{"author":"qazal","author_email":"77887910+Qazalin@users.noreply.github.com","author_time":1701821968,"commit_time":1701821968,"committer":"GitHub","committer_email":"noreply@github.com","hash":"be09cc87c11ce65adaa67415b2387fa80e5e9e7f","message":"Bitcast support / fast bf16 load (#2011)\n\n* bitcast renderers\r\n\r\n* fast llama load\r\n\r\n* make it one kernel\r\n\r\n* regression testing p1: re-enable test_dtype for all backends\r\n\r\nfix GPU\r\n\r\n* regression testing p2: fuzz all possible cases against numpy\r\n\r\nremove hancoded tests since the fuzzer covers them\r\n\r\n* define ushort\r\n\r\n* fix indent, probably need flake8 back for CI to catch\r\n\r\n---------\r\n\r\nCo-authored-by: George Hotz <72895+geohot@users.noreply.github.com>","parents":["232ed2af3f2ebb55617672adc0df8460e858f191"],"tree_hash":"f28c051050a5d46f096af9e1471c4e1b7cac8b62"}