{"author":"wozeparrot","author_email":"wozeparrot@gmail.com","author_time":1710794827,"commit_time":1710794827,"committer":"GitHub","committer_email":"noreply@github.com","hash":"a0ab755317ef1650e48dd1aaae325d3835a3146c","message":"threefry again (#3785)\n\n* feat: initial xor\r\n\r\n* feat: initial threefly\r\n\r\n* feat: remove custom random\r\n\r\n* fix: really need to install precommit\r\n\r\n* feat: lmao forgot that this is rotate not a shift\r\n\r\n* clean: put that there\r\n\r\n* feat: numpy xor\r\n\r\n* feat: quick test for xor\r\n\r\n* feat: llvm xor\r\n\r\n* feat: slightly working xor in torch\r\n\r\n* feat: rand works in jit\r\n\r\n* clean: save a line\r\n\r\n* feat: match jax\r\n\r\n* feat: maybe test against jax\r\n\r\n* feat: requires_grad\r\n\r\n* fix: fix test_symbolic_ops\r\n\r\n* feat: lower alpha\r\n\r\n* feat: just pad\r\n\r\n* fix: maybe fix training tests?\r\n\r\n* fix: fix some llvm stuff\r\n\r\n* feat: cursed realize on the way out\r\n\r\n* feat: testing jax\r\n\r\n* fix: why is the jax install process not simple\r\n\r\n* fix: maybe passing test\r\n\r\n* fix: symbolic workarounds\r\n\r\n* clean: still need that precommit\r\n\r\n* fix: aaaa\r\n\r\n* fix: more test fixes\r\n\r\n* fix: quick fix for wgsl\r\n\r\n* feat: need to set requires_grad on the final tensor\r\n\r\n* feat: one more tensor\r\n\r\n* feat: don't take forever\r\n\r\n* feat: seeing y ci is brok\r\n\r\n* feat: can't allocate 64GiB lmao\r\n\r\n* fix: fix this\r\n\r\n* feat: hope this doesn't break smth before i go to bed\r\n\r\n* feat: don't destroy ram\r\n\r\n* feat: int\r\n\r\n* feat: remove jax\r\n\r\n* feat: properish workaround?\r\n\r\n* feat: skip slow webgpu tests\r\n\r\n* feat: no longer fails\r\n\r\n* feat: use dtypes\r\n\r\n* feat: real number\r\n\r\n* fix: torch\r\n\r\n* fix: don't test against reference for torch\r\n\r\n* feat: to device\r\n\r\n* feat: fix advanced indexing\r\n\r\n* feat: correct casting\r\n\r\n* feat: even rng_counter\r\n\r\n* feat: match master\r\n\r\n* feat: this was actually bad\r\n\r\n* fix: maybe?\r\n\r\n* feat: store\r\n\r\n* feat: remove realizes\r\n\r\n* feat: somehow this is important\r\n\r\n* feat: somehow this is also important\r\n\r\n* feat: save a line\r\n\r\n* fix: don't need that anymore\r\n\r\n* feat: restore this\r\n\r\n* fix: linter\r\n\r\n* feat: remove realizes\r\n\r\n* fix: realized is in base now\r\n\r\n* fix: add back cast\r\n\r\n* fix: bump deadline\r\n\r\n* fix: bump deadline\r\n\r\n* fix: bump deadline\r\n\r\n* fix: bump deadline\r\n\r\n* fix: bump deadline\r\n\r\n* fix: :(\r\n\r\n* fix: :(\r\n\r\n* fix: not being dumb\r\n\r\n* feat: try changing less tests\r\n\r\n* feat: shouldn't have to change that\r\n\r\n* feat: contiguous bumps it by one\r\n\r\n* fix: hmm\r\n\r\n* fix: numpy memory moment\r\n\r\n* fix: cl_khr_fp16\r\n\r\n* fix: torch has different tensor count\r\n\r\n* fix: missing contiguous\r\n\r\n* hmm: hmm\r\n\r\n* fix: some fixes\r\n\r\n* fix: typing\r\n\r\n* feat: dont do that\r\n\r\n* feat: typing fixes\r\n\r\n* feat: why is this realize required?\r\n\r\n* feat: ngl kinda odd typing\r\n\r\n* feat: oh\r\n\r\n* feat: remove realizes\r\n\r\n* feat: why is this realize required?\r\n\r\n* fix: hacky patch for cudacpu\r\n\r\n* fix: without this realize pytest crashes?????\r\n\r\n* fix: shorter line\r\n\r\n* fix: cudacpu fixes\r\n\r\n* fix: cudacpu fixes\r\n\r\n* feat: real buffer\r\n\r\n* feat: don't search when searching lmao\r\n\r\n* fix: can't use contiguous things\r\n\r\n* fix: no more 100GB arrays\r\n\r\n* fix: revert\r\n\r\n* fix: skip 7 and 10\r\n\r\n* feat: working ish beam\r\n\r\n* feat: minimize changes\r\n\r\n* feat: seed 0 stable diffusion example changed\r\n\r\n* fix: different on ci\r\n\r\n* fix: no beam\r\n\r\n* feat: make threefry optional\r\n\r\n* fix: check value\r\n\r\n* fix: unused import\r\n\r\n* feat: threefry default\r\n\r\n* fix: 5d\r\n\r\n* feat: allow non upcast div\r\n\r\n* fix: 5d better\r\n\r\n* fix: 5d better\r\n\r\n* fix: save all dtype\r\n\r\n* feat: proper error\r\n\r\n* feat: lazyop key\r\n\r\n* fix: check float\r\n\r\n* feat: try removing this realize now\r\n\r\n* feat: disable threefry for uops hip tensor cores\r\n\r\n* feat: don't need that\r\n\r\n* feat: only check upcast\r\n\r\n* fix: disable threefry for some metal tests\r\n\r\n* feat: disable for metal tensor uops as well\r\n\r\n* feat: disable for most uops\r\n\r\n* fix: disable threefry for new uops tests\r\n\r\n* feat: multitensor\r\n\r\n* fix: typing\r\n\r\n* feat: threefry default off\r\n\r\n* feat: skip threefry half rand\r\n\r\n* feat: restore old\r\n\r\n* fix: bad git\r\n\r\n* clean: ruff\r\n\r\n* feat: bfloat16 fix\r\n\r\n* fix: :|\r\n\r\n* feat: restore old\r\n\r\n---------\r\n\r\nCo-authored-by: chenyu <chenyu@fastmail.com>","parents":["7af7467f22df9c2b232cd5b2d3f53e92cbf00819"],"tree_hash":"50b85e258dc922a5695ba818094b6b527fdb9418"}