commaai
/
tinygrad
mirror of https://github.com/commaai/tinygrad.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
3,087 Commits 10 Branches 0 Tags 52 MiB
Commit Graph

3087 Commits

Author SHA1 Message Date
chenyu 66d9eb10b6
arange default dtype to int and zeros/ones default to float (#2769) 2023-12-14 17:53:00 -05:00
qazal 3cf4376ce2
test_linearizer cleanup (#2766) 
* test_linearizer cleanup

* use unittest.skipIf

* update msg
 2023-12-14 17:20:09 -05:00
chenyu 57017c87e9
remove duplicated dtype in DEFINE_GLOBAL args (#2768) 
now DEFINE_GLOBAL uop.arg[1] is always the same as uop.dtype, we can remove the one in arg and just use uop.dtype
 2023-12-14 15:42:36 -05:00
chenyu 5235cdee3d
remove _arg_int32 internal type (#2767) 
in DEFINE_GLOBAL, PtrDtype(int32) is buffer and int32 is int
 2023-12-14 14:17:14 -05:00
chenyu 8a2a2257b4
minor onnx_op cleanups to prep dtype changes (#2764) 
* minor onnx_op cleanups to prep dtype changes

read through it and clean some minor stuff

* revert embedding - is it really being tested
 2023-12-14 13:01:27 -05:00
geohotstan 0398288b79
Getitem round3 .... (#2760) 
* refactor round 3

* comment

* oops

* oops

* oops2

* factored out multiple condition

* add a comment for type

* wooaah roundup is cool, thanks chenyu lol

* add another walrus for symmetry and some spaces

* lol wtf useless listcompre
 2023-12-14 12:22:37 -05:00
chenyu 0ae22b0f81
restore Tensor.default_type in test_hip_rdna3 (#2763) 
might cause flaky tests
 2023-12-14 11:35:38 -05:00
qazal 746cb5de21
Test coverage for matvec (#2762) 
* add test coverage for matvec

* skip devices that don't support locals
 2023-12-14 11:34:56 -05:00
chenyu 64fea9ff4a
Revert "minor onnx_op cleanups to prep dtype changes (#2758)" (#2759) 
This reverts commit 38da001b64.
 2023-12-14 03:12:14 -05:00
chenyu 38da001b64
minor onnx_op cleanups to prep dtype changes (#2758) 
read through it and clean some minor stuff
 2023-12-14 03:05:59 -05:00
jaredeh d8952fc575
updating to work with new internal apis (#2755) 2023-12-13 21:54:47 -08:00
chenyu 2c6814ba28
insert_before is None means insert at the end (#2757) 2023-12-13 21:05:10 -05:00
chenyu aad005e220
set default str for CStyleLanguage.arg_int_prefix (#2756) 
it's the same `const int` for clang, opencl, cuda and hip
metal overwrites with `constant int&` and webgl has its own thing
 2023-12-13 20:23:27 -05:00
chenyu 107dd8f3d7
fix a typo in test_dtype_alu (#2754) 2023-12-13 19:23:21 -05:00
chenyu fc6bca7ba8
update type annotation of _broadcasted (#2753) 
input can be Tensor, float, int.
also updated scaled_dot_product_attention that might add a None to a Tensor
 2023-12-13 19:03:14 -05:00
Maksym Sobolyev bf4165ccac
Fix double exception in __del__() when __init__() raises exception. (#2738) 2023-12-13 15:46:11 -08:00
chenyu 81a747fc63
more test cases in test_slice_fancy_indexing_with_idx (#2751) 2023-12-13 17:52:26 -05:00
chenyu 22feb7330e
simplify fancy index with negative Tensor entries (#2749) 2023-12-13 14:45:50 -05:00
chenyu b229879613
refactor _broadcasted (#2747) 
also moved the expand noop check to .expand.
 2023-12-13 13:36:25 -05:00
George Hotz 7e5b3e53fe
changes to prep for new lazy (#2748) 
* changes to prep for new lazy

* put those back
 2023-12-13 10:28:22 -08:00
Umut Zengin 8ad7cfeeb1
More simplification in to_image_idx and symbolic (#2679) 
* less valid

* add test

---------

Co-authored-by: chenyu <chenyu@fastmail.com>
 2023-12-13 12:30:44 -05:00
Ahmed Harmouche e7248b677c
Remove wgsl custom render_for (#2729) 
* Generic for

* remove custom render_if

* Simplify for loop

* 150 line-length constraint

* Put custom render_if back
 2023-12-13 09:04:17 -08:00
tomtom-95 6b0f07e94a
add decorator to preserve info about original function (#2743) 2023-12-13 09:03:50 -08:00
chenyu aa4a0de287
simpler Tensor.pow to integer (#2746) 2023-12-13 11:39:20 -05:00
chenyu 26f49869f4
minor tensor type annotation and cleanup (#2742) 2023-12-13 01:53:59 -05:00
chenyu 2ef33abd20
some unary functions cast int input into float (#2740) 
* some unary functions cast int input into float

* precision

* image dtype
 2023-12-13 00:10:29 -05:00
George Hotz 3e778fcc52 hotfix: *** 2023-12-12 19:44:31 -08:00
Shawn Hagler 51afe938f1
update onnx model links (#2737) 2023-12-12 19:11:11 -08:00
George Hotz 431fae5ed3 hotfix: update_stats cleanup, yellow is nicer than red 2023-12-12 17:50:22 -08:00
chenyu 0869e7a301
update onnx benchmark urls (#2735) 
onnx is remapping the models, old ones are in archive/
 2023-12-12 20:46:01 -05:00
George Hotz 6d6eb9302d
ruff checks the max line length is 150 (#2734) 
* ruff checks the max line length is 150

* fix tensor.py

* a lot more

* done
 2023-12-12 17:34:47 -08:00
George Hotz 3635540ddb
shorter line (#2733) 2023-12-12 15:34:17 -08:00
nimlgen ede7971ada
save some lines (#2731) 
* remove unsused mem_cached var

* one more
 2023-12-12 15:26:27 -08:00
chenyu 00b611c156
simplify type promotion - remove weak types (#2730) 2023-12-12 16:12:57 -05:00
Nguyen Nguyen Phuong 07cf45e133
fix cuda matmul (#2725) 2023-12-12 07:59:31 -08:00
chenyu ef6e942a23
dtype promotion helpers (#2724) 
* dtype promotion helpers

* better tests

* space
 2023-12-11 23:14:23 -05:00
Christopher Mauri Milan 0232db294d
fix tolist issue (#2723) 2023-12-11 19:14:00 -08:00
chenyu 4075208127
some dtype creation spec test cases (#2722) 2023-12-11 19:33:49 -05:00
Guy Leroy ee9e1d3662
Extend available types for `safe_save` (#2720) 
* Extend available types to save with

* Linter fix
 2023-12-11 14:50:35 -08:00
George Hotz b5fd160b39 hotfix: increase rtol on simple_matmul 2023-12-11 10:10:29 -08:00
Gregor Kikelj 4feaaa27aa
ensure shrink is valid (#2717) 2023-12-11 09:58:43 -08:00
qazal a43bc78804
fix dtypes helpers for integers (#2716) 
* scalar

* maybe do this instead

* Revert "scalar"

everything is a scalar

* add tests in test_dtype

* fuzz testing + fix unsigned ints

* fuzz everything
 2023-12-11 09:28:19 -08:00
nimlgen bc3c4ce50b
cuda set context before sync (#2715) 
* cuda set context before sync

* no helper
 2023-12-11 09:26:53 -08:00
Ivan Vnučec 8d206f6bfd
fix help message (#2705) 
llama -> mixtral
 2023-12-10 22:04:35 -08:00
George Hotz 59ab3675a3
faster mixtral + green for new kernels (#2701) 
* green for new kernels

* track ram
 2023-12-10 19:04:58 -08:00
chenyu 2ee6f689c5
simpler einsum (#2700) 2023-12-10 21:24:44 -05:00
George Hotz b01e3907a1 mixtral touch up: two lines 2023-12-10 17:21:49 -08:00
George Hotz b3982187d1
Mixtral Example (#2691) 
* mixtral

* simpler

* global counters

* simpler

* weights arg
 2023-12-10 17:18:31 -08:00
George Hotz 0fd44259cd
bf16 fix + cleanups from mixtral (#2698) 
* bf16 fix + cleanups from mixtral

* generic bf16 cast
 2023-12-10 16:31:52 -08:00
Davi Silva 7fbebb3df6
Implement einsum (#2686) 
* hopeful impl for Tensor.einsum

* satisfy mypy by having less typing. :(

* a few simple tests

* even more tests

* permute tests

* xfails for improper usage

* fix LLVM test fail

* use argfix

* more helpful error message on shape mismatch
 2023-12-10 15:56:01 -08:00