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

206 Commits

Author SHA1 Message Date
George Hotz ce0c80e1bd ugh, corrupt downloads. when did this break? 2020-11-07 08:04:43 -08:00
George Hotz 9332fe1989 weee, logsoftmax on GPU. very slow 2020-11-07 07:59:37 -08:00
George Hotz 38e112cccd logsoftmax test 2020-11-07 07:26:53 -08:00
George Hotz 237bbe3625 sigmoid on GPU 2020-11-07 07:23:46 -08:00
Rakib Fiha f40dbd791c
Use --upgrade since its in active dev (#63) 2020-11-07 07:15:05 -08:00
Rene Delgado cd54697fd8
fix gpu sum forward (#61) 
* ignore venv

* add sum test

* fix sum forward
 2020-11-05 21:59:16 -08:00
gallanoe aea1069f63
Div on CPU (#58) 
* Added Div on CPU

* Removed eps. value

* Fixed tabs
 2020-11-05 21:58:37 -08:00
NeuralLink cc605da36d
Stable Sigmoid op (#59) 
* 🔨 Added stable sigmoid

* ✅ added sigmoid test

* 🔧 suppressed overflow warning

* 🔧 clean up
 2020-11-05 21:57:50 -08:00
George Hotz cc852efa64 make OpenCL work elsewhere 2020-11-05 07:59:55 -08:00
Ryan Neph 22a5f9975d
pow+div on GPU (#57) 2020-11-05 07:49:45 -08:00
adamritter b16fadc5c6
Speed up sum backward pass (#54) 
Co-authored-by: holonomicjl <58403584+holonomicjl@users.noreply.github.com>
 2020-11-04 19:55:45 -08:00
George Hotz 0be3d2cdba install pyopencl 2020-11-04 11:22:11 -08:00
George Hotz 3efb4f4df4 chicken.jpg 2020-11-04 11:20:22 -08:00
Marcel Bischoff c69582744d
Update efficientnet.py (#53) 
from https://github.com/lukemelas/EfficientNet-PyTorch/blob/master/efficientnet_pytorch/utils.py
```
    blocks_args = [
        'r1_k3_s11_e1_i32_o16_se0.25',
        'r2_k3_s22_e6_i16_o24_se0.25',
        'r2_k5_s22_e6_i24_o40_se0.25',
       'r3_k3_s22_e6_i40_o80_se0.25',
        'r3_k5_s11_e6_i80_o112_se0.25',
        'r4_k5_s22_e6_i112_o192_se0.25',
        'r1_k3_s11_e6_i192_o320_se0.25',
    ]
```

now it's a persian cat.
 2020-11-04 09:36:27 -08:00
Marcel Bischoff c40862241d
Revert "choose aspect ration dependent on landscape or portrait (#48)" (#51) 
This reverts commit 2808411421.
 2020-11-02 19:31:07 -08:00
Anders Aaen Springborg 2808411421
choose aspect ration dependent on landscape or portrait (#48) 2020-11-02 18:10:22 -08:00
Marcel Bischoff d24363f421
Update setup.py (#49) 
I think `:=` in tinygrad/test/test_mnist.py actually needs 3.8
 2020-11-02 18:09:31 -08:00
George Hotz 940e14c6ca more readme 2020-11-02 08:33:48 -08:00
George Hotz 1e6bbdf4f8 readme updates 2020-11-02 08:30:43 -08:00
George Hotz f178d23ff3 gpu relu is good 2020-11-02 08:25:32 -08:00
George Hotz 8766346187 fix gpu sub 2020-11-02 08:18:58 -08:00
George Hotz 231c1134bd cute trick for GPU test 2020-11-02 08:17:17 -08:00
George Hotz 6a7896f3e9 opsgpu cl program caching 2020-11-02 07:46:26 -08:00
George Hotz 2e7f16bf3f the power of cheating 2020-11-02 07:42:11 -08:00
George Hotz fc358a07ad print sloccount 2020-11-02 07:38:13 -08:00
George Hotz 3da64acd6b the power of small 2020-11-02 07:36:34 -08:00
George Hotz 82fc842b40 in_place_op 2020-11-02 07:26:13 -08:00
George Hotz 1d793b8571 gpu SGD runs now 2020-11-02 07:19:12 -08:00
George Hotz 621c61b30d sgd now operates on Tensors 2020-11-02 07:14:54 -08:00
George Hotz 355402504e ops work 2020-11-02 07:03:23 -08:00
Marcel Bischoff af5a4e0f5a
alternative to einsum which is slow in extreme cases (#44) 
* streamlined numerical_jacobian

* Got rid of the g loop in Conv2D.forward

* ereased stupid line

* nothing

* no loops in Conv2D forward

* Conv2D backprop improved

* stupid things in examples

* alternative to einsum

* Conv2D backward einsum alternative

* tidying up

* tidied up

* no ravel

* got rid of print

* Update efficientnet.py

* Update efficientnet.py

* Update efficientnet.py

* only tensordot

* 255.0

* whitespace

* aspect ratio error in efficientnet

* noprint

Co-authored-by: Marcel Bischoff <marcel@Marcels-iMac.local>
 2020-11-02 06:36:07 -08:00
George Hotz 5201a8e89f matmul on GPU 2020-11-01 08:54:20 -08:00
George Hotz 41e7d59aed test dot 2020-11-01 07:51:35 -08:00
George Hotz 1f544d6ece test mnist on GPU 2020-11-01 07:46:17 -08:00
George Hotz 499604d69b convert tensor to gpu 2020-11-01 07:34:50 -08:00
George Hotz bac6bfda11 with that at least it's only 10x slower 2020-11-01 07:30:19 -08:00
George Hotz 9ac1ad40d6
Add GPU Support! (do not merge yet) (#41) 
* copy tensors to and from gpu

* add on GPU

* adding works

* we stick shapes in

* works on cpu and gpu

* test changes, not passing yet

* something else

* op tests pass

* add, mean, and sum have working forward/backward

* mul ops test

* no gpu support, no problem

* test pass, clean up later

* gpu cleanup

* cleanup test ops, don't let div fail

* revert more

* aimpler dispatcher

* clean up grad

* GPU and

* grad is a Tensor now

* gate test on GPU

* cleanups

* late loading gpu

* GPU as input option

* last cleanups
 2020-11-01 07:00:49 -08:00
liej6799 c06a4fcc80
Fix temp location for windows (#39) 
* Fix temp location for windows

* Make sure get tempfile for other os also.

* Update utils.py
 2020-10-31 17:21:14 -07:00
George Hotz 7c0dc8f48b more whitespace 2020-10-31 11:05:11 -07:00
George Hotz e01e35e545 14 ops to write for GPU 2020-10-31 10:59:30 -07:00
George Hotz 06928cf3cc ENET WORKS 2020-10-31 10:42:58 -07:00
George Hotz 68cba88e8f fix enet bugs, now is mousetrap 2020-10-31 10:28:07 -07:00
George Hotz a852143572 simpler conv2d 2020-10-31 09:20:20 -07:00
Marcel Bischoff e5153d401f
Conv2D forward without loops much faster, backward slightly faster (#38) 
* streamlined numerical_jacobian

* Got rid of the g loop in Conv2D.forward

* ereased stupid line

* nothing

* no loops in Conv2D forward

* Conv2D backprop improved

Co-authored-by: Marcel Bischoff <marcel@Marcels-iMac.local>
 2020-10-31 09:08:32 -07:00
George Hotz 053c319708 optional look at cat 2020-10-30 09:30:15 -07:00
George Hotz 2c7e75d733
group conv: forward pass works (#34) 
* forward pass works

* got the backward pass

* okay, it's now a coho
 2020-10-30 09:19:20 -07:00
George Hotz 339a35b081 div needs help 2020-10-30 08:32:16 -07:00
George Hotz 71aedc2309 it thinks it's washer. it's cat. bad net. you do bad. 2020-10-30 08:28:05 -07:00
George Hotz c14473f87d unit test for batchnorm2d 2020-10-30 08:19:58 -07:00
George Hotz 843b1cb7d6 only warn once 2020-10-30 08:04:48 -07:00