The invention discloses a mixed request-oriented large
language model adaptive scheduling
algorithm. The
algorithm comprises the following steps: S1, initializing queues Q1, Q2 and Q3 and a set CB to be empty; s2, when a new request arrives, calculating the priority of the new request, adding Q1 or Q2 according to the type of the new request, and then entering S3; when no new request arrives, entering S3; s3, judging whether the request is true or not, if so, entering S4, otherwise, terminating the
algorithm, and starting when a new request exists; s4, updating the priorities of the requests in the queues Q1, Q2 and Q3; s5, the updated Q1, Q2 and Q3 are traversed in sequence, and the request is selected to be added into the CB; s6, inputting the CB into the large
language model for reasoning for one time, removing the completed request from the CB, and recording the
completion time; s7, when the CB is empty, entering the step S2; when the CB is not empty, k is made to be equal to k + 1, and when k is smaller than Y-1, the step S6 is executed, and otherwise, the step S8 is executed; and S8, for uncompleted requests in the CB, the request from the Q1 is put back to the Q1, the requests from the Q2 and the Q3 are both put back to the Q3, and then the step S2 is returned.