github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-23 22:08:08 -05:00
Code Issues Packages Projects Releases Wiki Activity
3,196 Commits 205 Branches 14 Tags
d424babe2cd22ed5d09eff59af3bd7bf8db7eb51
Commit Graph

3196 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
chenyu
d424babe2c tensor.py cleanup around Tensor.slice (#2921) 
use None for no-op slice and pad
 2023-12-22 19:46:39 -05:00
chenyu
089703a390 cleanup test_dtype_alu (#2919) 
wrapped long lines and lowered atol for METAL.sin to 2 since atol of two sins are bounded by 2
 2023-12-22 17:29:31 -05:00
chenyu
3ba591c3fd less outdated abstraction.py (#2917) 
removed some old terms and updated types and code pointers
 2023-12-22 15:31:02 -05:00
chenyu
50927defad s/lazydata.realized/lazydata.base.realized/g (#2914) 
* s/lazydata.realized/lazydata.base.realized/g

* not that
 2023-12-22 14:45:13 -05:00
chenyu
2783e1b50d bugfix Tensor.item when it's unbased (#2913) 
it's possible for numel 1 tensor lazydata to be unbased and should call lazydata.base.realized
 2023-12-22 13:50:06 -05:00
Oleg Rybalko
c3133adb8c Disk shm refactor (#2912) 
* better support for platform dependent flags

* osx test support

* removed unused import and made line length <150

* changed osx ci shm

* lstrip in case SharedMemory._name is passed
 2023-12-22 09:23:37 -08:00
chenyu
3855432265 don't use numpy to create Tensor(None) (#2909) 
* don't use numpy to create Tensor(None)

empty suffices

* parentheses
 2023-12-22 01:07:44 -05:00
chenyu
50cfb1fb3a update onnx model links (#2908) 
updated in https://github.com/onnx/models/pull/644
 2023-12-22 00:19:41 -05:00
chenyu
1bbeb3fe2f remove the different rtol / atol for openpilot CUDA in benchmark (#2907) 
not sure what the issue was but seems to be fixed on master
 2023-12-21 22:23:39 -05:00
chenyu
a543d8bea8 fuzz default dtypes for some test_dtype tests (#2906) 
* fuzz default dtypes for some test_dtype tests

* ocd

* setUp and tearDown
 2023-12-21 22:00:21 -05:00
wozeparrot
5f3d5cfb02 catch cycles in print_tree (#2891) 
* feat: smaller tree on references

* fix: shorter line

* fix: huh

* fix: should be all

* feat: cleaner

* fix: extra imports

* fix: pass by reference
 2023-12-21 18:40:37 -08:00
George Hotz
4432cb17bb minor cleanups / remove that op (#2905) 2023-12-21 18:24:20 -08:00
chenyu
fd0ba33b38 onnx_ops formatting cleanup (#2904) 
also removed a case in safe_numpy that always convert 0-dim array to 1-dim
 2023-12-21 20:06:06 -05:00
George Hotz
5cac6338a4 apply the multitensor optimizations in lazy.py (#2901) 
* apply the multitensor optimizations in lazy.py

* less lines

* hack for webgpu

* save a line
 2023-12-21 13:55:49 -08:00
chenyu
5bf43c9634 reenable one onnx test failed due to dtype (#2902) 2023-12-21 15:50:02 -05:00
chenyu
677ae7673d use np.less and torch.lt for CMPLT (#2899) 
also removed one unused output_type
 2023-12-21 14:37:24 -05:00
qazal
d2e9245de8 render_locals takes a dtype (#2873) 
Co-authored-by: chenyu <chenyu@fastmail.com>
 2023-12-21 14:15:28 -05:00
chenyu
6116039f7b don't match dtype with first input in where (#2898) 
* don't match dtype with first input in where

`Tensor([1, 2, 3]).where(1.2, 2.3)` the first `[1, 2, 3]` can directly cast into bool without casting float (in broadcasted) first

* cast in one place
 2023-12-21 13:02:15 -05:00
chenyu
7dc3352877 increase stable diffusion validation threshold 1e-4 -> 3e-4 (#2897) 
saw a flaky CI failure with 1.1e-4, and 3e-4 is a good number
 2023-12-21 11:45:25 -05:00
qazal
24e79e0f53 Move the webgpu CMPLT hack to one place (#2895) 
* move hacks to one place

* no casting in mlops, move to tensor

* ruff fix
 2023-12-21 11:14:56 -05:00
George Hotz
852ef57ba4 fix readme typo 2023-12-21 08:06:24 -08:00
George Hotz
193109a88c hotfix: compare on ids 2023-12-20 23:47:50 -08:00
George Hotz
f6c7833f9f fast compare for lazyop (#2893) 2023-12-20 23:32:27 -08:00
chenyu
1500aca43d remove output_type in ops_cpu and ops_torch (#2892) 
now the input types are matched and checked in lazy, we can remove these output_type.
also remove the usage of least_upper_dtype in ops.py since we can just use the input type
 2023-12-21 02:11:27 -05:00
chenyu
2d2c4980fe assert for elementwise dtypes in lazy (#2888) 
* assert for elementwise dtypes in lazy

* no image hack

* check dtype of scalar for IMAGE=2
 2023-12-21 01:42:32 -05:00
George Hotz
41b2a25be6 Fix exponential behavior in lazyops (#2890) 
* add cache to ast_parse and lazyop builder

* add caches
 2023-12-20 22:06:50 -08:00
George Hotz
8c4a0f8e15 Fix int child count (#2882) 
* pad ops broke coder

* that contiguous fixes it

* Update lazy.py

* recursive add

* fix all

* revert that

* todo test
 2023-12-20 21:06:27 -08:00
chenyu
8a04107d30 move the op casting logic from mlops to tensor try 2 (#2887) 
* unary works

* where works

* add sub mul

* xor div

* CMPLT

* sparse_categorical_crossentropy

* image const

* sparse_categorical_crossentropy
 2023-12-20 23:50:37 -05:00
George Hotz
7da2325dc7 get_lazyops() -> lazyops (#2884) 
* get_lazyops() -> lazyops

* don't compare empty mem
 2023-12-20 18:04:49 -08:00
George Hotz
64dded27f0 pad ops broke coder (#2881) 
* pad ops broke coder

* that contiguous fixes it

* Update lazy.py
 2023-12-20 17:03:41 -08:00
George Hotz
e1861ab65e remove realize from optimizer (#2880) 
* remove realize from optimizer

* one still needed

* opt realize
 2023-12-20 16:42:41 -08:00
George Hotz
1765849937 new lazy, benchmark (#2878) 
* lazy rewrite, try 2

* min fix tests

* pass contig test

* put broken pads back

* move that to realize

* no contig child fixes array packing

* so wrong

* now that's correct

* base children

* fix bind issues

* disable to_image_idx

* fix tests

* that failure shouldn't break other tests

* more fixes

* fix torch

* skip failing tests in CI

* 1e-7

* half is broken

* 1e-6 margin of error
 2023-12-20 14:33:21 -08:00
Peter Cawley
dae8976889 Fix reshape merging with masks (#2877) 2023-12-20 14:00:58 -08:00
George Hotz
8fe24038d8 Revert "mulacc fusion cleanup (#2871)" (#2876) 
This reverts commit 863c5b26ed.
 2023-12-20 13:26:25 -08:00
qazal
863c5b26ed mulacc fusion cleanup (#2871) 
* add mulacc fusion tests

* cleanup the implementation

* fix indent in the test utility

* less verbose
 2023-12-20 15:39:54 -05:00
chenyu
e13b4964d7 remove the all_int(shape) check in Tensor._loadop (#2874) 
* remove the all_int(shape) check in Tensor._loadop

we can support jittable symbolic shape random with custom rand now, and we can formalize it in the test after threefry is ready

* MOCKHIP false positive
 2023-12-20 15:04:50 -05:00
qazal
5f07ef455e update dtypes (#2872) 2023-12-20 15:04:02 -05:00
chenyu
857c35d256 make gpt2 decode output just once at the end (#2869) 
also updated function name from greedy_until to generate, as it's not greedy nor until
 2023-12-20 12:14:55 -05:00
chenyu
e92069fb1c remove unused symbolic.is_sym_int (#2868) 2023-12-20 11:37:54 -05:00
George Hotz
ca59054463 fix shapetracker math (#2861) 
* proper test

* all st math good now

* fix real_strides bug
 2023-12-19 22:17:34 -08:00
chenyu
5a739e8c20 update one skipped pad_reshape test that was fine (#2860) 
* update one skipped pad_reshape test that was fine

had a typo

* this one passed
 2023-12-19 23:25:52 -05:00
chenyu
39af93ed7c minor tensor.py function cleanup (#2859) 
* minor tensor.py function cleanup

* where outputs not aligned yet
 2023-12-19 22:39:39 -05:00
George Hotz
94f71fe238 random and empty shouldn't reshape 2023-12-19 18:09:03 -08:00
George Hotz
637879af78 add direct install to readme 2023-12-19 18:04:00 -08:00
chenyu
ad233d557f disable reshape merging with masks (#2858) 
fuzzer found a bug, and it's not complete
 2023-12-19 19:06:16 -05:00
chenyu
1231ec5a02 run the sz.py line count at the end of linter ci (#2857) 2023-12-19 16:33:12 -05:00
George Hotz
ac6ec936cd update contributing 2023-12-19 12:19:14 -08:00
George Hotz
e477cc2f45 hotfix: README is ~25 ops to stop getting PRs about it 2023-12-19 11:53:35 -08:00
Oleg Rybalko
42a038c83f More readable torch_load ext check (#2853) 
* more readable extension check

* enable tarfile test

* detach tensor if requires grad in torch
 2023-12-19 14:53:15 -05:00
chenyu
172a88e719 skip slow test_indexing on METAL (#2852) 
LLVM still runs and is a lot faster, would be curious to know why.
also reworded some error messages and remove regex check
 2023-12-19 12:00:54 -05:00