Commit Graph

Select branches

Hide Pull Requests

geohot/updates

gh-pages

master

remove_anchors

#1

#2

#3

v0.10.0

v0.10.1

v0.10.2

v0.10.3

v0.11.0

v0.12.0

v0.13.0

v0.4.0

v0.5.0

v0.6.0

v0.7.0

v0.8.0

v0.9.0

v0.9.1

v0.9.2

Mono Color

• 81a11d891d ops rdna geohot 2023-05-21 11:45:38 -07:00
• ed038ba129 Contract float4 ALU operations (#780) George HotzandGitHub 2023-05-16 19:03:49 -07:00
• 90fff82c8a Rdna (#776) George HotzandGitHub 2023-05-16 05:33:57 -07:00
• 89b8b39d9c fix mypy geohot 2023-05-13 21:25:36 -07:00
• e0b2035023 fast imagenet eval, gets 76.14% across the set geohot 2023-05-13 12:39:01 -07:00
• c552f6f92b Inference test: add tests for ResNet50 (#773) Jacky LeeandGitHub 2023-05-13 21:18:15 -07:00
• e5b4b36cba add std to tensor.py (#767) Rabia Eda YılmazandGitHub 2023-05-13 22:20:44 +03:00
• b705510d5c getting 77% on imagenet eval geohot 2023-05-13 07:46:27 -07:00
• 810f03dafa conv3d + unet3d (#772) George HotzandGitHub 2023-05-12 13:54:07 -07:00
• 46d419060b start on mlperf models geohot 2023-05-10 16:30:49 -07:00
• d13629cb26 ResNet: match implementation with Nvidia and PyTorch (#770) Jacky LeeandGitHub 2023-05-10 09:01:22 -07:00
• b80cf9220c Statistics test: check if distributions match torch (#769) Jacky LeeandGitHub 2023-05-07 21:43:23 -07:00
• cb7c22beeb fix mypy geohot 2023-05-06 19:18:54 +00:00
• 5190037cbc rocm: disassembler for shader geohot 2023-05-06 18:56:09 +00:00
• 7fbf96b992 jit: TODO, use abstractions geohot 2023-05-05 22:51:30 -07:00
• 0cd3feb452 jit oops. should add that to commit tests geohot 2023-05-05 22:01:13 -07:00
• 5b2ae262db assertions for jit geohot 2023-05-05 21:56:32 -07:00
• 42256c0d9d rocm sniffer dumps code geohot 2023-05-05 18:36:53 +00:00
• 81aa3e546b exclude GPU on tiny (#766) George HotzandGitHub 2023-05-05 10:07:23 -07:00
• f2a964f447 nocopy (#764) George HotzandGitHub 2023-05-05 09:32:06 -07:00
• 466ffeb04f fast cifar on AMD geohot 2023-05-05 02:10:50 +00:00
• 3a2011ab2d rocm sniffer geohot 2023-05-04 22:22:39 +00:00
• a55c4f5000 better rocm build scripts geohot 2023-05-04 09:14:05 +00:00
• 987b1aaf96 rocm build scripts geohot 2023-05-04 08:45:23 +00:00
• f28df9900f multidevice works (#763) George HotzandGitHub 2023-05-04 01:04:58 -07:00
• 4f6d674ec0 use CPU tests in pre-commit geohot 2023-05-03 19:46:16 +00:00
• ed33a89d52 no werror in archprobe geohot 2023-05-03 19:34:17 +00:00
• 7ecf4dff68 multi cl_queue (#762) George HotzandGitHub 2023-05-03 12:15:28 -07:00
• 7757f5fed2 Fixed package description (#761) Rylan JusticeandGitHub 2023-05-03 13:21:05 -04:00
• 3b933b0a2f rocm setup script geohot 2023-05-03 16:01:17 +00:00
• 9628a3f190 Updated LICENSE year (#760) Rylan JusticeandGitHub 2023-05-01 18:35:23 -04:00
• 0b9d4126d0 Add Tensor.stack() and Tensor.repeat() (...trying to make einops work with tinygrad) (#758) JoqsanandGitHub 2023-05-01 19:37:46 +03:00
• 59d0d168cd FLOAT16 off works geohot 2023-04-19 15:34:56 -07:00
• 3d15769a8f 50 TFLOPS cuda matmul geohot 2023-04-19 14:38:24 -07:00
• 03b38864db fix batchnorm at training (#753) George HotzandGitHub 2023-04-19 08:01:04 -07:00
• 1aa0648d6a fix path linter issue geohot 2023-04-18 19:17:41 -07:00
• cbe2564b7b oops, no hip yet geohot 2023-04-18 19:10:36 -07:00
• e4db0c820f hlb_cifar10 init from torch weights geohot 2023-04-18 19:09:13 -07:00
• a6b9733256 GB/s can be higher geohot 2023-04-18 17:51:03 -07:00
• 9fb3f9ace3 Revert "move t.grad realize on SGD" geohot 2023-04-18 17:50:08 -07:00
• e93e04ed6e Revert "huh...this is faster" geohot 2023-04-18 17:50:07 -07:00
• aedd4685fa huh...this is faster geohot 2023-04-18 17:36:31 -07:00
• dbc99c243b why did that test break? geohot 2023-04-18 17:08:38 -07:00
• ccdc0290d6 move t.grad realize on SGD geohot 2023-04-18 16:47:51 -07:00
• 8b7ecd63bb Remove Zeroview (#748) George HotzandGitHub 2023-04-17 08:21:46 -07:00
• 4e17d27d09 Fix cuda errors when running llama example (#749) Jan Henrik HøilandandGitHub 2023-04-16 22:52:10 +02:00
• 0b5a0b9ba4 winograd comment geohot 2023-04-16 03:36:51 -07:00
• 8b777af571 metal_conv gets over 10.4 TFLOPS... geohot 2023-04-15 03:31:22 -07:00
• d66e682205 metal matmul from tcores branch geohot 2023-04-14 23:29:29 -07:00
• 732884653c osx in hlb_cifar10_torch geohot 2023-04-14 13:12:08 -07:00
• 17e37157b6 fix backward convs (#746) George HotzandGitHub 2023-04-14 10:42:11 -07:00
• f7f416d6f4 back to 6 for test_fold_conv_sgd geohot 2023-04-14 07:34:00 -07:00
• 133521e730 relu UnaryOp is back geohot 2023-04-14 07:12:53 -07:00
• 584ee6f616 don't graph consts geohot 2023-04-14 03:32:20 -07:00
• 9a39ebefde hlb_cifar10_torch gets 80% geohot 2023-04-14 02:47:03 -07:00
• 552a048a33 make maximum split the grad like torch when equal (#738) worldwalker2000andGitHub 2023-04-14 00:17:46 -07:00
• 06ed958abd Fix train_resnet example (#744) Jacky LeeandGitHub 2023-04-12 04:18:39 -04:00
• 70b9072663 add Pad onnx operator and rework _padding (#740) SohaibandGitHub 2023-04-06 11:37:36 +00:00
• 8e40ff8c8d Do not specify errors when trying to load devices. (#741) jintuzhangandGitHub 2023-04-06 19:35:36 +08:00
• 7a45b989a1 Device: make GPU default and METAL/CUDA if possible (#732) Jacky LeeandGitHub 2023-04-03 21:11:52 -07:00
• 94e2c49c35 test_cacheline_size that works in both places geohot 2023-03-30 06:47:20 +04:00
• b05c2828f7 better cacheline test geohot 2023-03-30 06:08:54 +04:00
• 76db1af6fc better archprobe geohot 2023-03-30 05:52:00 +04:00
• 1240c12ac5 download cifar to datasets dir geohot 2023-03-29 12:25:42 +04:00
• e5f430d8c6 Device: move below LazyBuffer (#733) Jacky LeeandGitHub 2023-03-28 23:35:11 -07:00
• b99798f08e acc function not needed geohot 2023-03-29 08:02:11 +04:00
• 20894991ed good changes from the M1 Tensor Core project (#730) George HotzandGitHub 2023-03-29 05:11:02 +04:00
• 156640e90d Permute examples (#731) Jacky LeeandGitHub 2023-03-28 18:07:06 -07:00
• 39d6e1525f Added activation ops + tests (#729) Andre SlavescuandGitHub 2023-03-28 05:17:53 -04:00
• fa5516dda0 fix lint, installed pre-commit on now computer geohot 2023-03-24 11:15:59 -07:00
• ebc4ad6223 color the jit nicer geohot 2023-03-24 10:54:20 -07:00
• 23f88fb026 synchronize for honest speed compare geohot 2023-03-24 10:24:27 -07:00
• 1cb5b2d015 test_enet_se geohot 2023-03-24 10:04:30 -07:00
• fafe8e9ce2 casting: support all backends and implement half (#726) Jacky LeeandGitHub 2023-03-24 09:58:03 -07:00
• e88b9bfe1e print gflops avg with DEBUG=2 geohot 2023-03-23 16:07:08 -07:00
• de04208247 hotcast bug fix George Hotz 2023-03-23 11:49:47 -07:00
• e009b6f341 Add tests for casting (#724) Jacky LeeandGitHub 2023-03-23 08:02:52 -07:00
• 68e45fca18 metal_matmul: bw and torch sync geohot 2023-03-23 08:02:04 -07:00
• bd6c3c31a9 compare to torch geohot 2023-03-22 23:58:37 -07:00
• c3a3db75c7 fix metal matmul example geohot 2023-03-22 23:42:51 -07:00
• f5aea472a3 latest torch and onnx should be fine geohot 2023-03-22 23:33:50 -07:00
• 51e19ac25c OPTLOCAL=2 makes stable diffusion a usable speed after the cache builds geohot 2023-03-22 19:19:11 -07:00
• 2e18469fd4 clean up display name geohot 2023-03-22 18:32:05 -07:00
• b12b60af20 fix binop, other tests failure (#723) George HotzandGitHub 2023-03-22 18:15:07 -07:00
• d6f4219952 LayerNorm2d for 2 lines geohot 2023-03-20 16:58:43 -07:00
• 128ca160ac lazy: remove required device geohot 2023-03-20 16:31:45 -07:00
• 120d7072bd indexing merge almost works geohot 2023-03-20 16:17:07 -07:00
• 06abbbfe7c remove the stupid register class (#721) George HotzandGitHub 2023-03-20 15:45:12 -07:00
• 30b795874a remove RMSprop, nobody uses it anymore geohot 2023-03-20 12:31:34 -07:00
• 25287a974e types (#720) George HotzandGitHub 2023-03-20 12:31:02 -07:00
• 9b314c6342 factor uops transformers into functions geohot 2023-03-20 08:19:48 -07:00
• 623fb1ef28 do test_conv_with_bn test geohot 2023-03-19 23:53:56 -07:00
• 5495c7d64e linearizer! (#714) George HotzandGitHub 2023-03-19 23:43:49 -07:00
• b629fd4cd8 add AdamW optimizer (#716) Cyril RoumégousandGitHub 2023-03-19 20:51:06 +01:00
• 1012b68f7e finally, some speedups geohot 2023-03-18 18:14:41 -07:00
• 902906f909 Fix constant folding (#713) George HotzandGitHub 2023-03-18 17:52:46 -07:00
• 73bd0b217b add int64 as supported dtype from numpy (#699) Fernando VidalandGitHub 2023-03-18 17:15:04 -07:00
• f355b02987 remove comments and reorder geohot 2023-03-18 14:48:39 -07:00
• f5467cfedc Devicebufferless (#708) George HotzandGitHub 2023-03-18 14:40:23 -07:00
• 26a3888ab8 Fix llama 13B RAM usage (#710) KirillandGitHub 2023-03-18 23:50:09 +03:00