Commit Graph

77 Commits

Author SHA1 Message Date
Will Brickner 03af140e12
Implement Quiet Softmax (`Attention Is Off By One`) (#692)
* Added quiet_softmax

* Undid bad formatting

---------

Co-authored-by: Dilshod Tadjibaev <939125+antimora@users.noreply.github.com>
2023-11-30 12:58:30 -05:00
Dilshod Tadjibaev fed618b265
Add Cos, Exp, Gelu, Log and Neg unary ONNX ops (#1013)
* Add ONNX Cos OP

* Add ONNX Exp OP

* Add ONNX Gelu OP

* Add ONNX Log OP

* Allow approx_constant clippy rule in generated model code

* Add ONNX Neg OP

* Fix tests with custom imports for unary nodes

* Add scalar tests for Sqrt
2023-11-29 14:57:52 -06:00
David Chavez fda9e5d388
feature(tensor): Add triu and tril ops (#1002) 2023-11-29 09:39:42 -05:00
Edmondo Porcu b86bc58761
Implementing ONNX support for sqrt (#991)
* Implementing ONNX support for sqrt

* Formatting, my bad

* Implementing feedback from pull request

* Fixing codegen

* Fixing tests

* Fixing tests
2023-11-22 16:12:35 -06:00
Dilshod Tadjibaev 445f41bb7b
Support count_include_pad attr in avg_pool2d ONNX (#978)
Fixes #636
2023-11-21 12:21:12 -06:00
Luni-4 445603401d
ci/Check dependencies (#895) 2023-11-19 10:35:03 -05:00
Luni-4 ec9df53d4c
ci/Fix `cargo clippy` action (#942) 2023-11-16 19:35:38 -05:00
Zsombor 4fc0c27e31
Implement tensor.recip() function to calculate elementwise reciprocals (#953) 2023-11-15 09:17:32 -05:00
Ariel Cohen cb4c23b39d
add gather (#947) 2023-11-13 08:12:00 -05:00
Nathaniel Simard 96524d40a1
[Breaking] Refactor Backend Names (#904) 2023-10-29 18:27:49 -04:00
Nathaniel Simard 233922d60c
Chore: Bump version for next release (#900) 2023-10-24 19:31:13 -04:00
Luni-4 9add42442f
Generalize model usage in burn-import README (#889) 2023-10-24 09:53:42 -04:00
AuruTus 84e74df3b9
Feat/onnx import erf (#859) 2023-10-13 09:59:44 -04:00
Dilshod Tadjibaev 097fd956d0
Upgrade dependency versions (#854)
This updates dependencies including tch to 0.14.0, which uses Torch 2.1.
2023-10-09 14:29:44 -04:00
Dilshod Tadjibaev e2a17e4295
Add image classification web demo with WebGPU, CPU backends (#840) 2023-10-05 10:29:13 -04:00
Dilshod Tadjibaev 3ac9abbacb
Allow custom file location for loading record for ONNX model (#824) 2023-09-22 08:40:36 -04:00
Dilshod Tadjibaev bdcd5873c5
Support no-std for models generated by burn-import (#822) 2023-09-21 11:43:07 -04:00
Dilshod Tadjibaev 35ca839e18
Add support for different record types in ONNX (#816) 2023-09-21 09:06:57 -04:00
Nathaniel Simard 251ec00070
Fix clamp API (#799) 2023-09-12 14:20:11 -04:00
Nathaniel Simard af0be5cfeb
Chore: bump version (#777) 2023-09-06 12:15:13 -04:00
Damien Elmes 3669d2a6d4
Migrate from log4rs to tracing (#739) 2023-08-31 21:07:26 -04:00
Nathaniel Simard c95b34c511
Book: backend extension + custom wgpu kernel (#728) 2023-08-31 09:55:43 -04:00
Louis Fortier-Dubois 7c34e21424
Perf/tensor ops/more tests (#718) 2023-08-30 09:08:18 -04:00
Dilshod Tadjibaev bb71f17409
Update burn-import README (#727) 2023-08-30 08:20:53 -04:00
Dilshod Tadjibaev 13342b45d1
Account for additional cases in Linear node conversion (#709) 2023-08-28 16:24:59 -04:00
Dilshod Tadjibaev 03ee987cda
ONNX Clip Operator (#696) 2023-08-26 17:58:51 -04:00
AuruTus 3fdfeeae2b
Feat/onnx import tanh (#683) 2023-08-25 14:57:07 -04:00
Louis Fortier-Dubois bc27a87e9d
new clippy stuff (#687) 2023-08-24 13:20:58 -04:00
Will Brickner d1a708e317
Made TchTensor fields public (#677) 2023-08-22 21:37:58 -04:00
Caio Piccirillo 2fefc82099
Dilation maxpool (#668) 2023-08-21 14:14:25 -04:00
Dilshod Tadjibaev b06fe660e8
Onnx tests and bug fixes - BatchNorm, Identity, Relu, Sigmoid and Transpose (#661) 2023-08-21 11:39:59 -04:00
Dilshod Tadjibaev feabc14d2e
Fix Conv1d kernel for ONNX importing (#658) 2023-08-17 08:41:22 -04:00
Dilshod Tadjibaev 0a1ba971cc
Fix bug related to calculation of flatten op output dim (#652)
Fixes #646
2023-08-16 18:22:23 -04:00
Dilshod Tadjibaev 8448611908
License fixes (#648) 2023-08-16 12:45:35 -04:00
Dilshod Tadjibaev 3264b1007c
Changes and fixes required for squeezenet1.1-opset7 onnx (#639) 2023-08-15 15:37:52 -04:00
Louis Fortier-Dubois d659f11639
Perf/wgpu/autotune (#609) 2023-08-15 11:26:00 -04:00
Mitchell Robert Vollger 434975c410
Add conv1d to burn-import (onnx) (#638) 2023-08-14 17:48:02 -04:00
Dilshod Tadjibaev b79fa0748f
Fix bugs with MaxPool2d in ONNX conversation (#623) 2023-08-10 09:35:15 -04:00
Dilshod Tadjibaev c1ba355c26
Softmax OP ONNX support (#620) 2023-08-09 16:48:45 -04:00
Dilshod Tadjibaev 894783f08d
Global avg pool (#611) 2023-08-09 16:15:33 -04:00
Dilshod Tadjibaev 716b7569de
Add Dropout ONNX OP and other refactoring (#606) 2023-08-09 09:53:20 -04:00
Caio Piccirillo 1d3bbaab13
Typos (#608) 2023-08-08 17:57:51 -04:00
Gadersd ed255c5561
Use buffered io for massive performance gains when loading and saving… (#593) 2023-08-06 12:56:27 -04:00
Dilshod Tadjibaev 5cc32cc8cb
Remove topological sort of nodes (#589)
ONNX nodes should come as topologically sorted, so we are removing it instead, we are making sure the nodes are topologically sorted.
2023-08-06 10:51:28 -04:00
Dilshod Tadjibaev 1554a3c898
Full support for ONNX scalar operators and Constants (#578) 2023-08-04 16:51:51 -04:00
Nathaniel Simard 8436d4ff66
Feat/tensor/adaptive avg pool2d (#572) 2023-08-04 10:23:59 -04:00
Dilshod Tadjibaev f453f47efb
ONNX Concat op (#561) 2023-07-31 15:46:37 -04:00
Dilshod Tadjibaev 6b459fde82
Update Conv2d source generation to support all attributes (#558) 2023-07-28 12:52:23 -04:00
Luni-4 67ce5ec62c
Implement binary operators for `burn-import` (#532) 2023-07-27 19:37:10 -04:00
Luni-4 2148ac43a4
burn-import: Add transpose operator (#546) 2023-07-27 19:34:43 -04:00