chenyu
b89ee1ac83
lazy type annotation and cleanups ( #1897 )
2023-09-22 14:20:23 +08:00
George Hotz
78576915de
Add needed contiguous to DiskBuffer. SHM support on OSX ( #1891 )
...
* add some contiguous
* remove second contig
* Revert "remove second contig"
This reverts commit fc164f7dca1ad75b1e466e4e45a05eca58b7e0e0.
* shm on osx
* can repro bug
* don't contig zeros and ones
2023-09-22 09:16:42 +08:00
qazal
d0e752003d
fixes ( #1893 )
2023-09-22 07:20:27 +08:00
wozeparrot
009a99a0b1
feat: way cleaner hip wrapper ( #1895 )
2023-09-22 07:20:03 +08:00
Yixiang Gao
cb5d6576cb
cifar step time 65ms while stay above 94% ( #1888 )
...
* change reduceop heruistics
* add model ema and jit hack
* add ema eval
* have to create a duplicate eval function for jit
* remove manual seed
* 94% achieveable with normal eval
* ema is outputting the same results as normal
* fix ema bug
* ema achieves 94% with fix seed
* multigpu tested
* constant fold decay, fix jit, adjust message for multigpu
* pull SpeedyResNet out of train_cifar()
2023-09-21 11:19:32 +08:00
kormann
864746d6aa
polish print_tree ( #1868 )
...
* fix
* isinstance
2023-09-21 11:13:10 +08:00
chenyu
a5090f0ee9
remove NumNode.int() ( #1876 )
2023-09-21 10:29:16 +08:00
Gijs Koning
9eb6310686
Fix gpt optimization ( #1885 )
...
* fix for gpt
* the actual fix
* Remove change in symbolic
* small comment
2023-09-21 10:28:18 +08:00
Szymon Ożóg
bd3444797b
make ssa assign r[u] ( #1887 )
2023-09-21 10:20:20 +08:00
nimlgen
9450e41f70
no import when Python is shutting down ( #1875 )
2023-09-20 12:47:02 -04:00
Yixiang Gao
84ab47a90a
add branch up-to-date check ( #1879 )
2023-09-20 12:41:51 -04:00
nimlgen
504bb6d0ea
support symbolic jit in HIP ( #1877 )
2023-09-20 01:44:26 -04:00
chenyu
cd66c9e249
no numnode in shape ( #1871 )
2023-09-17 07:49:45 +08:00
Yixiang Gao
18ec5a9e09
add comment bot to CI ( #1873 )
2023-09-16 12:22:06 -04:00
Yixiang Gao
a27f6c7d62
add diff mode to sz.py ( #1872 )
2023-09-16 00:43:47 -04:00
nimlgen
4c31dfafb3
add seed to gpt-2 ( #1869 )
2023-09-15 17:34:14 -04:00
wozeparrot
c870764940
Revert "add line changes diff bot to CI ( #1863 )" ( #1870 )
2023-09-15 16:56:42 -04:00
Yixiang Gao
789c84a7a3
add line changes diff bot to CI ( #1863 )
2023-09-15 16:29:58 -04:00
chenyu
29ac8293d7
run gpt2 in CI ( #1866 )
2023-09-15 04:37:02 +08:00
chenyu
1b46de1a3e
fix type of helpers.prod, add test cases ( #1859 )
2023-09-14 05:16:55 +08:00
chenyu
e67306ba04
symbolic shape type with TypeGuard ( #1852 )
2023-09-13 05:27:22 +08:00
Roelof van Dijk
c91b44f7bf
refactor: move size to view ( #1848 )
...
* refactor: move size to view
* fix: pylint
---------
Co-authored-by: Roelof van Dijk <roelof.van.dijk@vitestro.com >
2023-09-11 07:16:04 -07:00
chenyu
9e9ea20784
Fix view, CI cpu test with python 3.8 ( #1845 )
2023-09-10 22:37:58 -04:00
chenyu
3ec301c2d7
apply view.py patch ( #1844 )
2023-09-10 17:32:15 -07:00
Yixiang Gao
a32951a001
add test_tensor_copy ( #1840 )
...
* add test_tensor_copy
* fix whitespace
* add value check
2023-09-10 16:01:58 -07:00
Roelof van Dijk
1bc52c60df
fix: minor tweaks to view ( #1842 )
...
Co-authored-by: Roelof van Dijk <roelof.van.dijk@vitestro.com >
2023-09-10 15:55:57 -07:00
George Hotz
47e602f717
view: do not trade complexity for speed ( #1839 )
...
* view: do not trade complexity for speed
* staticmethods
* view create
2023-09-10 11:29:53 -07:00
chenyu
c0bc4cfbaf
DivNode.b is int ( #1833 )
2023-09-10 09:04:29 -07:00
nimlgen
13790b1e20
cast types in render_load ( #1837 )
2023-09-10 07:58:13 -07:00
David Hou
e74a6ca7e4
expand in terms of substitute ( #1827 )
2023-09-09 14:43:00 -07:00
George Hotz
0e3e2bac13
amd wino: upload results
2023-09-09 13:57:14 -07:00
George Hotz
6f95c5f284
winograd speed test for AMD ( #1826 )
2023-09-09 13:56:33 -07:00
George Hotz
0f2bd10d00
add winograd CIFAR to mac tests ( #1825 )
...
* add winograd CIFAR to mac tests
* symlink already done
2023-09-09 13:45:24 -07:00
nimlgen
31fca43706
kopt works with local+grouped reduce and tests ( #1824 )
2023-09-09 13:22:09 -07:00
chenyu
9da40c8448
move Node.__lt__ SumNode special case to SumNode ( #1823 )
2023-09-09 13:20:38 -07:00
Francis Lam
651205fa5c
linearizer: support local and group_for_reduce dimensions together ( #1821 )
...
also minor changes to test_speed_v_torch.py and size of UOps.SPECIAL
2023-09-08 12:39:27 -07:00
segf00lt
9e8c1dbf34
patch to remove hack from stable_diffusion.py ( #1814 )
...
* patch to remove hack from stable_diffusion.py
* sorry linter
* realize after assign?
* float16 broken in llvmlite use float64 for now
* int32
* idiot forgot to change test array dtype
2023-09-08 09:26:50 -07:00
chenyu
ebcda8a714
Move var_vals from ShapeTracker to LazyBuffer ( #1819 )
2023-09-08 09:25:10 -07:00
kormann
7ac65a93b4
utils.printtree ( #1816 )
...
* utils.printtree
* linter compliance
* rename to print_tree
2023-09-07 23:08:57 -07:00
George Hotz
4613c9e77c
add tvm example, formatting ( #1813 )
...
* add tvm example
* no realize
2023-09-07 11:50:41 -07:00
nimlgen
5b15a972b5
no functions with same names in test/ ( #1811 )
2023-09-07 11:27:31 -07:00
George Hotz
722823dee1
stable diffusion: force fp16 free
2023-09-06 15:11:05 -07:00
chenyu
928cb1a64a
AndNode.substitute short circuit ( #1800 )
...
* AndNode substitute short circuit
* Node.__bool__ is faster than Node.__eq__
2023-09-06 14:58:49 -07:00
nimlgen
a78a1fa499
fix jit buffer reuse when freed ( #1802 )
...
* fix jit buffer reuse when freed
* Firbid output_buffer reusage
2023-09-06 14:41:57 -07:00
Yixiang Gao
22cf15e9d0
convert function into tinygrad ( #1803 )
2023-09-06 14:41:26 -07:00
Pavol Rusnak
52a92bf95d
use class Foo: instead of class Foo(): ( #1797 )
...
* use class Foo: instead of class Foo():
* add ruff linter, copy settings from .flake8 to ruff.toml
2023-09-06 12:20:25 -07:00
badcc
fd25792c8b
Ensure freqs as type float32 in freqs_cis ( #1798 )
2023-09-06 10:24:15 -07:00
chenyu
35072877ef
sym_infer is noop for int input ( #1795 )
2023-09-06 09:17:20 -07:00
George Hotz
f67638b27a
delete broken DDPG example
2023-09-06 08:01:12 -07:00
George Hotz
78a43ad2c7
add uop fixup ( #1793 )
2023-09-06 07:55:22 -07:00
