fairness between different units is ensured, and bandwidth can be quickly transferred to more urgent flows at a business peak moment. Weighted round-robin weights are dynamically updated according to the emergency degree, the SLA urgency degree and the quota surplus degree, hysteresis and minimum interval control are superimposed, and the agility of scheduling is improved.

This invention relates to the field of large pressure vessel installation technology, specifically disclosing a vertical pressure vessel positioning and installation device and method. The vertical pressure vessel positioning and installation device includes: an installation platform having a first through hole for the vertical pressure vessel to pass through; a positioning adjustment mechanism disposed on the installation platform, used to adjust the position and attitude of the vertical pressure vessel; and a support frame disposed above and connected to the positioning adjustment mechanism, used to support the vertical pressure vessel. The positioning adjustment mechanism includes a fixed base, a horizontal adjustment component, and a vertical adjustment component. The fixed base is disposed on the installation platform, the horizontal adjustment component is movably connected to the top wall of the fixed base, and the vertical adjustment component is disposed above the horizontal adjustment component and connected to both the support frame and the horizontal adjustment component. This invention has the advantages of high installation accuracy, high safety, and high installation efficiency.

A multi-agent context sharing method, system, and medium are disclosed, comprising: parsing the context content output by a first agent to generate corresponding metadata; associating the context content with the corresponding metadata and storing it in a distributed cache; establishing a topic set indexed by a session identifier in the distributed cache and adding the topic identifiers to the topic set; responding to a task to be executed by a second agent, obtaining the topic set corresponding to the current session identifier, reading the topic descriptions corresponding to each topic identifier in the topic set, and generating a list of topic descriptions; inputting the task to be executed and the corresponding list of topic descriptions into a large language model, receiving the topic identifier set returned by the large language model as the target topic identifiers semantically related to the task to be executed; and reading the corresponding context content from the distributed cache according to the target topic identifiers, associating the context content with the task to be executed, and inputting it into the second agent.