{"author":"uuuvn","author_email":"83587632+uuuvn@users.noreply.github.com","author_time":1710958801,"commit_time":1710958801,"committer":"GitHub","committer_email":"noreply@github.com","hash":"c5bf9e4c96cd1fd1ca5191d74a461a70888f1d70","message":"Ring allreduce in multitensor (#3000)\n\n* Ring allreduce v3\r\n\r\n* Configurable size, number of gpus and jit in benchmark\r\n\r\n* ScheduleBarrier v0\r\n\r\n* GB/s that make sense\r\n\r\n* ScheduleBarrier v0.1\r\n\r\n* Fallback on 2 GPUs\r\n\r\n* ScheduleBarrier v0.2\r\n\r\n* ScheduleBarrier v0.3\r\n\r\n* ScheduleBarrier v0.3.1\r\n\r\n* ScheduleBarrier v0.3.2\r\n\r\n* Replace ScheduleBarrier with automatic optimization\r\n\r\n* unused import\r\n\r\n* fix comment\r\n\r\n* typing\r\n\r\n* better fallback\r\n\r\n* python 3.8\r\n\r\n---------\r\n\r\nCo-authored-by: George Hotz <72895+geohot@users.noreply.github.com>\r\nCo-authored-by: chenyu <chenyu@fastmail.com>\r\nCo-authored-by: nimlgen <138685161+nimlgen@users.noreply.github.com>","parents":["455f7bea9b675df49ef8d158a98b68e5eb0e9fad"],"tree_hash":"2ff36f22b76d868ee766bb8bfda50d97d9a432b7"}