3results about How to "Avoid obsolete" patented technology

The utility model discloses a forming device for expansive cement expansion rate test, which belongs to the technical field of expansion rate test, and comprises a base, a pair of detachable end plates are arranged on the base, a plurality of mutually parallel partition plates are arranged between the pair of end plates, small holes are respectively arranged in the areas of the end plates between the adjacent partition plates, and the small holes are communicated with the end plates. A small hole is formed in the end plate and used for being connected with a nail head, a limiting protrusion is arranged in the small hole, a limiting groove matched with the limiting protrusion is formed in the nail head, and the end plate is detachable in the limiting direction of the limiting protrusion. According to the forming device for the expansive cement expansion rate test, a limiting protrusion is arranged in a small hole, a limiting groove matched with the limiting protrusion is formed in a nail head, an end plate is detachable in the limiting direction of the limiting protrusion, the limiting protrusion is inserted into the limiting groove, and therefore the nail head can be limited; the nail head is prevented from rotating or moving in the small hole, so that the nail head is prevented from falling off or inclining from the small hole when the test body vibrates.

This invention provides a video download method, apparatus, electronic device, and readable storage medium. The method includes: in response to detecting that a first video downloaded by a client is interrupted due to transcoding on the server side, obtaining the download progress of the first video already downloaded by the client; the download progress is used to indicate the amount of data downloaded and is associated with the identification information of the last downloaded video frame; based on the download progress, determining a set of segmented videos to be downloaded from a second video generated after transcoding the first video, and downloading the set of segmented videos; matching the last downloaded video frame of the first video with the video frame content of the second video to determine the mapping time point corresponding to the last video frame in the second video; in response to playing to the last video frame of the first video, resuming playback of the second video from the mapping time point.

This invention relates to the field of artificial intelligence technology, and more particularly to a multimodal large-model inference error correction method based on multi-agent collaborative self-feedback. This method maintains an error correction window during streaming inference, dividing the inference output into a frozen region, an error-correctable region, and a region being generated, using semantic units as the basic management unit. Upon detecting an error, the method calculates the attention dependency of each semantic unit on the error location, classifying subsequent inference segments into those affected and those unaffected. For affected segments, an error correction agent is invoked for incremental correction; for unaffected segments, a lightweight verification agent is invoked for consistency verification. Verified segments retain their original output and corresponding key-value cache. This invention avoids the waste of computational resources caused by full regeneration, significantly reducing the computational overhead of error correction while ensuring inference quality.