3 results about "Double threshold" patented technology

The application provides a memory control method and a memory storage device. The method comprises: receiving a data write request from a host system, the data write request comprising to-be-written data; determining a frequency of the data write request issued by the host system or a number of target blocks in a plurality of entity blocks; dynamically adjusting a first threshold and a second threshold for cold and hot data classification according to the frequency of the data write request issued by the host system or the number of target blocks; calculating a hotness value of the to-be-written data according to the data write request; determining a data state of the to-be-written data according to the first threshold, the second threshold and the hotness value, and writing the to-be-written data into an entity block corresponding to the data state. Through the distinction of the double thresholds and the plurality of data states, more refined cold and hot data management is realized, the frequent jumping (ping-pong effect) of data between different storage blocks due to the cold and hot data classification error is avoided, and the data migration flow is smoothed.

This invention discloses a method and system for playing back air traffic control data based on message timing. The method includes: a reading thread extracting the timestamp and hexadecimal message data of each line of the message based on a two-level timestamp extraction strategy; a decoding thread extracting the message and sorting it using a sliding time window double threshold sorting algorithm before sending it to the decoding queue; a sorting thread transmitting the message data; and a sending thread and a dynamic scheduling algorithm based on a reference time calculating the precise sending time of each message and sending it to the target system for playback. The advantages of this invention are as follows: (1) The playback timing accuracy is naturally matched with the protocol accuracy, and the time error is strictly controlled within 1 / 128 seconds; (2) It effectively solves the trajectory distortion caused by the disorder of the original data; (3) The system resource occupancy rate is low, supporting high-concurrency playback tasks; (4) The startup delay is short, meeting the fast response requirements of the playback system; (5) The processing throughput is high, adapting to high-density air traffic control data playback application scenarios.

The invention provides a large language model quaternary weight compression method and device and a storage medium. The large language model quaternary weight compression method comprises the steps that a model and data are prepared, specifically, a pre-trained large language model is selected, and a Wikitext2 data set is adopted as calibration data; executing a first-stage PTQ process, and carrying out weight transformation: activating quantization, defining a quaternary quantization function, carrying out multi-parameter grid search, and obtaining an optimal weight scaling factor and double thresholds; and executing a second-stage PTQ process, and carrying out weight optimization: defining block reconstruction loss, carrying out greedy iterative quantization, and generating a quaternary weight compression model. The technical problems of large model performance loss and low high-precision quantization compression ratio caused by insufficient information retention of a ternary quantization scheme in the prior art can be solved.