{"author":"qazal","author_email":"77887910+Qazalin@users.noreply.github.com","author_time":1703518914,"commit_time":1703518914,"committer":"GitHub","committer_email":"noreply@github.com","hash":"12996d3a7d37b26f75fe2ade778654ae11378613","message":"green linearizer asserts for ops (#2800)\n\n* these asserts should pass\r\n\r\n* fix that assert\r\n\r\n* ALU dtypes\r\n\r\n* acc dtype for group_for_reduce\r\n\r\n* cast image ALUs to the base dtype\r\n\r\n* remove all casts from linearizer\r\n\r\n* fix argmax\r\n\r\n* fix multinomial\r\n\r\n* fix __getitem__\r\n\r\n* Revert \"fix __getitem__\"\r\n\r\nThis reverts commit 62ad719bfa5a2e1fcbfa931360f54897f8977602.\r\n\r\n* fix MemBuffer outputs being wrong when there is an arange + ALU with a different dtype\r\n\r\neg. fancy slicing (int, float), bert embeddings (int, long)\r\n\r\nthis should be fixed in lazy instead of having to break the kernel\r\n\r\n* cleanup argmax fix\r\n\r\n* fix matmul in ints\r\n\r\ncast in the end\r\n\r\n* fix llama\r\n\r\n* skip wrong hardcoded asts in the worlds dataset\r\n\r\n* fix llama p2\r\n\r\n* cleanup missing parts of the diff\r\n\r\n---------\r\n\r\nCo-authored-by: George Hotz <geohot@gmail.com>","parents":["1fb815e77e846945065a8c0e4e327d0252a677c3"],"tree_hash":"e1bfe5ed601456c2981632cb62cff3f6d9d3d013"}