{"author":"Maksym Sobolyev","author_email":"sobomax@sippysoft.com","author_time":1702839866,"commit_time":1702839866,"committer":"GitHub","committer_email":"noreply@github.com","hash":"887f3d99334884e17be7056000ff85e9d4de7ee5","message":"Make torch backend more usable, fix bfloat support in the llvm backend (#2765)\n\n* Uncripple dtype tests, TestBFloat16DType never actually runs.\r\n\r\n* Fix conversion from/to bfloat16.\r\n\r\nCall cast() recursively, so that it works for any type combo.\r\n\r\n* Run this test on torch backend as well.\r\n\r\n* Add torch.bfloat16.\r\n\r\n* Add support for ushort and uint.\r\n\r\n* Convert np.uint32 to np.int32 when loading.\r\n\r\n* Fix warning.","parents":["9c32474a1fe90b0e3de3d08eeea51c44fd88c68f"],"tree_hash":"63975b7230d25538d8a853af3fa006ed57a99230"}