Web Machine Learning Working Group - Activity in 2024
No matches.
TR publications
-
1 document published.
-
Web Neural Network API published as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
3 participants joined.
-
Joshua Lochner (Hugging Face) joined.
-
Jiajia Qin (Microsoft Corporation) joined.
-
Yang Gu (Microsoft Corporation) joined.
-
-
1 participant left.
-
Daniel LaLiberte (Google LLC) left.
-
GitHub
webmachinelearning/webnn
-
2 issues opened.
-
Allow NaN and infinite values for float attributes webmachinelearning/webnn#800, by @lisa0314.
-
Uncaught (in promise) NotSupportedError: Failed to execute 'createContext' on 'ML': DirectML: WebNN is blocklisted for NPU. webmachinelearning/webnn#798, by @sandeepks1.
-
-
2 issues closed.
-
Remove `MLContext.compute()` webmachinelearning/webnn#791.
-
2 issues received 4 new comments.
-
Allow NaN and infinite values for float attributes webmachinelearning/webnn#800 (3 comments by @fdwr, @huningxin, @inexorabletash)
-
Uncaught (in promise) NotSupportedError: Failed to execute 'createContext' on 'ML': DirectML: WebNN is blocklisted for NPU. webmachinelearning/webnn#798 (1 comment by @huningxin)
-
-
3 pull requests opened.
-
Reject promises for in-progress operations on an MLTensor if it is destroyed webmachinelearning/webnn#799, by @a-sully.
-
Editorial: Various style and wording tweaks webmachinelearning/webnn#797, by @inexorabletash.
-
Complain about RFC2119 terms, and fix usage webmachinelearning/webnn#796, by @inexorabletash.
-
-
2 pull requests merged.
-
Complain about RFC2119 terms, and fix usage webmachinelearning/webnn#796.
-
Remove the MLContext.compute() method webmachinelearning/webnn#795.
-
-
5 pull requests received 24 new comments.
-
Complain about RFC2119 terms, and fix usage webmachinelearning/webnn#796 (11 comments by @fdwr, @inexorabletash, @zolkis)
-
Editorial: Various style and wording tweaks webmachinelearning/webnn#797 (8 comments by @huningxin, @inexorabletash)
-
Reject promises for in-progress operations on an MLTensor if it is destroyed webmachinelearning/webnn#799 (2 comments by @a-sully, @inexorabletash)
-
Rename pool2d MLRoundingType - Simplify the operand layout support of conv2d and pooling 2d operations webmachinelearning/webnn#770 (1 comment by @jlewis-austin)
-
TR publications
-
6 documents published.
-
Web Neural Network API published 6 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
-
1 meeting.
Participants
-
1 participant joined.
-
Kaushik Satpathy (Yahoo Holdings Inc.) joined.
-
-
3 participants left.
-
Joshua Lochner (Hugging Face) left.
-
Yang Gu (Microsoft Corporation) left.
-
Jason Stephen (Microsoft Corporation) left.
-
-
1 organization joined.
-
Yahoo Holdings Inc. joined.
-
GitHub
webmachinelearning/webnn
-
9 issues opened.
-
Disallow operations on scalar tensors that are no-ops webmachinelearning/webnn#794, by @reillyeon.
-
Create new objects with a Realm webmachinelearning/webnn#793, by @a-sully.
-
Define equality for non-trivial data types webmachinelearning/webnn#792, by @a-sully.
-
Remove `MLContext.compute()` webmachinelearning/webnn#791, by @a-sully.
-
`MLGraphBuilder.constant()` should take an `AllowSharedBufferSource` webmachinelearning/webnn#788, by @a-sully.
-
Proposed style/wording tweaks webmachinelearning/webnn#783, by @inexorabletash.
-
Idea for Discussion: Device Memory Management Primitives webmachinelearning/webnn#780, by @wacky6.
-
Support block-wise quantization webmachinelearning/webnn#779, by @huningxin.
-
Proposal: Report non-fatal errors from the WebNN timeline webmachinelearning/webnn#778, by @a-sully.
-
-
5 issues closed.
-
`MLGraphBuilder.constant()` should take an `AllowSharedBufferSource` webmachinelearning/webnn#788.
-
API lacks handling for async ML device errors on the context webmachinelearning/webnn#477.
-
Don't transfer input ArrayBuffers webmachinelearning/webnn#566.
-
Fix the build webmachinelearning/webnn#771.
-
WebNN should support int8 quantized models webmachinelearning/webnn#128.
-
-
22 issues received 49 new comments.
-
Proposal: Report non-fatal errors from the WebNN timeline webmachinelearning/webnn#778 (9 comments by @a-sully, @bbernhar, @RafaelCintron)
-
Idea for Discussion: Device Memory Management Primitives webmachinelearning/webnn#780 (6 comments by @bbernhar, @reillyeon, @wacky6)
-
Disallow operations on scalar tensors that are no-ops webmachinelearning/webnn#794 (5 comments by @fdwr, @inexorabletash, @reillyeon)
-
Create new objects with a Realm webmachinelearning/webnn#793 (2 comments by @a-sully, @inexorabletash)
-
Request the decomposition for gatherElements, scatterElements and scatterND webmachinelearning/webnn#767 (2 comments by @fujunwei, @huningxin)
-
API lacks handling for async ML device errors on the context webmachinelearning/webnn#477 (2 comments by @anssiko, @bbernhar)
-
Request standards positions from Mozilla and WebKit webmachinelearning/webnn#763 (2 comments by @inexorabletash, @reillyeon)
-
Consider removing `lstm` and `gru` operators webmachinelearning/webnn#689 (2 comments by @huningxin, @philloooo)
-
heads up re: proposal to add Float16Array to JavaScript webmachinelearning/webnn#373 (1 comment by @anssiko)
-
Define equality for non-trivial data types webmachinelearning/webnn#792 (1 comment by @inexorabletash)
-
`MLGraphBuilder.constant()` should take an `AllowSharedBufferSource` webmachinelearning/webnn#788 (1 comment by @huningxin)
-
Support coordinate transformation modes for Resample2d webmachinelearning/webnn#270 (1 comment by @fdwr)
-
MLContextOptions.deviceType seems unnecessary outside of conformance testing webmachinelearning/webnn#749 (1 comment by @anssiko)
-
Consider adding int64/uint64 data type support for some reduce operators webmachinelearning/webnn#694 (1 comment by @anssiko)
-
webmachinelearning/webnn#128 (1 comment by @reillyeon)
-
-
10 pull requests opened.
-
Remove the MLContext.compute() method webmachinelearning/webnn#795, by @a-sully.
-
Use `AllowSharedBufferSource` for `MLGraphBuilder.constant()` webmachinelearning/webnn#790, by @huningxin.
-
Include output operand type/rank in op constraint tables webmachinelearning/webnn#789, by @inexorabletash.
-
Specify MLTensor webmachinelearning/webnn#787, by @a-sully.
-
Remove descriptions of outdated MLContext requirements webmachinelearning/webnn#786, by @a-sully.
-
Editorial: Link to error handling proposal from the MLTensor explainer webmachinelearning/webnn#785, by @a-sully.
-
Add device selection explainer webmachinelearning/webnn#784, by @zolkis.
-
Wording: Fix some minor typos and grammar glitches webmachinelearning/webnn#782, by @inexorabletash.
-
Add deprecation warning for compute() -> dispatch() webmachinelearning/webnn#781, by @anssiko.
-
Wording: Use newfangled multiple assignment from lists webmachinelearning/webnn#777, by @inexorabletash.
-
-
8 pull requests merged.
-
Use `AllowSharedBufferSource` for `MLGraphBuilder.constant()` webmachinelearning/webnn#790.
-
Specify MLTensor webmachinelearning/webnn#787.
-
Editorial: Link to error handling proposal from the MLTensor explainer webmachinelearning/webnn#785.
-
Include output operand type/rank in op constraint tables webmachinelearning/webnn#789.
-
Make operand data type and rank validation table-driven webmachinelearning/webnn#657.
-
Wording: Fix some minor typos and grammar glitches webmachinelearning/webnn#782.
-
Add deprecation warning for compute() -> dispatch() webmachinelearning/webnn#781.
-
Wording: Use newfangled multiple assignment from lists webmachinelearning/webnn#777.
-
-
2 pull requests closed.
-
Remove descriptions of outdated MLContext requirements webmachinelearning/webnn#786.
-
Copy input ArrayBuffers webmachinelearning/webnn#700.
-
-
10 pull requests received 184 new comments.
-
Specify MLTensor webmachinelearning/webnn#787 (91 comments by @a-sully, @fdwr, @huningxin, @inexorabletash, @reillyeon)
-
Make operand data type and rank validation table-driven webmachinelearning/webnn#657 (35 comments by @fdwr, @huningxin, @inexorabletash)
-
Use `AllowSharedBufferSource` for `MLGraphBuilder.constant()` webmachinelearning/webnn#790 (17 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Include output operand type/rank in op constraint tables webmachinelearning/webnn#789 (16 comments by @fdwr, @huningxin, @inexorabletash)
-
Remove the MLContext.compute() method webmachinelearning/webnn#795 (11 comments by @a-sully, @huningxin)
-
Rename pool2d MLRoundingType - Simplify the operand layout support of conv2d and pooling 2d operations webmachinelearning/webnn#770 (6 comments by @anssiko, @fdwr, @huningxin, @inexorabletash)
-
Remove descriptions of outdated MLContext requirements webmachinelearning/webnn#786 (5 comments by @a-sully, @huningxin)
-
Wording: Fix some minor typos and grammar glitches webmachinelearning/webnn#782 (1 comment by @inexorabletash)
-
Past meetings
Participants
-
6 participants joined.
-
Christian Liebel (Thinktecture) joined.
-
Islam El-Ashi (Dfinity Stiftung) joined.
-
Domenic Denicola (Google LLC) joined.
-
Sohum Chatterjee (Microsoft Corporation) joined.
-
Talha Gorsi (Qualcomm) joined.
-
Robert Simpson (Qualcomm) joined.
-
-
2 organizations joined.
-
Dfinity Stiftung joined.
-
Qualcomm joined.
-
-
1 organization left.
-
Wiley left.
-
GitHub
webmachinelearning/webnn
-
7 issues opened.
-
The split's opSupportLimits should be of `MLSplitSupportLimits` webmachinelearning/webnn#775, by @BruceDai.
-
Support `reverse` operator webmachinelearning/webnn#773, by @huningxin.
-
Support strides option for `slice` operator webmachinelearning/webnn#772, by @huningxin.
-
Fix the build webmachinelearning/webnn#771, by @anssiko.
-
Consider removing or redesigning the `triangular` operator webmachinelearning/webnn#768, by @a-sully.
-
Request the decomposition for gatherElements, scatterElements and scatterND webmachinelearning/webnn#767, by @fujunwei.
-
Allow setting label for constant operand webmachinelearning/webnn#766, by @huningxin.
-
-
5 issues closed.
-
Reconsider `MLOperand` methods webmachinelearning/webnn#666.
-
The split's opSupportLimits should be of `MLSplitSupportLimits` webmachinelearning/webnn#775.
-
Support for LocalResponseNormalization (LRN) operation webmachinelearning/webnn#228.
-
Allow to hint number of threads for CPU MLContext webmachinelearning/webnn#436.
-
Google Chrome Feedback on WebNN: aiming for broad device coverage and maintainability webmachinelearning/webnn#453.
-
-
1 issue reopened.
-
Define the maximum number of operand dimensions (maximum rank) webmachinelearning/webnn#456, by @huningxin.
-
-
14 issues received 26 new comments.
-
Support for LocalResponseNormalization (LRN) operation webmachinelearning/webnn#228 (5 comments by @a-sully, @fdwr, @huningxin)
-
Support strides option for `slice` operator webmachinelearning/webnn#772 (3 comments by @a-sully, @fdwr, @huningxin)
-
MLContextOptions.deviceType seems unnecessary outside of conformance testing webmachinelearning/webnn#749 (3 comments by @mwyrzykowski, @zolkis)
-
Consider removing or redesigning the `triangular` operator webmachinelearning/webnn#768 (2 comments by @a-sully, @fdwr)
-
Allow to hint number of threads for CPU MLContext webmachinelearning/webnn#436 (2 comments by @a-sully, @huningxin)
-
Bitwise operators and logical operators naming (rename not to logicalNot) webmachinelearning/webnn#496 (2 comments by @fdwr, @mwyrzykowski)
-
Request the decomposition for gatherElements, scatterElements and scatterND webmachinelearning/webnn#767 (1 comment by @huningxin)
-
Google Chrome Feedback on WebNN: aiming for broad device coverage and maintainability webmachinelearning/webnn#453 (1 comment by @a-sully)
-
Define the maximum number of operand dimensions (maximum rank) webmachinelearning/webnn#456 (1 comment by @huningxin)
-
Allow checking whether operators/types are supported for a backend before creating a graph webmachinelearning/webnn#463 (1 comment by @huningxin)
-
Consider separately pass each activation and direction's weights&bias for lstm and gru webmachinelearning/webnn#751 (1 comment by @philloooo)
-
-
5 pull requests opened.
-
Bugfix: Fix the error of opSupportLimits for split op webmachinelearning/webnn#776, by @BruceDai.
-
Convert MLOperand methods into readonly attributes webmachinelearning/webnn#774, by @a-sully.
-
Rename pool2d MLRoundingType - Simplify the operand layout support of conv2d and pooling 2d operations webmachinelearning/webnn#770, by @fdwr.
-
Add Implementation Status to metadata webmachinelearning/webnn#769, by @anssiko.
-
Add resource contention considerations webmachinelearning/webnn#765, by @anssiko.
-
-
5 pull requests merged.
-
Convert MLOperand methods into readonly attributes webmachinelearning/webnn#774.
-
Bugfix: Fix the error of opSupportLimits for split op webmachinelearning/webnn#776.
-
Add MLTensor explainer webmachinelearning/webnn#754.
-
Add resource contention considerations webmachinelearning/webnn#765.
-
Add Implementation Status to metadata webmachinelearning/webnn#769.
-
-
5 pull requests received 42 new comments.
-
Rename pool2d MLRoundingType - Simplify the operand layout support of conv2d and pooling 2d operations webmachinelearning/webnn#770 (9 comments by @a-sully, @anssiko, @fdwr, @huningxin, @inexorabletash)
-
Convert MLOperand methods into readonly attributes webmachinelearning/webnn#774 (3 comments by @a-sully, @fdwr)
TR publications
-
3 documents published.
-
Web Neural Network API published 3 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
2 participants joined.
-
Erik Anderson (Microsoft Corporation) joined.
-
Tarek Ziade (Hugging Face) joined.
-
-
2 participants left.
-
Geoff Gustafson (Intel Corporation) left.
-
Tzviya Siegman (Wiley) left.
-
-
1 organization joined.
-
Mozilla Foundation joined.
-
GitHub
webmachinelearning/webnn
-
4 issues opened.
-
Request standards positions from Mozilla and WebKit webmachinelearning/webnn#763, by @reillyeon.
-
Support building graphs from `MLTensor` containing constants webmachinelearning/webnn#760, by @bbernhar.
-
MLOpSupportLimits should be opt-in with base functionality webmachinelearning/webnn#759, by @mwyrzykowski.
-
Make `MLOperandDescriptor.shape` a `required` dictionary member webmachinelearning/webnn#758, by @a-sully.
-
-
8 issues closed.
-
Do we need an `MLConstantOperand`? webmachinelearning/webnn#668.
-
Executing operations webmachinelearning/webnn#11.
-
Need to restrict the value of alpha to be positive for elu operation webmachinelearning/webnn#383.
-
Make `MLOperandDescriptor.shape` a `required` dictionary member webmachinelearning/webnn#758.
-
Simplify `resample2d` op webmachinelearning/webnn#474.
-
[MLBuffer] Support interop with WebGPU webmachinelearning/webnn#688.
-
Rename `MLOperandDescriptor`'s `dimensions` to `shape` webmachinelearning/webnn#669.
-
Rename MLBuffer => MLTensor webmachinelearning/webnn#753.
-
-
2 issues reopened.
-
Simplify the operand layout support of conv2d and pooling 2d operations webmachinelearning/webnn#324, by @inexorabletash.
-
Add "implementation consideration" about how out-of-bound indices of Gather/Scatter should be handled webmachinelearning/webnn#486, by @huningxin.
-
-
23 issues received 46 new comments.
-
Make `MLOperandDescriptor.shape` a `required` dictionary member webmachinelearning/webnn#758 (4 comments by @a-sully, @fdwr, @huningxin)
-
Simplify the operand layout support of conv2d and pooling 2d operations webmachinelearning/webnn#324 (4 comments by @fdwr, @huningxin, @inexorabletash)
-
Support building graphs from `MLTensor` containing constants webmachinelearning/webnn#760 (4 comments by @bbernhar, @fdwr, @mmccool)
-
Need to restrict the value of alpha to be positive for elu operation webmachinelearning/webnn#383 (3 comments by @fdwr, @huningxin, @inexorabletash)
-
Add "implementation consideration" about how out-of-bound indices of Gather/Scatter should be handled webmachinelearning/webnn#486 (3 comments by @a-sully, @huningxin)
-
Consider separately pass each activation and direction's weights&bias for lstm and gru webmachinelearning/webnn#751 (3 comments by @fdwr, @huningxin, @shiyi9801)
-
Rename `MLOperandDescriptor`'s `dimensions` to `shape` webmachinelearning/webnn#669 (2 comments by @fdwr, @zolkis)
-
webmachinelearning/webnn#11 (1 comment by @inexorabletash)
-
Behavior for numeric overflows/underflows/etc in algorithms webmachinelearning/webnn#636 (1 comment by @inexorabletash)
-
webmachinelearning/webnn#128 (1 comment by @reillyeon)
-
Should `scale` and `bias` be required inputs for `batchNormalization` op? webmachinelearning/webnn#481 (1 comment by @huningxin)
-
Request standards positions from Mozilla and WebKit webmachinelearning/webnn#763 (1 comment by @anssiko)
-
Consider dropping the support of uint64/int64 data type for some operators webmachinelearning/webnn#654 (1 comment by @fdwr)
-
-
2 pull requests opened.
-
Make MLOperandDescriptor.shape required webmachinelearning/webnn#764, by @a-sully.
-
Bugfix: Rename member `output` to `outputs` of `MLLstmCellSupportLimits` webmachinelearning/webnn#757, by @huningxin.
-
-
3 pull requests merged.
-
Make MLOperandDescriptor.shape required webmachinelearning/webnn#764.
-
Rename MLOperandDescriptor's "dimensions" key to "shape" webmachinelearning/webnn#676.
-
Bugfix: Rename member `output` to `outputs` of `MLLstmCellSupportLimits` webmachinelearning/webnn#757.
-
-
2 pull requests closed.
-
Introduce MLConstantOperand webmachinelearning/webnn#747.
-
Add MLBuffer exploration doc webmachinelearning/webnn#541.
-
-
7 pull requests received 115 new comments.
-
Add MLTensor explainer webmachinelearning/webnn#754 (86 comments by @a-sully, @anssiko, @bbernhar, @bjjones, @domenic, @fdwr, @Kangz, @RafaelCintron)
-
Rename MLOperandDescriptor's "dimensions" key to "shape" webmachinelearning/webnn#676 (20 comments by @a-sully, @fdwr, @huningxin, @reillyeon, @zolkis)
-
Bugfix: Add missing 64-bit integers support for some reduction operators webmachinelearning/webnn#695 (3 comments by @a-sully, @fdwr, @huningxin)
-
Bugfix: Rename member `output` to `outputs` of `MLLstmCellSupportLimits` webmachinelearning/webnn#757 (3 comments by @fdwr, @huningxin, @shiyi9801)
-
TR publications
-
3 documents published.
-
Web Neural Network API published 3 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
1 participant left.
-
Barbara Hochgesang (Intel Corporation) left.
-
GitHub
webmachinelearning/webnn
-
5 issues opened.
-
Rename destroy() to dispose() webmachinelearning/webnn#756, by @reillyeon.
-
Rename MLBuffer => MLTensor webmachinelearning/webnn#753, by @bbernhar.
-
Consider separately pass each activation and direction's weights&bias for lstm and gru webmachinelearning/webnn#751, by @philloooo.
-
MLContextOptions.deviceType seems unnecessary outside of conformance testing webmachinelearning/webnn#749, by @mwyrzykowski.
-
Is there a reason to support out of order `axes` for layerNormalization? webmachinelearning/webnn#748, by @philloooo.
-
-
6 issues closed.
-
Allow checking whether operators/types are supported for a backend before creating a graph webmachinelearning/webnn#463.
-
Need clarify the usage of axes=[0,1] for resample2d webmachinelearning/webnn#624.
-
How to define the algorithm of L2_Pool2d? webmachinelearning/webnn#278.
-
Some thoughts of defining ULP tolerance of exp op webmachinelearning/webnn#288.
-
WPT tests tracker / 2 for remaining ops webmachinelearning/webnn#338.
-
Add samples of shape broadcasting webmachinelearning/webnn#378.
-
-
16 issues received 53 new comments.
-
How to define the algorithm of L2_Pool2d? webmachinelearning/webnn#278 (15 comments by @a-sully, @fdwr, @fujunwei, @huningxin)
-
Do we need an `MLConstantOperand`? webmachinelearning/webnn#668 (6 comments by @fdwr, @huningxin, @inexorabletash, @philloooo)
-
Support for transformers webmachinelearning/webnn#375 (6 comments by @anssiko, @fdwr, @huningxin, @sushraja-msft)
-
Allow 0 size dimensions (dimensions containing a 0 in the list of sizes, not a rank of 0 which is valid) webmachinelearning/webnn#391 (6 comments by @bbernhar, @fdwr)
-
Need clarify the usage of axes=[0,1] for resample2d webmachinelearning/webnn#624 (4 comments by @fdwr, @huningxin, @philloooo)
-
Consider removing `lstm` and `gru` operators webmachinelearning/webnn#689 (3 comments by @a-sully, @BruceDai)
-
[MLBuffer] Creation and representing MLBuffer on a XPU devices webmachinelearning/webnn#542 (3 comments by @a-sully, @bbernhar)
-
webmachinelearning/webnn#93 (1 comment by @fdwr)
-
Some thoughts of defining ULP tolerance of exp op webmachinelearning/webnn#288 (1 comment by @BruceDai)
-
webmachinelearning/webnn#128 (1 comment by @fdwr)
-
-
4 pull requests opened.
-
Define opSupportLimits() webmachinelearning/webnn#755, by @philloooo.
-
Add MLTensor explainer webmachinelearning/webnn#754, by @a-sully.
-
Allow arbitrary axes for resample2d webmachinelearning/webnn#752, by @philloooo.
-
Add int64, uint64 data type support for reduceL1, reduceProduct, reduceSum and reduceSumSquare webmachinelearning/webnn#750, by @shiyi9801.
-
-
3 pull requests merged.
-
Define opSupportLimits() webmachinelearning/webnn#755.
-
Allow arbitrary axes for resample2d webmachinelearning/webnn#752.
-
Add blurb explaining broadcasting in more detail webmachinelearning/webnn#745.
-
-
1 pull request closed.
-
Add int64, uint64 data type support for reduceL1, reduceProduct, reduceSum and reduceSumSquare webmachinelearning/webnn#750.
-
-
6 pull requests received 93 new comments.
-
Add MLTensor explainer webmachinelearning/webnn#754 (56 comments by @a-sully, @anssiko, @bbernhar, @fdwr, @huningxin, @philloooo, @RafaelCintron)
-
Define opSupportLimits() webmachinelearning/webnn#755 (18 comments by @fdwr, @huningxin, @philloooo, @reillyeon)
-
Specify destroy() methods and behavior around context loss and error reporting webmachinelearning/webnn#744 (8 comments by @bbernhar, @fdwr, @huningxin, @inexorabletash, @mingmingtasd, @reillyeon)
-
Add int64, uint64 data type support for reduceL1, reduceProduct, reduceSum and reduceSumSquare webmachinelearning/webnn#750 (4 comments by @fdwr, @huningxin, @shiyi9801)
-
Bugfix: Add missing 64-bit integers support for some reduction operators webmachinelearning/webnn#695 (3 comments by @fdwr, @Honry, @philloooo)
-
TR publications
-
13 documents published.
-
Web Neural Network API published 13 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
GitHub
webmachinelearning/webnn
-
4 issues opened.
-
Question on reduction with empty axes and scalar input webmachinelearning/webnn#740, by @philloooo.
-
Limited support for pad on CoreML backend webmachinelearning/webnn#739, by @philloooo.
-
Restrict dimensions size to [0, INT32_MAX] webmachinelearning/webnn#734, by @philloooo.
-
MLBuffer destruction timeline webmachinelearning/webnn#716, by @reillyeon.
-
-
19 issues closed.
-
Question on reduction with empty axes and scalar input webmachinelearning/webnn#740.
-
Consider adding node `label`s for more diagnosable error messages for async errors. webmachinelearning/webnn#585.
-
Consider adopting new broadcasting rules webmachinelearning/webnn#590.
-
Restrict dimensions size to [0, INT32_MAX] webmachinelearning/webnn#734.
-
Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type webmachinelearning/webnn#653.
-
argMax/Min only support scalar axis in TFLite runtime webmachinelearning/webnn#629.
-
Clarify the casting behavior from floating-point / signed integers <-> unsigned integers webmachinelearning/webnn#489.
-
Add security consideration for computation control-flow attack based on weights / constants change webmachinelearning/webnn#443.
-
Can an MLGraphBuilder be reused? webmachinelearning/webnn#567.
-
Need to define error handling of MLNamedArrayBufferViews transfer algorithm webmachinelearning/webnn#351.
-
Consider alternate styling/linking for method argument definitions webmachinelearning/webnn#574.
-
Missing TypeScript Type Declaration webmachinelearning/webnn#677.
-
ArgMax/Min `selectLastIndex` is not supported on CoreML webmachinelearning/webnn#652.
-
Use USVString for operand name webmachinelearning/webnn#713.
-
Specifies scalar values casted to match input type. webmachinelearning/webnn#678.
-
Clarify the usage of 32 bit floating point type and consider using double webmachinelearning/webnn#325.
-
Type of some parameters should match the input data type webmachinelearning/webnn#442.
-
Support output tensor in native memory format webmachinelearning/webnn#173.
-
Implement web platform tests for input validation steps webmachinelearning/webnn#545.
-
-
28 issues received 99 new comments.
-
MLBuffer destruction timeline webmachinelearning/webnn#716 (9 comments by @bbernhar, @huningxin, @reillyeon)
-
Define the maximum number of operand dimensions (maximum rank) webmachinelearning/webnn#456 (9 comments by @a-sully, @fdwr, @Honry, @huningxin, @inexorabletash)
-
[MLBuffer] Uploading/downloading tensor data webmachinelearning/webnn#543 (8 comments by @bbernhar, @fdwr, @huningxin, @reillyeon)
-
Allow checking whether operators/types are supported for a backend before creating a graph webmachinelearning/webnn#463 (8 comments by @fdwr, @huningxin, @inexorabletash, @philloooo)
-
argMax/Min only support scalar axis in TFLite runtime webmachinelearning/webnn#629 (8 comments by @fdwr, @huningxin, @philloooo)
-
Consider removing `lstm` and `gru` operators webmachinelearning/webnn#689 (7 comments by @a-sully, @BruceDai, @fdwr, @huningxin, @miaobin)
-
ArgMax/Min `selectLastIndex` is not supported on CoreML webmachinelearning/webnn#652 (7 comments by @fdwr, @huningxin, @inexorabletash, @mwyrzykowski, @philloooo)
-
Question on reduction with empty axes and scalar input webmachinelearning/webnn#740 (5 comments by @fdwr, @inexorabletash, @philloooo)
-
Where is scatter and gather op? webmachinelearning/webnn#467 (5 comments by @huningxin, @inexorabletash)
-
Do we need an `MLConstantOperand`? webmachinelearning/webnn#668 (3 comments by @a-sully, @fdwr, @inexorabletash)
-
Restrict dimensions size to [0, INT32_MAX] webmachinelearning/webnn#734 (3 comments by @fdwr, @huningxin, @inexorabletash)
-
Missing TypeScript Type Declaration webmachinelearning/webnn#677 (3 comments by @egalli, @inexorabletash)
-
Support multiple op sets / builders webmachinelearning/webnn#714 (3 comments by @reillyeon, @zolkis)
-
Clarify the casting behavior from floating-point / signed integers <-> unsigned integers webmachinelearning/webnn#489 (2 comments by @fdwr, @inexorabletash)
-
Consider alternate styling/linking for method argument definitions webmachinelearning/webnn#574 (2 comments by @inexorabletash)
-
Implement web platform tests for input validation steps webmachinelearning/webnn#545 (2 comments by @inexorabletash)
-
webmachinelearning/webnn#128 (1 comment by @inexorabletash)
-
Support for LocalResponseNormalization (LRN) operation webmachinelearning/webnn#228 (1 comment by @fdwr)
-
Need clarify the usage of axes=[0,1] for resample2d webmachinelearning/webnn#624 (1 comment by @inexorabletash)
-
Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type webmachinelearning/webnn#653 (1 comment by @fdwr)
-
Add security consideration for computation control-flow attack based on weights / constants change webmachinelearning/webnn#443 (1 comment by @inexorabletash)
-
Improve "underlying platform" references in spec webmachinelearning/webnn#462 (1 comment by @inexorabletash)
-
Inconsistency between MLGraphBuilder and MLBuffer construction webmachinelearning/webnn#697 (1 comment by @bbernhar)
-
Support output tensor in native memory format webmachinelearning/webnn#173 (1 comment by @huningxin)
-
-
28 pull requests opened.
-
Introduce MLConstantOperand webmachinelearning/webnn#747, by @a-sully.
-
Bug fix: softmax()'s axis argument should EnforceRange webmachinelearning/webnn#746, by @inexorabletash.
-
Add blurb explaining broadcasting in more detail webmachinelearning/webnn#745, by @inexorabletash.
-
Specify destroy() methods and behavior around context loss and error reporting webmachinelearning/webnn#744, by @inexorabletash.
-
Make prelu() bidirectionally broadcast, improve broadcast wording webmachinelearning/webnn#743, by @inexorabletash.
-
Add optional operator labels for more diagnosable error messages webmachinelearning/webnn#742, by @inexorabletash.
-
Clarify reduction with empty axes and scalar input webmachinelearning/webnn#741, by @inexorabletash.
-
Update dimension valid range to signed integer webmachinelearning/webnn#738, by @philloooo.
-
Add helper for creating MLOperandDescriptor to simplify steps webmachinelearning/webnn#737, by @inexorabletash.
-
Fixed typo webmachinelearning/webnn#736, by @BruceDai.
-
Editorial: Don't reference IDL boolean for values in tensors webmachinelearning/webnn#735, by @inexorabletash.
-
Editorial: Fix an instance of "nchw" enum member styling/linking webmachinelearning/webnn#733, by @inexorabletash.
-
Reference newly landed WebIDL "transferable" definition webmachinelearning/webnn#732, by @inexorabletash.
-
Update documentation and fix some minor issues with lint tool. webmachinelearning/webnn#731, by @inexorabletash.
-
Add outputDataType to argmin/argmax webmachinelearning/webnn#730, by @philloooo.
-
Avoid "sequence" in prose, linkify use in types webmachinelearning/webnn#729, by @inexorabletash.
-
Workflow: Add steps to run tools/lint.mjs after spec is built webmachinelearning/webnn#728, by @inexorabletash.
-
Lint: Improve logic for validating algorithm steps. webmachinelearning/webnn#727, by @inexorabletash.
-
Clarify the cast() op behavior between different data types webmachinelearning/webnn#726, by @inexorabletash.
-
Add security consideration for computation control-flow attacks webmachinelearning/webnn#725, by @inexorabletash.
-
change argmin/argmax to take scalar axis webmachinelearning/webnn#724, by @philloooo.
-
Define error handling of MLNamedArrayBufferViews transfer algorithm webmachinelearning/webnn#723, by @inexorabletash.
-
Remove argmin/max selectLastIndex parameter webmachinelearning/webnn#722, by @philloooo.
-
Style: Link method argument definitions webmachinelearning/webnn#721, by @inexorabletash.
-
Ensure every method dfn is correctly associated with an interface webmachinelearning/webnn#720, by @a-sully.
-
Rename where()'s parameters "input" and "other" to "trueValue" and "falseValue" webmachinelearning/webnn#719, by @shiyi9801.
-
Replace MLActivation with MLRecurrentNetworkActivation webmachinelearning/webnn#718, by @a-sully.
-
Allow MLGraphBuilder.build() to be called only once webmachinelearning/webnn#717, by @a-sully.
-
-
27 pull requests merged.
-
Bug fix: softmax()'s axis argument should EnforceRange webmachinelearning/webnn#746.
-
Clarify reduction with empty axes and scalar input webmachinelearning/webnn#741.
-
Add optional operator labels for more diagnosable error messages webmachinelearning/webnn#742.
-
Make prelu() bidirectionally broadcast, improve broadcast wording webmachinelearning/webnn#743.
-
Add helper for creating MLOperandDescriptor to simplify steps webmachinelearning/webnn#737.
-
Update dimension valid range to signed integer webmachinelearning/webnn#738.
-
Add outputDataType to argmin/argmax webmachinelearning/webnn#730.
-
Fixed typo webmachinelearning/webnn#736.
-
Editorial: Fix an instance of "nchw" enum member styling/linking webmachinelearning/webnn#733.
-
Editorial: Don't reference IDL boolean for values in tensors webmachinelearning/webnn#735.
-
Replace MLActivation with MLRecurrentNetworkActivation webmachinelearning/webnn#718.
-
Reference newly landed WebIDL "transferable" definition webmachinelearning/webnn#732.
-
Update documentation and fix some minor issues with lint tool. webmachinelearning/webnn#731.
-
Avoid "sequence" in prose, linkify use in types webmachinelearning/webnn#729.
-
Workflow: Add steps to run tools/lint.mjs after spec is built webmachinelearning/webnn#728.
-
change argmin/argmax to take scalar axis webmachinelearning/webnn#724.
-
Clarify the cast() op behavior between different data types webmachinelearning/webnn#726.
-
Add security consideration for computation control-flow attacks webmachinelearning/webnn#725.
-
Lint: Improve logic for validating algorithm steps. webmachinelearning/webnn#727.
-
Allow MLGraphBuilder.build() to be called only once webmachinelearning/webnn#717.
-
Define error handling of MLNamedArrayBufferViews transfer algorithm webmachinelearning/webnn#723.
-
Style: Link method argument definitions webmachinelearning/webnn#721.
-
Remove argmin/max selectLastIndex parameter webmachinelearning/webnn#722.
-
Rename where()'s parameters "input" and "other" to "trueValue" and "falseValue" webmachinelearning/webnn#719.
-
Ensure every method dfn is correctly associated with an interface webmachinelearning/webnn#720.
-
Fix #713: DOMString to USVString webmachinelearning/webnn#715.
-
Introduce MLNumber for specifying numeric inputs of any type webmachinelearning/webnn#647.
-
-
30 pull requests received 185 new comments.
-
Allow MLGraphBuilder.build() to be called only once webmachinelearning/webnn#717 (23 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Introduce MLNumber for specifying numeric inputs of any type webmachinelearning/webnn#647 (21 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Clarify reduction with empty axes and scalar input webmachinelearning/webnn#741 (16 comments by @fdwr, @huningxin, @inexorabletash)
-
Add optional operator labels for more diagnosable error messages webmachinelearning/webnn#742 (16 comments by @fdwr, @huningxin, @inexorabletash, @lisa0314, @philloooo)
-
Add outputDataType to argmin/argmax webmachinelearning/webnn#730 (11 comments by @a-sully, @fdwr, @huningxin, @philloooo)
-
Specify destroy() methods and behavior around context loss and error reporting webmachinelearning/webnn#744 (10 comments by @inexorabletash, @mingmingtasd, @reillyeon)
-
Replace MLActivation with MLRecurrentNetworkActivation webmachinelearning/webnn#718 (10 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Clarify the cast() op behavior between different data types webmachinelearning/webnn#726 (10 comments by @fdwr, @inexorabletash)
-
change argmin/argmax to take scalar axis webmachinelearning/webnn#724 (9 comments by @fdwr, @inexorabletash, @philloooo)
-
Add helper for creating MLOperandDescriptor to simplify steps webmachinelearning/webnn#737 (8 comments by @fdwr, @huningxin, @inexorabletash)
-
Make prelu() bidirectionally broadcast, improve broadcast wording webmachinelearning/webnn#743 (6 comments by @a-sully, @fdwr, @inexorabletash)
-
Avoid "sequence" in prose, linkify use in types webmachinelearning/webnn#729 (5 comments by @fdwr, @inexorabletash)
-
Define error handling of MLNamedArrayBufferViews transfer algorithm webmachinelearning/webnn#723 (5 comments by @inexorabletash, @reillyeon)
-
Style: Link method argument definitions webmachinelearning/webnn#721 (4 comments by @fdwr, @inexorabletash)
-
Rename where()'s parameters "input" and "other" to "trueValue" and "falseValue" webmachinelearning/webnn#719 (3 comments by @fdwr, @inexorabletash, @shiyi9801)
-
Add blurb explaining broadcasting in more detail webmachinelearning/webnn#745 (2 comments by @fdwr, @inexorabletash)
-
Editorial: Fix an instance of "nchw" enum member styling/linking webmachinelearning/webnn#733 (2 comments by @huningxin, @inexorabletash)
-
Editorial: Don't reference IDL boolean for values in tensors webmachinelearning/webnn#735 (2 comments by @fdwr, @inexorabletash)
-
Make operand data type and rank validation table-driven webmachinelearning/webnn#657 (2 comments by @fdwr, @inexorabletash)
-
Workflow: Add steps to run tools/lint.mjs after spec is built webmachinelearning/webnn#728 (2 comments by @inexorabletash)
-
Bugfix: Add missing 64-bit integers support for some reduction operators webmachinelearning/webnn#695 (2 comments by @fdwr, @philloooo)
-
Bug fix: softmax()'s axis argument should EnforceRange webmachinelearning/webnn#746 (1 comment by @inexorabletash)
-
Reference newly landed WebIDL "transferable" definition webmachinelearning/webnn#732 (1 comment by @inexorabletash)
-
Update documentation and fix some minor issues with lint tool. webmachinelearning/webnn#731 (1 comment by @inexorabletash)
-
Lint: Improve logic for validating algorithm steps. webmachinelearning/webnn#727 (1 comment by @inexorabletash)
-
Add security consideration for computation control-flow attacks webmachinelearning/webnn#725 (1 comment by @inexorabletash)
-
TR publications
-
10 documents published.
-
Web Neural Network API published 10 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
4 participants joined.
-
Andrew Nolan (Microsoft Corporation) joined.
-
Natasha Gaitonde (Microsoft Corporation) joined.
-
Sushanth Rajasankar (Microsoft Corporation) joined.
-
Rob Kochman (Google LLC) joined.
-
GitHub
webmachinelearning/webnn
-
3 issues opened.
-
Support multiple op sets / builders webmachinelearning/webnn#714, by @zolkis.
-
Use USVString for operand name webmachinelearning/webnn#713, by @philloooo.
-
Failed to execute 'createContext' on 'ML': DirectML: Unable to find a capable adapter. webmachinelearning/webnn#701, by @josefhernandez.
-
-
8 issues closed.
-
Deprecated squeeze with emulation path is missing its own section webmachinelearning/webnn#551.
-
Make `MLOperandDescriptor.dimensions` a `FrozenArray` webmachinelearning/webnn#670.
-
Bitwise operators and logical operators naming (rename not to logicalNot) webmachinelearning/webnn#496.
-
"Unidirectionally broadcast the shapes" steps have issues webmachinelearning/webnn#662.
-
Failed to execute 'createContext' on 'ML': DirectML: Unable to find a capable adapter. webmachinelearning/webnn#701.
-
Remove `builder.constant(value, type)` variant webmachinelearning/webnn#475.
-
`split` into sizes are not widely supported webmachinelearning/webnn#392.
-
MLOperandDataType and ArrayBufferView compatibility table missing int64/uint64 webmachinelearning/webnn#693.
-
-
1 issue reopened.
-
How to define the algorithm of L2_Pool2d? webmachinelearning/webnn#278, by @huningxin.
-
-
18 issues received 53 new comments.
-
WebNN should support NPU and QDQ operations webmachinelearning/webnn#623 (7 comments by @fdwr, @huningxin, @mwyrzykowski, @zolkis)
-
Inconsistency between MLGraphBuilder and MLBuffer construction webmachinelearning/webnn#697 (6 comments by @a-sully, @bbernhar, @RafaelCintron, @zolkis)
-
Allow checking whether operators/types are supported for a backend before creating a graph webmachinelearning/webnn#463 (6 comments by @Honry, @huningxin, @philloooo, @reillyeon, @wacky6)
-
Consider adding node `label`s for more diagnosable error messages for async errors. webmachinelearning/webnn#585 (6 comments by @fdwr, @lisa0314, @mingmingtasd, @philloooo)
-
Don't transfer input ArrayBuffers webmachinelearning/webnn#566 (5 comments by @a-sully, @inexorabletash, @reillyeon, @zolkis)
-
[MLBuffer] Creation and representing MLBuffer on a XPU devices webmachinelearning/webnn#542 (3 comments by @bbernhar, @huningxin)
-
`split` into sizes are not widely supported webmachinelearning/webnn#392 (3 comments by @a-sully, @fdwr, @huningxin)
-
[MLBuffer] Support interop with WebGPU webmachinelearning/webnn#688 (2 comments by @bbernhar, @RafaelCintron)
-
Consider removing `lstm` and `gru` operators webmachinelearning/webnn#689 (2 comments by @a-sully, @fdwr)
-
Bitwise operators and logical operators naming (rename not to logicalNot) webmachinelearning/webnn#496 (2 comments by @fdwr, @inexorabletash)
-
Clarify the usage of 32 bit floating point type and consider using double webmachinelearning/webnn#325 (1 comment by @inexorabletash)
-
Failed to execute 'createContext' on 'ML': DirectML: Unable to find a capable adapter. webmachinelearning/webnn#701 (1 comment by @huningxin)
-
-
14 pull requests opened.
-
Fix #713: DOMString to USVString webmachinelearning/webnn#715, by @zolkis.
-
Process: Add "device selection" label webmachinelearning/webnn#712, by @anssiko.
-
Remove unnecessary subsections, make order consistent webmachinelearning/webnn#711, by @inexorabletash.
-
Style: Consistently use backticks for expressions in prose webmachinelearning/webnn#710, by @inexorabletash.
-
Write example decompositions as complete functions webmachinelearning/webnn#709, by @inexorabletash.
-
Link fix: Link to "is empty" not "empty" when an adjective is intended webmachinelearning/webnn#708, by @inexorabletash.
-
Style: Ensure algorithm steps end in '.' or ':' webmachinelearning/webnn#707, by @inexorabletash.
-
Bugfix: Add missing validation for conv2d() and convTranspose2d() webmachinelearning/webnn#706, by @huningxin.
-
Add a section for updating an existing operation webmachinelearning/webnn#705, by @huningxin.
-
Workflows: use ubuntu-latest to get up-to-date python/Bikeshed webmachinelearning/webnn#704, by @inexorabletash.
-
Remove MLActivations definitely not usable with recurrent ops webmachinelearning/webnn#703, by @a-sully.
-
Add tools/ dir and some handy utilities webmachinelearning/webnn#702, by @inexorabletash.
-
Copy input ArrayBuffers webmachinelearning/webnn#700, by @a-sully.
-
Bug fix: Correct a handful of typos/JS errors in decompositions webmachinelearning/webnn#699, by @inexorabletash.
-
-
17 pull requests merged.
-
Process: Add "device selection" label webmachinelearning/webnn#712.
-
Style: Consistently use backticks for expressions in prose webmachinelearning/webnn#710.
-
Remove unnecessary subsections, make order consistent webmachinelearning/webnn#711.
-
Write example decompositions as complete functions webmachinelearning/webnn#709.
-
Add MLDeviceType npu webmachinelearning/webnn#696.
-
Link fix: Link to "is empty" not "empty" when an adjective is intended webmachinelearning/webnn#708.
-
Content: Introduce emulation section, move reshape examples there webmachinelearning/webnn#598.
-
Add a section for updating an existing operation webmachinelearning/webnn#705.
-
Bugfix: Add missing validation for conv2d() and convTranspose2d() webmachinelearning/webnn#706.
-
Style: Ensure algorithm steps end in '.' or ':' webmachinelearning/webnn#707.
-
Add tools/ dir and some handy utilities webmachinelearning/webnn#702.
-
Remove MLActivations definitely not usable with recurrent ops webmachinelearning/webnn#703.
-
Workflows: use ubuntu-latest to get up-to-date python/Bikeshed webmachinelearning/webnn#704.
-
Fix unidirectionally broadcast the shapes steps webmachinelearning/webnn#663.
-
Bug fix: Correct a handful of typos/JS errors in decompositions webmachinelearning/webnn#699.
-
Swap parameters of scalar constant() operand method webmachinelearning/webnn#650.
-
Add int64/uint64 to ArrayBufferView compatibility table webmachinelearning/webnn#698.
-
-
17 pull requests received 137 new comments.
-
Add tools/ dir and some handy utilities webmachinelearning/webnn#702 (32 comments by @fdwr, @huningxin, @inexorabletash)
-
Add MLDeviceType npu webmachinelearning/webnn#696 (23 comments by @anssiko, @fdwr, @inexorabletash, @mwyrzykowski, @zolkis)
-
Write example decompositions as complete functions webmachinelearning/webnn#709 (15 comments by @fdwr, @huningxin, @inexorabletash, @zolkis)
-
Remove MLActivations definitely not usable with recurrent ops webmachinelearning/webnn#703 (12 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Swap parameters of scalar constant() operand method webmachinelearning/webnn#650 (12 comments by @fdwr, @inexorabletash)
-
Link fix: Link to "is empty" not "empty" when an adjective is intended webmachinelearning/webnn#708 (10 comments by @fdwr, @huningxin, @inexorabletash)
-
Bugfix: Add missing 64-bit integers support for some reduction operators webmachinelearning/webnn#695 (10 comments by @a-sully, @fdwr, @huningxin, @mwyrzykowski, @philloooo)
-
Bug fix: Correct a handful of typos/JS errors in decompositions webmachinelearning/webnn#699 (6 comments by @fdwr, @huningxin, @inexorabletash)
-
Process: Add "device selection" label webmachinelearning/webnn#712 (3 comments by @anssiko, @inexorabletash)
-
Bugfix: Add missing validation for conv2d() and convTranspose2d() webmachinelearning/webnn#706 (3 comments by @fdwr, @huningxin, @inexorabletash)
-
Introduce MLNumber for specifying numeric inputs of any type webmachinelearning/webnn#647 (2 comments by @anssiko, @inexorabletash)
-
Style: Consistently use backticks for expressions in prose webmachinelearning/webnn#710 (2 comments by @inexorabletash)
-
Content: Introduce emulation section, move reshape examples there webmachinelearning/webnn#598 (2 comments by @fdwr, @huningxin)
-
Fix unidirectionally broadcast the shapes steps webmachinelearning/webnn#663 (2 comments by @huningxin)
-
Remove unnecessary subsections, make order consistent webmachinelearning/webnn#711 (1 comment by @inexorabletash)
-
Style: Ensure algorithm steps end in '.' or ':' webmachinelearning/webnn#707 (1 comment by @inexorabletash)
-
Workflows: use ubuntu-latest to get up-to-date python/Bikeshed webmachinelearning/webnn#704 (1 comment by @inexorabletash)
-
TR publications
-
12 documents published.
-
Web Neural Network API published 12 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
4 participants joined.
-
Jan Williams (TPGi) joined.
-
Dr. Humera Noor Minhas (W3C Invited Experts) joined.
-
Muthaiah Venkatachalam (Intel Corporation) joined.
-
Rahul Unnikrishnan Nair (Intel Corporation) joined.
-
-
1 participant left.
-
Dr. Humera Noor Minhas (W3C Invited Experts) left.
-
-
1 organization joined.
-
TPGi joined.
-
GitHub
webmachinelearning/webnn
-
12 issues opened.
-
Inconsistency between MLGraphBuilder and MLBuffer construction webmachinelearning/webnn#697, by @reillyeon.
-
Consider adding int64/uint64 data type support for some reduce operators webmachinelearning/webnn#694, by @lisa0314.
-
MLOperandDataType and ArrayBufferView compatibility table missing int64/uint64 webmachinelearning/webnn#693, by @inexorabletash.
-
Divide-by-zero outcome should be standardized webmachinelearning/webnn#691, by @huningxin.
-
Consider removing `lstm` and `gru` operators webmachinelearning/webnn#689, by @a-sully.
-
[MLBuffer] Support interop with WebGPU webmachinelearning/webnn#688, by @bbernhar.
-
`layerNormalization()` method steps should validate the values of `options.axes` webmachinelearning/webnn#686, by @huningxin.
-
Shall we update the spec to add constraints on the axes and the input rank for the reduction operator? webmachinelearning/webnn#681, by @mei1127.
-
Specifies scalar values casted to match input type. webmachinelearning/webnn#678, by @philloooo.
-
Missing TypeScript Type Declaration webmachinelearning/webnn#677, by @egalli.
-
why `gather` indices only accept uint32 or int64? webmachinelearning/webnn#675, by @philloooo.
-
Meta: Introduce "Interop" label? webmachinelearning/webnn#673, by @inexorabletash.
-
-
13 issues closed.
-
[investigation] buffer sharing between GPU and ML accelerator webmachinelearning/webnn#33.
-
Clarify where GPU related states are stored for WebGPU interop (MLContext, MLCommandEncoder and MLGraph) webmachinelearning/webnn#342.
-
Support for device-based tensor storage objects webmachinelearning/webnn#482.
-
`layerNormalization()` method steps should validate the values of `options.axes` webmachinelearning/webnn#686.
-
Shall we update the spec to add constraints on the axes and the input rank for the reduction operator? webmachinelearning/webnn#681.
-
why `gather` indices only accept uint32 or int64? webmachinelearning/webnn#675.
-
Synchronously validate input operands/activations webmachinelearning/webnn#572.
-
Meta: Introduce "Interop" label? webmachinelearning/webnn#673.
-
Clarify the restriction for `minValue` and `maxValue` of `MLClampOptions` webmachinelearning/webnn#396.
-
Should softmax be a fused activation ? webmachinelearning/webnn#471.
-
Consider removing `MLActivation` parameters used for op fusion webmachinelearning/webnn#658.
-
Remove use of `null` as parameter to `reshape` in code examples webmachinelearning/webnn#667.
-
Need clarify the maximum limit of the hidden size of the Gru operator webmachinelearning/webnn#625.
-
-
1 issue reopened.
-
Meta: Introduce "Interop" label? webmachinelearning/webnn#673, by @anssiko.
-
-
32 issues received 102 new comments.
-
Allow checking whether operators/types are supported for a backend before creating a graph webmachinelearning/webnn#463 (10 comments by @Honry, @huningxin, @philloooo, @reillyeon)
-
[MLBuffer] Creation and representing MLBuffer on a XPU devices webmachinelearning/webnn#542 (7 comments by @a-sully, @bbernhar)
-
Specifies scalar values casted to match input type. webmachinelearning/webnn#678 (6 comments by @inexorabletash)
-
Consider removing `MLActivation` parameters used for op fusion webmachinelearning/webnn#658 (6 comments by @a-sully, @fdwr, @huningxin, @mingmingtasd)
-
Specify the operand data type constraints of operation webmachinelearning/webnn#283 (5 comments by @fdwr, @huningxin, @inexorabletash, @philloooo)
-
Shall we update the spec to add constraints on the axes and the input rank for the reduction operator? webmachinelearning/webnn#681 (5 comments by @fdwr, @inexorabletash)
-
Should softmax be a fused activation ? webmachinelearning/webnn#471 (5 comments by @a-sully, @fdwr, @huningxin)
-
[MLBuffer] Support for MLBuffer in graph execution webmachinelearning/webnn#544 (4 comments by @bbernhar, @inexorabletash)
-
WebNN should support NPU and QDQ operations webmachinelearning/webnn#623 (4 comments by @fdwr, @inexorabletash, @mwyrzykowski, @zolkis)
-
Meta: Introduce "Interop" label? webmachinelearning/webnn#673 (4 comments by @anssiko, @inexorabletash)
-
API lacks handling for async ML device errors on the context webmachinelearning/webnn#477 (3 comments by @bbernhar, @inexorabletash, @zolkis)
-
Make `MLOperandDescriptor.dimensions` a `FrozenArray` webmachinelearning/webnn#670 (3 comments by @a-sully, @anssiko, @inexorabletash)
-
Support output tensor in native memory format webmachinelearning/webnn#173 (3 comments by @a-sully, @anssiko)
-
Missing TypeScript Type Declaration webmachinelearning/webnn#677 (3 comments by @anssiko, @bbernhar, @egalli)
-
Consider adding node `label`s for more diagnosable error messages for async errors. webmachinelearning/webnn#585 (3 comments by @lisa0314, @philloooo, @zolkis)
-
Do we need an `MLConstantOperand`? webmachinelearning/webnn#668 (3 comments by @a-sully, @huningxin)
-
ArgMax/Min `selectLastIndex` is not supported on CoreML webmachinelearning/webnn#652 (3 comments by @fdwr, @mwyrzykowski, @philloooo)
-
Clarify where GPU related states are stored for WebGPU interop (MLContext, MLCommandEncoder and MLGraph) webmachinelearning/webnn#342 (2 comments by @bbernhar, @huningxin)
-
Allow 0 size dimensions (dimensions containing a 0 in the list of sizes, not a rank of 0 which is valid) webmachinelearning/webnn#391 (2 comments by @fdwr, @reillyeon)
-
Google Chrome Feedback on WebNN: aiming for broad device coverage and maintainability webmachinelearning/webnn#453 (2 comments by @anssiko)
-
Bitwise operators and logical operators naming (rename not to logicalNot) webmachinelearning/webnn#496 (2 comments by @inexorabletash, @shiyi9801)
-
why `gather` indices only accept uint32 or int64? webmachinelearning/webnn#675 (2 comments by @fdwr, @huningxin)
-
Remove use of `null` as parameter to `reshape` in code examples webmachinelearning/webnn#667 (2 comments by @inexorabletash)
-
Inconsistency between MLGraphBuilder and MLBuffer construction webmachinelearning/webnn#697 (1 comment by @bbernhar)
-
Divide-by-zero outcome should be standardized webmachinelearning/webnn#691 (1 comment by @inexorabletash)
-
webmachinelearning/webnn#33 (1 comment by @bbernhar)
-
Consider dropping the support of uint64/int64 data type for some operators webmachinelearning/webnn#654 (1 comment by @huningxin)
-
Support for device-based tensor storage objects webmachinelearning/webnn#482 (1 comment by @bbernhar)
-
Synchronously validate input operands/activations webmachinelearning/webnn#572 (1 comment by @inexorabletash)
-
Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type webmachinelearning/webnn#653 (1 comment by @philloooo)
-
-
16 pull requests opened.
-
Add int64/uint64 to ArrayBufferView compatibility table webmachinelearning/webnn#698, by @inexorabletash.
-
Add MLDeviceType npu webmachinelearning/webnn#696, by @fdwr.
-
Bugfix: Add missing 64-bit integers support for some reduction operators webmachinelearning/webnn#695, by @huningxin.
-
Rename the logical operation "not" to "logicalNot" webmachinelearning/webnn#692, by @shiyi9801.
-
Add missing validation for pad(), slice(), and split() webmachinelearning/webnn#690, by @inexorabletash.
-
Validate layerNormalization options.axes webmachinelearning/webnn#687, by @inexorabletash.
-
Support int32 data type for 'indices' operand of 'gather' operator webmachinelearning/webnn#685, by @huningxin.
-
Remove note about interop issues with clamp()'s minValue == maxValue webmachinelearning/webnn#684, by @inexorabletash.
-
Validate no duplicate axes for reduction ops webmachinelearning/webnn#683, by @inexorabletash.
-
Process: Add "interop" label webmachinelearning/webnn#682, by @anssiko.
-
Add missing definitions of inputShape to conv2d algorithms webmachinelearning/webnn#680, by @inexorabletash.
-
Build fix: Correct link for "transferred" term webmachinelearning/webnn#679, by @inexorabletash.
-
Rename MLOperandDescriptor's "dimensions" key to "shape" webmachinelearning/webnn#676, by @a-sully.
-
Use consistent phrasing for operator creation webmachinelearning/webnn#674, by @inexorabletash.
-
Handful of algorithm and convention fixes webmachinelearning/webnn#672, by @inexorabletash.
-
Remove uses of null as parameter to reshape() in code samples webmachinelearning/webnn#671, by @a-sully.
-
-
15 pull requests merged.
-
Rename the logical operation "not" to "logicalNot" webmachinelearning/webnn#692.
-
Add missing validation for pad(), slice(), and split() webmachinelearning/webnn#690.
-
Validate layerNormalization options.axes webmachinelearning/webnn#687.
-
Validate no duplicate axes for reduction ops webmachinelearning/webnn#683.
-
Support int32 data type for 'indices' operand of 'gather' operator webmachinelearning/webnn#685.
-
Use consistent phrasing for operator creation webmachinelearning/webnn#674.
-
Remove note about interop issues with clamp()'s minValue == maxValue webmachinelearning/webnn#684.
-
Process: Add "interop" label webmachinelearning/webnn#682.
-
Build fix: Correct link for "transferred" term webmachinelearning/webnn#679.
-
Add missing definitions of inputShape to conv2d algorithms webmachinelearning/webnn#680.
-
Remove MLActivation parameters used for op fusion webmachinelearning/webnn#664.
-
Remove uses of null as parameter to reshape() in code samples webmachinelearning/webnn#671.
-
Handful of algorithm and convention fixes webmachinelearning/webnn#672.
-
Simplify, correct, and add validation for GRU/LSTM and friends webmachinelearning/webnn#659.
-
Validate the hidden size of GRU and LSTM operators webmachinelearning/webnn#644.
-
-
21 pull requests received 82 new comments.
-
Add missing validation for pad(), slice(), and split() webmachinelearning/webnn#690 (11 comments by @fdwr, @huningxin, @inexorabletash)
-
Remove MLActivation parameters used for op fusion webmachinelearning/webnn#664 (11 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Swap parameters of scalar constant() operand method webmachinelearning/webnn#650 (8 comments by @huningxin, @inexorabletash, @zolkis)
-
Introduce MLNumber for specifying numeric inputs of any type webmachinelearning/webnn#647 (6 comments by @inexorabletash, @philloooo, @zolkis)
-
Validate no duplicate axes for reduction ops webmachinelearning/webnn#683 (6 comments by @fdwr, @huningxin, @inexorabletash)
-
Handful of algorithm and convention fixes webmachinelearning/webnn#672 (6 comments by @fdwr, @inexorabletash)
-
Rename the logical operation "not" to "logicalNot" webmachinelearning/webnn#692 (5 comments by @anssiko, @huningxin, @inexorabletash, @w3cbot)
-
Use consistent phrasing for operator creation webmachinelearning/webnn#674 (5 comments by @fdwr, @inexorabletash)
-
Add MLDeviceType npu webmachinelearning/webnn#696 (4 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Bugfix: Add missing 64-bit integers support for some reduction operators webmachinelearning/webnn#695 (3 comments by @a-sully, @Honry, @huningxin)
-
Make operand data type and rank validation table-driven webmachinelearning/webnn#657 (3 comments by @inexorabletash, @zolkis)
-
Build fix: Correct link for "transferred" term webmachinelearning/webnn#679 (2 comments by @inexorabletash)
-
Add missing definitions of inputShape to conv2d algorithms webmachinelearning/webnn#680 (2 comments by @inexorabletash)
-
Remove uses of null as parameter to reshape() in code samples webmachinelearning/webnn#671 (2 comments by @a-sully)
-
Simplify, correct, and add validation for GRU/LSTM and friends webmachinelearning/webnn#659 (2 comments by @fdwr, @inexorabletash)
-
Add int64/uint64 to ArrayBufferView compatibility table webmachinelearning/webnn#698 (1 comment by @inexorabletash)
-
Fix unidirectionally broadcast the shapes steps webmachinelearning/webnn#663 (1 comment by @inexorabletash)
-
Validate layerNormalization options.axes webmachinelearning/webnn#687 (1 comment by @inexorabletash)
-
Remove note about interop issues with clamp()'s minValue == maxValue webmachinelearning/webnn#684 (1 comment by @inexorabletash)
-
Content: Introduce emulation section, move reshape examples there webmachinelearning/webnn#598 (1 comment by @inexorabletash)
-
TR publications
-
13 documents published.
-
Web Neural Network API published as a Candidate Recommendation Snapshot. History of Web Neural Network API
-
Web Neural Network API published 12 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
4 participants joined.
-
Yunchao He (ByteDance) joined.
-
Kenji Baheux (Google LLC) joined.
-
Michael McCool (Intel Corporation) joined.
-
Juan Deng (Alibaba Group) joined.
-
GitHub
webmachinelearning/webnn
-
17 issues opened.
-
Make `MLOperandDescriptor.dimensions` a `FrozenArray` webmachinelearning/webnn#670, by @a-sully.
-
Rename `MLOperandDescriptor`'s `dimensions` to `shape` webmachinelearning/webnn#669, by @a-sully.
-
Do we need an `MLConstantOperand`? webmachinelearning/webnn#668, by @a-sully.
-
Remove use of `null` as parameter to `reshape` in code examples webmachinelearning/webnn#667, by @a-sully.
-
Reconsider `MLOperand` methods webmachinelearning/webnn#666, by @a-sully.
-
"Unidirectionally broadcast the shapes" steps have issues webmachinelearning/webnn#662, by @huningxin.
-
prelu steps should validate the data type of slope is equal to the data type of input webmachinelearning/webnn#660, by @huningxin.
-
Consider removing `MLActivation` parameters used for op fusion webmachinelearning/webnn#658, by @a-sully.
-
Consider dropping the support of uint64/int64 data type for some operators webmachinelearning/webnn#654, by @lisa0314.
-
Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type webmachinelearning/webnn#653, by @philloooo.
-
ArgMax/Min `selectLastIndex` is not supported on CoreML webmachinelearning/webnn#652, by @philloooo.
-
Consider removing `steepness` parameter of softplus webmachinelearning/webnn#645, by @shiyi9801.
-
Behavior for numeric overflows/underflows/etc in algorithms webmachinelearning/webnn#636, by @inexorabletash.
-
`MLContext.compute()` explicitly rejects promises where the sub-septs may currently throw webmachinelearning/webnn#634, by @huningxin.
-
Should `MLContext.compute()` throw an exception if the inputs validation steps fail? webmachinelearning/webnn#633, by @huningxin.
-
Need to add the value restriction of output padding in convTranspose2d webmachinelearning/webnn#630, by @mei1127.
-
argMax/Min only support scalar axis in TFLite runtime webmachinelearning/webnn#629, by @fujunwei.
-
-
22 issues closed.
-
Allow no-op graphs? webmachinelearning/webnn#614.
-
Constant sequential filling operation needs output shape parameter webmachinelearning/webnn#492.
-
prelu steps should validate the data type of slope is equal to the data type of input webmachinelearning/webnn#660.
-
Specify the operand data type constraints of operation webmachinelearning/webnn#283.
-
Softmax axis absent webmachinelearning/webnn#466.
-
Consider removing `steepness` parameter of softplus webmachinelearning/webnn#645.
-
`keepDimensions` of reduction ops webmachinelearning/webnn#465.
-
WebIDL definition for constant(start, end, step, type) is missing webmachinelearning/webnn#571.
-
Should Gather's indices data type be integers and support negative value? webmachinelearning/webnn#484.
-
Need to add the value restriction of output padding in convTranspose2d webmachinelearning/webnn#630.
-
Is "validate graph resources" backwards? webmachinelearning/webnn#602.
-
BatchNormalization should be an optional operation webmachinelearning/webnn#187.
-
`MLContext.compute()` explicitly rejects promises where the sub-septs may currently throw webmachinelearning/webnn#634.
-
Need clarify scale factor for resample2d webmachinelearning/webnn#610.
-
reduceLogSum, reduceLogSumExp, reduceSumSquare are not supported on OV/NNAPI webmachinelearning/webnn#209.
-
Candidate Recommendation readiness tracker webmachinelearning/webnn#240.
-
Supported operations query mechanism webmachinelearning/webnn#504.
-
The valid range of `output_padding` of MLConvTranspose2dOptions should be clarified webmachinelearning/webnn#368.
-
Need Gelu operation webmachinelearning/webnn#626.
-
Should `MLContext.compute()` throw an exception if the inputs validation steps fail? webmachinelearning/webnn#633.
-
Rename inputSize variables as inputRank in algorithms webmachinelearning/webnn#588.
-
Graph with no input webmachinelearning/webnn#615.
-
-
36 issues received 118 new comments.
-
[MLBuffer] Creation and representing MLBuffer on a XPU devices webmachinelearning/webnn#542 (14 comments by @a-sully, @bbernhar, @huningxin, @mwyrzykowski, @RafaelCintron, @reillyeon)
-
Constant sequential filling operation needs output shape parameter webmachinelearning/webnn#492 (11 comments by @a-sully, @fdwr, @wacky6)
-
Allow no-op graphs? webmachinelearning/webnn#614 (9 comments by @a-sully, @anssiko, @fdwr, @huningxin, @inexorabletash)
-
Should `MLContext.compute()` throw an exception if the inputs validation steps fail? webmachinelearning/webnn#633 (8 comments by @a-sully, @huningxin, @inexorabletash)
-
Specify the operand data type constraints of operation webmachinelearning/webnn#283 (7 comments by @fdwr, @huningxin, @inexorabletash, @philloooo, @wacky6)
-
Consider removing `steepness` parameter of softplus webmachinelearning/webnn#645 (6 comments by @a-sully, @fdwr, @huningxin, @wacky6)
-
Need clarify scale factor for resample2d webmachinelearning/webnn#610 (6 comments by @huningxin, @inexorabletash, @mingmingtasd)
-
Should softmax be a fused activation ? webmachinelearning/webnn#471 (5 comments by @fdwr, @huningxin, @inexorabletash, @wacky6)
-
Consider adding node `label`s for more diagnosable error messages for async errors. webmachinelearning/webnn#585 (4 comments by @lisa0314, @mingmingtasd)
-
Consider dropping the support of uint64/int64 data type for some operators webmachinelearning/webnn#654 (3 comments by @a-sully, @inexorabletash, @lisa0314)
-
WebNN should support NPU and QDQ operations webmachinelearning/webnn#623 (3 comments by @fdwr, @philloooo, @zolkis)
-
argMax/Min only support scalar axis in TFLite runtime webmachinelearning/webnn#629 (3 comments by @a-sully, @philloooo)
-
Type of some parameters should match the input data type webmachinelearning/webnn#442 (3 comments by @fdwr, @huningxin, @inexorabletash)
-
Behavior for numeric overflows/underflows/etc in algorithms webmachinelearning/webnn#636 (3 comments by @inexorabletash, @zolkis)
-
Default axis being 1 of batchNormalization operation is invalid for 1D input tensor webmachinelearning/webnn#307 (2 comments by @inexorabletash, @zolkis)
-
"Unidirectionally broadcast the shapes" steps have issues webmachinelearning/webnn#662 (2 comments by @huningxin, @inexorabletash)
-
Clarify the restriction for `minValue` and `maxValue` of `MLClampOptions` webmachinelearning/webnn#396 (2 comments by @fdwr, @inexorabletash)
-
Remove `builder.constant(value, type)` variant webmachinelearning/webnn#475 (2 comments by @fdwr, @inexorabletash)
-
Need clarify the maximum limit of the hidden size of the Gru operator webmachinelearning/webnn#625 (2 comments by @inexorabletash)
-
`keepDimensions` of reduction ops webmachinelearning/webnn#465 (2 comments by @huningxin, @inexorabletash)
-
Need clarify the usage of axes=[0,1] for resample2d webmachinelearning/webnn#624 (1 comment by @inexorabletash)
-
ArgMax/Min `selectLastIndex` is not supported on CoreML webmachinelearning/webnn#652 (1 comment by @fdwr)
-
WebIDL definition for constant(start, end, step, type) is missing webmachinelearning/webnn#571 (1 comment by @inexorabletash)
-
Add "implementation consideration" about how out-of-bound indices of Gather/Scatter should be handled webmachinelearning/webnn#486 (1 comment by @fdwr)
-
Allow checking whether operators/types are supported for a backend before creating a graph webmachinelearning/webnn#463 (1 comment by @philloooo)
-
The valid range of `output_padding` of MLConvTranspose2dOptions should be clarified webmachinelearning/webnn#368 (1 comment by @inexorabletash)
-
`MLContext.compute()` explicitly rejects promises where the sub-septs may currently throw webmachinelearning/webnn#634 (1 comment by @inexorabletash)
-
Is "validate graph resources" backwards? webmachinelearning/webnn#602 (1 comment by @inexorabletash)
-
Need to add the value restriction of output padding in convTranspose2d webmachinelearning/webnn#630 (1 comment by @inexorabletash)
-
-
25 pull requests opened.
-
Add note about no-op graphs webmachinelearning/webnn#665, by @inexorabletash.
-
Remove MLActivation parameters used for op fusion webmachinelearning/webnn#664, by @a-sully.
-
Fix unidirectionally broadcast the shapes steps webmachinelearning/webnn#663, by @huningxin.
-
Bugfix: prelu should validate slope operand webmachinelearning/webnn#661, by @huningxin.
-
Simplify, correct, and add validation for GRU/LSTM and friends webmachinelearning/webnn#659, by @inexorabletash.
-
Make operand data type and rank validation table-driven webmachinelearning/webnn#657, by @inexorabletash.
-
Remove sequential filling overload of constant() webmachinelearning/webnn#656, by @a-sully.
-
gru/lstm: Fix missing/incorrect variables and options use webmachinelearning/webnn#655, by @inexorabletash.
-
Remove softplus() steepness option webmachinelearning/webnn#651, by @a-sully.
-
Swap parameters of scalar constant() operand method webmachinelearning/webnn#650, by @inexorabletash.
-
Add axis argument to softmax() webmachinelearning/webnn#649, by @inexorabletash.
-
Editorial: Add note about keepDimensions of reduction ops webmachinelearning/webnn#648, by @inexorabletash.
-
Introduce MLNumber for specifying numeric inputs of any type webmachinelearning/webnn#647, by @inexorabletash.
-
Specify the operand data type constraints of operations webmachinelearning/webnn#646, by @inexorabletash.
-
Validate the hidden size of GRU and LSTM operators webmachinelearning/webnn#644, by @inexorabletash.
-
Fix/simplify some validation steps webmachinelearning/webnn#643, by @inexorabletash.
-
gather(): Address indices validation and other algorithm nits webmachinelearning/webnn#642, by @inexorabletash.
-
Introduce "valid dimension", used as needed when calculating operand shapes webmachinelearning/webnn#641, by @inexorabletash.
-
Float16Array has landed in Web IDL - remove workarounds webmachinelearning/webnn#640, by @inexorabletash.
-
BugFix: `compute()` explicitly rejects a promise if buffer transferring fails webmachinelearning/webnn#639, by @huningxin.
-
Editorial: Make "generically emulated" text a macro, update wording webmachinelearning/webnn#638, by @inexorabletash.
-
Decompositions for reduceLogSum, reduceLogSumExp, and reduceSumSquare webmachinelearning/webnn#637, by @inexorabletash.
-
README: Fix badge image URL webmachinelearning/webnn#635, by @inexorabletash.
-
add a note for empty input webmachinelearning/webnn#632, by @philloooo.
-
Validate restriction of output padding in convTranspose2d webmachinelearning/webnn#631, by @inexorabletash.
-
-
21 pull requests merged.
-
Add note about no-op graphs webmachinelearning/webnn#665.
-
Bugfix: prelu should validate slope operand webmachinelearning/webnn#661.
-
Remove sequential filling overload of constant() webmachinelearning/webnn#656.
-
Specify the operand data type constraints of operations webmachinelearning/webnn#646.
-
Add axis argument to softmax() webmachinelearning/webnn#649.
-
gru/lstm: Fix missing/incorrect variables and options use webmachinelearning/webnn#655.
-
Remove softplus() steepness option webmachinelearning/webnn#651.
-
Editorial: Add note about keepDimensions of reduction ops webmachinelearning/webnn#648.
-
Fix/simplify some validation steps webmachinelearning/webnn#643.
-
gather(): Address indices validation and other algorithm nits webmachinelearning/webnn#642.
-
Validate restriction of output padding in convTranspose2d webmachinelearning/webnn#631.
-
Revise graph resource validation webmachinelearning/webnn#622.
-
Editorial: Make "generically emulated" text a macro, update wording webmachinelearning/webnn#638.
-
BugFix: `compute()` explicitly rejects a promise if buffer transferring fails webmachinelearning/webnn#639.
-
Introduce "valid dimension", used as needed when calculating operand shapes webmachinelearning/webnn#641.
-
Decompositions for reduceLogSum, reduceLogSumExp, and reduceSumSquare webmachinelearning/webnn#637.
-
Float16Array has landed in Web IDL - remove workarounds webmachinelearning/webnn#640.
-
README: Fix badge image URL webmachinelearning/webnn#635.
-
Define Gelu operation webmachinelearning/webnn#628.
-
Conventions: Use "rank" for variable names, when appropriate webmachinelearning/webnn#627.
-
add a note for empty input webmachinelearning/webnn#632.
-
-
27 pull requests received 193 new comments.
-
Add axis argument to softmax() webmachinelearning/webnn#649 (43 comments by @fdwr, @huningxin, @inexorabletash)
-
Specify the operand data type constraints of operations webmachinelearning/webnn#646 (19 comments by @fdwr, @huningxin, @inexorabletash)
-
Introduce MLNumber for specifying numeric inputs of any type webmachinelearning/webnn#647 (17 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Fix/simplify some validation steps webmachinelearning/webnn#643 (14 comments by @fdwr, @inexorabletash, @zolkis)
-
gather(): Address indices validation and other algorithm nits webmachinelearning/webnn#642 (14 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Introduce "valid dimension", used as needed when calculating operand shapes webmachinelearning/webnn#641 (13 comments by @fdwr, @huningxin, @inexorabletash)
-
Simplify, correct, and add validation for GRU/LSTM and friends webmachinelearning/webnn#659 (10 comments by @huningxin, @inexorabletash)
-
Float16Array has landed in Web IDL - remove workarounds webmachinelearning/webnn#640 (7 comments by @fdwr, @huningxin, @inexorabletash)
-
Validate the hidden size of GRU and LSTM operators webmachinelearning/webnn#644 (6 comments by @huningxin, @inexorabletash, @zolkis)
-
BugFix: `compute()` explicitly rejects a promise if buffer transferring fails webmachinelearning/webnn#639 (6 comments by @fdwr, @huningxin, @inexorabletash)
-
Fix unidirectionally broadcast the shapes steps webmachinelearning/webnn#663 (5 comments by @fdwr, @huningxin, @inexorabletash)
-
Remove softplus() steepness option webmachinelearning/webnn#651 (5 comments by @a-sully, @BruceDai, @fdwr, @Honry, @huningxin)
-
Revise graph resource validation webmachinelearning/webnn#622 (5 comments by @fdwr, @huningxin, @inexorabletash)
-
Remove sequential filling overload of constant() webmachinelearning/webnn#656 (4 comments by @a-sully, @fdwr, @inexorabletash)
-
Make operand data type and rank validation table-driven webmachinelearning/webnn#657 (2 comments by @inexorabletash, @zolkis)
-
Bugfix: prelu should validate slope operand webmachinelearning/webnn#661 (2 comments by @huningxin, @inexorabletash)
-
gru/lstm: Fix missing/incorrect variables and options use webmachinelearning/webnn#655 (2 comments by @inexorabletash)
-
Editorial: Add note about keepDimensions of reduction ops webmachinelearning/webnn#648 (2 comments by @fdwr, @inexorabletash)
-
Editorial: Make "generically emulated" text a macro, update wording webmachinelearning/webnn#638 (2 comments by @huningxin, @inexorabletash)
-
Decompositions for reduceLogSum, reduceLogSumExp, and reduceSumSquare webmachinelearning/webnn#637 (2 comments by @fdwr, @huningxin)
-
Validate restriction of output padding in convTranspose2d webmachinelearning/webnn#631 (2 comments by @inexorabletash)
-
Conventions: Use "rank" for variable names, when appropriate webmachinelearning/webnn#627 (2 comments by @fdwr, @inexorabletash)
-
Swap parameters of scalar constant() operand method webmachinelearning/webnn#650 (1 comment by @inexorabletash)
-
TR publications
-
14 documents published.
-
Web Neural Network API published 14 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
2 participants joined.
-
Vasil Dedejski (Netcetera) joined.
-
Ilya Rezvov (Google LLC) joined.
-
-
1 participant left.
-
Vivek Sekhar (Google LLC) left.
-
-
1 organization joined.
-
Netcetera joined.
-
GitHub
webmachinelearning/webnn
-
12 issues opened.
-
Need Gelu operation webmachinelearning/webnn#626, by @mingmingtasd.
-
Need clarify the maximum limit of the hidden size of the Gru operator webmachinelearning/webnn#625, by @miaobin.
-
Need clarify the usage of axes=[0,1] for resample2d webmachinelearning/webnn#624, by @BruceDai.
-
WebNN should support NPU and QDQ operations webmachinelearning/webnn#623, by @wchao1115.
-
Graph with no input webmachinelearning/webnn#615, by @philloooo.
-
Allow no-op graphs? webmachinelearning/webnn#614, by @inexorabletash.
-
Alternate argumentdef styling ? webmachinelearning/webnn#612, by @inexorabletash.
-
Need clarify scale factor for resample2d webmachinelearning/webnn#610, by @BruceDai.
-
The steps of conv2d and convTranspose2d have issues webmachinelearning/webnn#607, by @BruceDai.
-
Broken link webmachinelearning/webnn#604, by @xfq.
-
Is "validate graph resources" backwards? webmachinelearning/webnn#602, by @inexorabletash.
-
Update anchor link for compute method webmachinelearning/webnn#595, by @ibelem.
-
-
21 issues closed.
-
Need to define the algorithm for the calculation of the output shape of Resample2d webmachinelearning/webnn#360.
-
Consistency for dictionary member definitions webmachinelearning/webnn#483.
-
Should MLBufferView + MLOperandDescriptor be strongly typed (as a MLTensor)? webmachinelearning/webnn#275.
-
Standardize on × or * for multiplication, and ÷ or / for division in formulas webmachinelearning/webnn#506.
-
Drop "The permissions and context validity have been checked..." notes webmachinelearning/webnn#501.
-
Reshape: what if |newShape| is a scalar number webmachinelearning/webnn#452.
-
API simplification: context owns builder, graph becomes internal slot webmachinelearning/webnn#303.
-
Fix description and formula in pooling operations webmachinelearning/webnn#449.
-
Empty tensors are valid for new shape webmachinelearning/webnn#472.
-
Broken link webmachinelearning/webnn#604.
-
The steps of conv2d and convTranspose2d have issues webmachinelearning/webnn#607.
-
Alternate argumentdef styling ? webmachinelearning/webnn#612.
-
Documenting implementation status webmachinelearning/webnn#223.
-
Complete the build algorithms webmachinelearning/webnn#457.
-
Decide how to specify graph initialization webmachinelearning/webnn#552.
-
Add traversal algorithm to buildSync() webmachinelearning/webnn#448.
-
Clarify the expected behavior for `axes` being empty `[]` webmachinelearning/webnn#493.
-
Update the programming model section webmachinelearning/webnn#549.
-
Update anchor link for compute method webmachinelearning/webnn#595.
-
Consider throwing `TypeError` for invalid input webmachinelearning/webnn#583.
-
Apply [EnforceRange] to integer types? webmachinelearning/webnn#562.
-
-
34 issues received 94 new comments.
-
Allow no-op graphs? webmachinelearning/webnn#614 (15 comments by @a-sully, @bbernhar, @fdwr, @huningxin)
-
Can an MLGraphBuilder be reused? webmachinelearning/webnn#567 (7 comments by @fdwr, @huningxin, @inexorabletash, @reillyeon)
-
Consider adding node `label`s for more diagnosable error messages for async errors. webmachinelearning/webnn#585 (6 comments by @fdwr, @inexorabletash, @philloooo, @zolkis)
-
Need Gelu operation webmachinelearning/webnn#626 (4 comments by @fdwr, @huningxin, @mingmingtasd, @zolkis)
-
Need clarify scale factor for resample2d webmachinelearning/webnn#610 (4 comments by @BruceDai, @huningxin, @inexorabletash)
-
Reshape: what if |newShape| is a scalar number webmachinelearning/webnn#452 (4 comments by @BruceDai, @wacky6, @zolkis)
-
Graph with no input webmachinelearning/webnn#615 (4 comments by @a-sully, @huningxin, @inexorabletash, @wacky6)
-
API simplification: context types, context options, createContext() webmachinelearning/webnn#302 (4 comments by @inexorabletash, @wchao1115)
-
Define the maximum number of operand dimensions (maximum rank) webmachinelearning/webnn#456 (3 comments by @alankelly, @fdwr, @Maratyszcza)
-
Allow 0 size dimensions (dimensions containing a 0 in the list of sizes, not a rank of 0 which is valid) webmachinelearning/webnn#391 (3 comments by @fdwr, @huningxin, @sushraja-msft)
-
Consider alternate styling/linking for method argument definitions webmachinelearning/webnn#574 (3 comments by @fdwr, @inexorabletash)
-
WebNN should support NPU and QDQ operations webmachinelearning/webnn#623 (2 comments by @anssiko, @philloooo)
-
Drop "The permissions and context validity have been checked..." notes webmachinelearning/webnn#501 (2 comments by @inexorabletash, @zolkis)
-
Update the programming model section webmachinelearning/webnn#549 (2 comments by @anssiko, @inexorabletash)
-
Need to define the algorithm for the calculation of the output shape of Resample2d webmachinelearning/webnn#360 (1 comment by @BruceDai)
-
Should MLBufferView + MLOperandDescriptor be strongly typed (as a MLTensor)? webmachinelearning/webnn#275 (1 comment by @a-sully)
-
Is "validate graph resources" backwards? webmachinelearning/webnn#602 (1 comment by @inexorabletash)
-
Consistency for dictionary member definitions webmachinelearning/webnn#483 (1 comment by @inexorabletash)
-
API simplification: context owns builder, graph becomes internal slot webmachinelearning/webnn#303 (1 comment by @zolkis)
-
Fix description and formula in pooling operations webmachinelearning/webnn#449 (1 comment by @zolkis)
-
The steps of conv2d and convTranspose2d have issues webmachinelearning/webnn#607 (1 comment by @huningxin)
-
Implement web platform tests for input validation steps webmachinelearning/webnn#545 (1 comment by @inexorabletash)
-
Synchronously validate input operands/activations webmachinelearning/webnn#572 (1 comment by @inexorabletash)
-
Add "implementation consideration" about how out-of-bound indices of Gather/Scatter should be handled webmachinelearning/webnn#486 (1 comment by @a-sully)
-
Rename inputSize variables as inputRank in algorithms webmachinelearning/webnn#588 (1 comment by @inexorabletash)
-
-
24 pull requests opened.
-
Define Gelu operation webmachinelearning/webnn#628, by @mingmingtasd.
-
Conventions: Use "rank" for variable names, when appropriate webmachinelearning/webnn#627, by @inexorabletash.
-
Revise graph resource validation webmachinelearning/webnn#622, by @inexorabletash.
-
Conventions: Ensure all dict members have definitions webmachinelearning/webnn#621, by @inexorabletash.
-
Bug fix: Drop "... have been checked..." notes webmachinelearning/webnn#620, by @inexorabletash.
-
Bugfix: Drop "re-throw" in MLActivation creation steps webmachinelearning/webnn#619, by @inexorabletash.
-
Conventions: Add and apply a few more spec coding conventions webmachinelearning/webnn#618, by @inexorabletash.
-
Update NSNet2 reference webmachinelearning/webnn#617, by @anssiko.
-
Bugfix: Unbalanced autolink brackets webmachinelearning/webnn#616, by @inexorabletash.
-
Programming Model grammar fix webmachinelearning/webnn#613, by @anssiko.
-
Bugfix: Updated the step of calculating output sizes by scales for resample2d webmachinelearning/webnn#611, by @BruceDai.
-
Bugfix: Add missing input operands and activations webmachinelearning/webnn#609, by @inexorabletash.
-
BugFix: Fix algorithm steps of conv2d and convTranspose2d webmachinelearning/webnn#608, by @huningxin.
-
Drop definitions for "platform operator" and "platform operand" webmachinelearning/webnn#606, by @inexorabletash.
-
Synchronously validate input operands/validations webmachinelearning/webnn#605, by @inexorabletash.
-
Content: Define build() steps more rigorously webmachinelearning/webnn#603, by @inexorabletash.
-
reshape(): Remove "nullable" in prose description of newShape webmachinelearning/webnn#601, by @inexorabletash.
-
Content: Define output for gru() webmachinelearning/webnn#600, by @inexorabletash.
-
Content: Clarify the expected behavior for axes being empty webmachinelearning/webnn#599, by @inexorabletash.
-
Content: Introduce emulation section, move reshape examples there webmachinelearning/webnn#598, by @inexorabletash.
-
Bugfix: Duplicate word, variable name consistency webmachinelearning/webnn#597, by @inexorabletash.
-
Update anchor link for compute() after dropping synchronous execution support webmachinelearning/webnn#596, by @ibelem.
-
Styles: simplify algorithm markup, improve step & dfn visibility webmachinelearning/webnn#594, by @inexorabletash.
-
Define task source, construct with realms, improve parallel usage webmachinelearning/webnn#593, by @inexorabletash.
-
-
23 pull requests merged.
-
Conventions: Ensure all dict members have definitions webmachinelearning/webnn#621.
-
Conventions: Add and apply a few more spec coding conventions webmachinelearning/webnn#618.
-
Bug fix: Drop "... have been checked..." notes webmachinelearning/webnn#620.
-
Bugfix: Drop "re-throw" in MLActivation creation steps webmachinelearning/webnn#619.
-
Update NSNet2 reference webmachinelearning/webnn#617.
-
BugFix: Fix algorithm steps of conv2d and convTranspose2d webmachinelearning/webnn#608.
-
Bugfix: Unbalanced autolink brackets webmachinelearning/webnn#616.
-
Programming Model grammar fix webmachinelearning/webnn#613.
-
Bugfix: Updated the step of calculating output sizes by scales for resample2d webmachinelearning/webnn#611.
-
Drop definitions for "platform operator" and "platform operand" webmachinelearning/webnn#606.
-
Bugfix: Add missing input operands and activations webmachinelearning/webnn#609.
-
Synchronously validate input operands/validations webmachinelearning/webnn#605.
-
Content: Define build() steps more rigorously webmachinelearning/webnn#603.
-
Content: Clarify the expected behavior for axes being empty webmachinelearning/webnn#599.
-
Content: Define output for gru() webmachinelearning/webnn#600.
-
Styles: simplify algorithm markup, improve step & dfn visibility webmachinelearning/webnn#594.
-
reshape(): Remove "nullable" in prose description of newShape webmachinelearning/webnn#601.
-
Content: Define operand concept, simplify graph connection steps webmachinelearning/webnn#591.
-
Bugfix: Duplicate word, variable name consistency webmachinelearning/webnn#597.
-
Update anchor link for compute() after dropping synchronous execution support webmachinelearning/webnn#596.
-
Define task source, construct with realms, improve parallel usage webmachinelearning/webnn#593.
-
Throw TypeError for all input validation webmachinelearning/webnn#589.
-
Apply [EnforceRange] to all integer types webmachinelearning/webnn#592.
-
-
24 pull requests received 227 new comments.
-
Content: Define operand concept, simplify graph connection steps webmachinelearning/webnn#591 (58 comments by @anssiko, @fdwr, @huningxin, @inexorabletash)
-
BugFix: Fix algorithm steps of conv2d and convTranspose2d webmachinelearning/webnn#608 (31 comments by @BruceDai, @fdwr, @huningxin, @inexorabletash)
-
Content: Define build() steps more rigorously webmachinelearning/webnn#603 (28 comments by @fdwr, @huningxin, @inexorabletash, @reillyeon)
-
Bugfix: Updated the step of calculating output sizes by scales for resample2d webmachinelearning/webnn#611 (16 comments by @BruceDai, @fdwr, @huningxin, @inexorabletash)
-
Conventions: Ensure all dict members have definitions webmachinelearning/webnn#621 (15 comments by @fdwr, @huningxin, @inexorabletash, @zolkis)
-
Define task source, construct with realms, improve parallel usage webmachinelearning/webnn#593 (13 comments by @a-sully, @huningxin, @inexorabletash, @reillyeon)
-
Bugfix: Add missing input operands and activations webmachinelearning/webnn#609 (9 comments by @fdwr, @inexorabletash)
-
Styles: simplify algorithm markup, improve step & dfn visibility webmachinelearning/webnn#594 (8 comments by @huningxin, @inexorabletash, @zolkis)
-
Revise graph resource validation webmachinelearning/webnn#622 (7 comments by @fdwr, @huningxin, @inexorabletash, @zolkis)
-
Throw TypeError for all input validation webmachinelearning/webnn#589 (5 comments by @a-sully, @fdwr, @huningxin)
-
Conventions: Add and apply a few more spec coding conventions webmachinelearning/webnn#618 (4 comments by @fdwr, @inexorabletash)
-
Bugfix: Drop "re-throw" in MLActivation creation steps webmachinelearning/webnn#619 (4 comments by @huningxin, @inexorabletash, @zolkis)
-
Bugfix: Unbalanced autolink brackets webmachinelearning/webnn#616 (4 comments by @anssiko, @inexorabletash)
-
Bug fix: Drop "... have been checked..." notes webmachinelearning/webnn#620 (3 comments by @fdwr, @inexorabletash, @zolkis)
-
Synchronously validate input operands/validations webmachinelearning/webnn#605 (3 comments by @huningxin)
-
Content: Clarify the expected behavior for axes being empty webmachinelearning/webnn#599 (3 comments by @BruceDai, @inexorabletash)
-
Conventions: Use "rank" for variable names, when appropriate webmachinelearning/webnn#627 (2 comments by @fdwr, @inexorabletash)
-
Drop definitions for "platform operator" and "platform operand" webmachinelearning/webnn#606 (2 comments by @huningxin, @inexorabletash)
-
Content: Define output for gru() webmachinelearning/webnn#600 (2 comments by @fdwr, @inexorabletash)
-
reshape(): Remove "nullable" in prose description of newShape webmachinelearning/webnn#601 (2 comments by @fdwr, @inexorabletash)
-
Content: Introduce emulation section, move reshape examples there webmachinelearning/webnn#598 (2 comments by @inexorabletash, @zolkis)
-
Bugfix: Duplicate word, variable name consistency webmachinelearning/webnn#597 (1 comment by @inexorabletash)
-
TR publications
-
14 documents published.
-
Web Neural Network API published 14 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
7 participants joined.
-
Mike Wyrzykowski (Apple Inc.) joined.
-
Anne van Kesteren (Apple Inc.) joined.
-
Benjamin Poulain (Apple Inc.) joined.
-
Benjamin Poulain (Apple Inc.) joined.
-
Cameron McCormack (Apple Inc.) joined.
-
Marcos Caceres (Apple Inc.) joined.
-
Theresa O'Connor (Apple Inc.) joined.
-
-
1 organization joined.
-
Apple Inc. joined.
-
GitHub
webmachinelearning/webnn
-
17 issues opened.
-
Consider adopting new broadcasting rules webmachinelearning/webnn#590, by @a-sully.
-
Rename inputSize variables as inputRank in algorithms webmachinelearning/webnn#588, by @inexorabletash.
-
Consider adding node `label`s for more diagnosable error messages for async errors. webmachinelearning/webnn#585, by @philloooo.
-
Consider throwing `TypeError` for invalid input webmachinelearning/webnn#583, by @a-sully.
-
Revisiting use of partial interfaces webmachinelearning/webnn#579, by @inexorabletash.
-
Consider alternate styling/linking for method argument definitions webmachinelearning/webnn#574, by @inexorabletash.
-
Core operator set webmachinelearning/webnn#573, by @philloooo.
-
Synchronously validate input operands/activations webmachinelearning/webnn#572, by @inexorabletash.
-
WebIDL definition for constant(start, end, step, type) is missing webmachinelearning/webnn#571, by @inexorabletash.
-
Can an MLGraphBuilder be reused? webmachinelearning/webnn#567, by @reillyeon.
-
Don't transfer input ArrayBuffers webmachinelearning/webnn#566, by @reillyeon.
-
Apply [EnforceRange] to integer types? webmachinelearning/webnn#562, by @inexorabletash.
-
Consider removing `MLBufferView` in its current form webmachinelearning/webnn#560, by @a-sully.
-
Control flow operations: if, while webmachinelearning/webnn#559, by @philloooo.
-
Decide how to specify graph initialization webmachinelearning/webnn#552, by @a-sully.
-
Deprecated squeeze with emulation path is missing its own section webmachinelearning/webnn#551, by @anssiko.
-
Update the programming model section webmachinelearning/webnn#549, by @anssiko.
-
-
24 issues closed.
-
Revisiting use of partial interfaces webmachinelearning/webnn#579.
-
Define "calculating output dimensions" webmachinelearning/webnn#500.
-
Define the algorithm of calculating the effective padding for "same-upper" and "same-lower" option webmachinelearning/webnn#326.
-
Should validate MLGraph.[[context]] in MLContext.compute() and MLContext.computeSync() steps webmachinelearning/webnn#341.
-
OperandType of gemm / matmul return webmachinelearning/webnn#84.
-
Implement webnn-baseline and WPT for operations needed for well-known transformers webmachinelearning/webnn#485.
-
Clarify intermediate values (activations) vs MLActivation vs operations webmachinelearning/webnn#335.
-
Consider removing `MLBufferView` in its current form webmachinelearning/webnn#560.
-
concat() assertions around input shape/datatypes seem invalid webmachinelearning/webnn#503.
-
Convert "If descriptor.dimensions does not exist, then descriptor defines a scalar input." from assertion to note webmachinelearning/webnn#502.
-
Use web spec best practices webmachinelearning/webnn#450.
-
Define algorithms for dictionaries with lists as default values webmachinelearning/webnn#211.
-
Use deep copy for setting internal slots? webmachinelearning/webnn#395.
-
Subclass MLGraph based on the context that creates it webmachinelearning/webnn#344.
-
MLOperand::dimensions() or MLGraphBuilder::shape() for debugging purposes and diagnostics webmachinelearning/webnn#379.
-
Drop the support of synchronous execution webmachinelearning/webnn#531.
-
0D scalars webmachinelearning/webnn#390.
-
The input of `triangular` should be at least 2D tensor webmachinelearning/webnn#494.
-
Consider removing `MLCommandEncoder` webmachinelearning/webnn#528.
-
Define power preference and device type in terms of WebIDL enums webmachinelearning/webnn#497.
-
Reserve asserts only for non-obvious things webmachinelearning/webnn#455.
-
Use modern WebIDL and Infra standard conventions webmachinelearning/webnn#210.
-
OperandDescriptor's dimension should match ECMAScript's TypedArray dimension webmachinelearning/webnn#29.
-
Validate op arguments early to better support eager execution webmachinelearning/webnn#234.
-
-
2 issues reopened.
-
Candidate Recommendation readiness tracker webmachinelearning/webnn#240, by @anssiko.
-
Wide review tracker webmachinelearning/webnn#239, by @anssiko.
-
-
41 issues received 103 new comments.
-
Synchronously validate input operands/activations webmachinelearning/webnn#572 (11 comments by @a-sully, @huningxin, @inexorabletash)
-
Deprecated squeeze with emulation path is missing its own section webmachinelearning/webnn#551 (6 comments by @anssiko, @huningxin, @inexorabletash)
-
Constant sequential filling operation needs output shape parameter webmachinelearning/webnn#492 (5 comments by @fdwr, @inexorabletash, @miaobin)
-
Update the programming model section webmachinelearning/webnn#549 (5 comments by @anssiko, @inexorabletash)
-
Consider adding node `label`s for more diagnosable error messages for async errors. webmachinelearning/webnn#585 (4 comments by @inexorabletash, @philloooo, @zolkis)
-
Candidate Recommendation readiness tracker webmachinelearning/webnn#240 (4 comments by @anssiko, @dontcallmedom)
-
Revisiting use of partial interfaces webmachinelearning/webnn#579 (4 comments by @anssiko, @inexorabletash, @zolkis)
-
WebIDL definition for constant(start, end, step, type) is missing webmachinelearning/webnn#571 (4 comments by @huningxin, @inexorabletash, @zolkis)
-
Don't transfer input ArrayBuffers webmachinelearning/webnn#566 (3 comments by @huningxin, @reillyeon)
-
API simplification: context types, context options, createContext() webmachinelearning/webnn#302 (3 comments by @huningxin, @inexorabletash)
-
Subclass MLGraph based on the context that creates it webmachinelearning/webnn#344 (3 comments by @huningxin, @inexorabletash, @zolkis)
-
Type of some parameters should match the input data type webmachinelearning/webnn#442 (3 comments by @fdwr, @inexorabletash)
-
Add traversal algorithm to buildSync() webmachinelearning/webnn#448 (3 comments by @anssiko, @inexorabletash)
-
How to define the algorithm of L2_Pool2d? webmachinelearning/webnn#278 (3 comments by @fdwr, @mingmingtasd)
-
WPT tests tracker / 2 for remaining ops webmachinelearning/webnn#338 (3 comments by @anssiko, @BruceDai)
-
Should validate MLGraph.[[context]] in MLContext.compute() and MLContext.computeSync() steps webmachinelearning/webnn#341 (2 comments by @anssiko, @inexorabletash)
-
Clarify intermediate values (activations) vs MLActivation vs operations webmachinelearning/webnn#335 (2 comments by @inexorabletash, @zolkis)
-
Use deep copy for setting internal slots? webmachinelearning/webnn#395 (2 comments by @inexorabletash, @zolkis)
-
Reshape: what if |newShape| is a scalar number webmachinelearning/webnn#452 (2 comments by @huningxin, @inexorabletash)
-
Standardize on × or * for multiplication, and ÷ or / for division in formulas webmachinelearning/webnn#506 (2 comments by @fdwr, @inexorabletash)
-
Implement web platform tests for input validation steps webmachinelearning/webnn#545 (2 comments by @BruceDai)
-
Implement webnn-baseline and WPT for operations needed for well-known transformers webmachinelearning/webnn#485 (1 comment by @anssiko)
-
reduceLogSum, reduceLogSumExp, reduceSumSquare are not supported on OV/NNAPI webmachinelearning/webnn#209 (1 comment by @inexorabletash)
-
Define algorithms for dictionaries with lists as default values webmachinelearning/webnn#211 (1 comment by @inexorabletash)
-
webmachinelearning/webnn#84 (1 comment by @inexorabletash)
-
Consider alternate styling/linking for method argument definitions webmachinelearning/webnn#574 (1 comment by @zolkis)
-
MLOperand::dimensions() or MLGraphBuilder::shape() for debugging purposes and diagnostics webmachinelearning/webnn#379 (1 comment by @inexorabletash)
-
Reserve asserts only for non-obvious things webmachinelearning/webnn#455 (1 comment by @inexorabletash)
-
Use modern WebIDL and Infra standard conventions webmachinelearning/webnn#210 (1 comment by @anssiko)
-
Fix description and formula in pooling operations webmachinelearning/webnn#449 (1 comment by @anssiko)
-
webmachinelearning/webnn#29 (1 comment by @anssiko)
-
Validate op arguments early to better support eager execution webmachinelearning/webnn#234 (1 comment by @anssiko)
-
-
28 pull requests opened.
-
Apply [EnforceRange] to all integer types webmachinelearning/webnn#592, by @a-sully.
-
Content: Define operand concept, simplify graph connection steps webmachinelearning/webnn#591, by @inexorabletash.
-
Throw TypeError for all input validation webmachinelearning/webnn#589, by @a-sully.
-
Remove MLAutoPad webmachinelearning/webnn#587, by @huningxin.
-
Style: Only use partial interfaces for MLGraphBuilder webmachinelearning/webnn#586, by @inexorabletash.
-
Bug fix: Validate MLGraph's context in MLContext's compute() webmachinelearning/webnn#584, by @inexorabletash.
-
Content: Specify output size calculations webmachinelearning/webnn#582, by @inexorabletash.
-
Bug fix: Correct an unclosed autolink webmachinelearning/webnn#581, by @inexorabletash.
-
Style: Turn on "Assume Explicit For" and scope autolinks webmachinelearning/webnn#580, by @inexorabletash.
-
Bug fix: gather() steps shouldn't invoke methods webmachinelearning/webnn#578, by @inexorabletash.
-
Bug fix: Make concat() validation and initialization more precise. webmachinelearning/webnn#577, by @inexorabletash.
-
Style: Make grouped algorithm steps consistent webmachinelearning/webnn#576, by @inexorabletash.
-
Bug fix: Fix MLGraphBuilder.input()'s handling of scalars webmachinelearning/webnn#575, by @inexorabletash.
-
Style: Prefer characters to entities webmachinelearning/webnn#570, by @inexorabletash.
-
Remove MLBufferView webmachinelearning/webnn#569, by @a-sully.
-
Update Triage Guidance webmachinelearning/webnn#568, by @anssiko.
-
Wording: Use shorter possessive form for abstract object properties webmachinelearning/webnn#565, by @inexorabletash.
-
Fix Bikeshed build errors webmachinelearning/webnn#564, by @anssiko.
-
Bug fix: Fix linking for a handful of property references webmachinelearning/webnn#563, by @inexorabletash.
-
Editor change webmachinelearning/webnn#561, by @anssiko.
-
Bug fix: Missing variable name in conv2d webmachinelearning/webnn#558, by @inexorabletash.
-
Remove hard-coded colors which break dark mode webmachinelearning/webnn#557, by @reillyeon.
-
Style change: Scope the links for MLContext internal slots webmachinelearning/webnn#556, by @inexorabletash.
-
Content change: Add definitions for MLOperand's shape and dataType webmachinelearning/webnn#555, by @inexorabletash.
-
Acknowledge more people webmachinelearning/webnn#554, by @anssiko.
-
Remove MLBufferResourceView webmachinelearning/webnn#553, by @a-sully.
-
Content: The input of triangular should be at least 2D tensor webmachinelearning/webnn#550, by @inexorabletash.
-
Drop the support of synchronous execution webmachinelearning/webnn#548, by @huningxin.
-
-
38 pull requests merged.
-
Style: Only use partial interfaces for MLGraphBuilder webmachinelearning/webnn#586.
-
Content: Specify output size calculations webmachinelearning/webnn#582.
-
Remove MLAutoPad webmachinelearning/webnn#587.
-
Bug fix: Validate MLGraph's context in MLContext's compute() webmachinelearning/webnn#584.
-
Bug fix: Correct an unclosed autolink webmachinelearning/webnn#581.
-
Style: Turn on "Assume Explicit For" and scope autolinks webmachinelearning/webnn#580.
-
Remove MLBufferView webmachinelearning/webnn#569.
-
Bug fix: gather() steps shouldn't invoke methods webmachinelearning/webnn#578.
-
Bug fix: Make concat() validation and initialization more precise. webmachinelearning/webnn#577.
-
Style: Make grouped algorithm steps consistent webmachinelearning/webnn#576.
-
Bug fix: Fix MLGraphBuilder.input()'s handling of scalars webmachinelearning/webnn#575.
-
Style: Prefer characters to entities webmachinelearning/webnn#570.
-
Wording: Use shorter possessive form for abstract object properties webmachinelearning/webnn#565.
-
Update Triage Guidance webmachinelearning/webnn#568.
-
Bug fix: Link operator names to the MLGraphBuilder methods webmachinelearning/webnn#547.
-
Bug fix: Fix linking for a handful of property references webmachinelearning/webnn#563.
-
Remove MLBufferResourceView webmachinelearning/webnn#553.
-
Drop the support of synchronous execution webmachinelearning/webnn#548.
-
Process: Add documentation for labels, current and proposed webmachinelearning/webnn#533.
-
Wording change: Tighten up output shape calculation algorithms webmachinelearning/webnn#523.
-
Content change: Add definitions for MLOperand's shape and dataType webmachinelearning/webnn#555.
-
Update Status of this document for CR Snapshot webmachinelearning/webnn#532.
-
New content: Add definition for shape broadcasting webmachinelearning/webnn#534.
-
Fix dimensions for 0D scalars webmachinelearning/webnn#476.
-
Fix Bikeshed build errors webmachinelearning/webnn#564.
-
Bug fix: Provide default for MLLayerNormalizationOptions axes in steps webmachinelearning/webnn#538.
-
Content: The input of triangular should be at least 2D tensor webmachinelearning/webnn#550.
-
Bug fix: Optional sequences shouldn't be defaulted to null webmachinelearning/webnn#539.
-
Remove MLCommandEncoder webmachinelearning/webnn#546.
-
Editor change webmachinelearning/webnn#561.
-
Style change: Scope the links for MLContext internal slots webmachinelearning/webnn#556.
-
Bug fix: Missing variable name in conv2d webmachinelearning/webnn#558.
-
Remove hard-coded colors which break dark mode webmachinelearning/webnn#557.
-
Wording change: Add types to all internal algorithm members webmachinelearning/webnn#536.
-
Wording change: Dedupe dictionary member type and default definitions webmachinelearning/webnn#526.
-
Wording change: Define power prefs & device type using WebIDL enums webmachinelearning/webnn#535.
-
Acknowledge more people webmachinelearning/webnn#554.
-
Wording change: drop links to File API's "object", define platform operand/operator webmachinelearning/webnn#540.
-
-
34 pull requests received 317 new comments.
-
Content: Specify output size calculations webmachinelearning/webnn#582 (30 comments by @anssiko, @fdwr, @huningxin, @inexorabletash, @zolkis)
-
Process: Add documentation for labels, current and proposed webmachinelearning/webnn#533 (25 comments by @anssiko, @fdwr, @huningxin, @inexorabletash)
-
Add MLBuffer exploration doc webmachinelearning/webnn#541 (23 comments by @a-sully, @anssiko, @bbernhar, @huningxin, @RafaelCintron, @zolkis)
-
Bug fix: Fix MLGraphBuilder.input()'s handling of scalars webmachinelearning/webnn#575 (22 comments by @fdwr, @huningxin, @inexorabletash)
-
Drop the support of synchronous execution webmachinelearning/webnn#548 (22 comments by @anssiko, @huningxin, @inexorabletash, @zolkis)
-
Fix dimensions for 0D scalars webmachinelearning/webnn#476 (22 comments by @a-sully, @fdwr, @inexorabletash, @zolkis)
-
New content: Add definition for shape broadcasting webmachinelearning/webnn#534 (16 comments by @fdwr, @huningxin, @inexorabletash)
-
Throw TypeError for all input validation webmachinelearning/webnn#589 (15 comments by @a-sully, @huningxin)
-
Remove MLBufferResourceView webmachinelearning/webnn#553 (15 comments by @a-sully, @anssiko, @bbernhar, @huningxin)
-
Content change: Add definitions for MLOperand's shape and dataType webmachinelearning/webnn#555 (15 comments by @fdwr, @huningxin, @inexorabletash, @zolkis)
-
Bug fix: Provide default for MLLayerNormalizationOptions axes in steps webmachinelearning/webnn#538 (13 comments by @fdwr, @inexorabletash)
-
Remove MLCommandEncoder webmachinelearning/webnn#546 (12 comments by @a-sully, @anssiko, @huningxin)
-
Bug fix: Link operator names to the MLGraphBuilder methods webmachinelearning/webnn#547 (12 comments by @huningxin, @inexorabletash, @zolkis)
-
Content: The input of triangular should be at least 2D tensor webmachinelearning/webnn#550 (10 comments by @anssiko, @fdwr, @inexorabletash)
-
Wording change: Tighten up output shape calculation algorithms webmachinelearning/webnn#523 (9 comments by @a-sully, @fdwr, @huningxin, @inexorabletash)
-
Update Status of this document for CR Snapshot webmachinelearning/webnn#532 (9 comments by @anssiko, @dontcallmedom, @huningxin)
-
Style: Make grouped algorithm steps consistent webmachinelearning/webnn#576 (6 comments by @fdwr, @huningxin, @inexorabletash)
-
Wording change: Dedupe dictionary member type and default definitions webmachinelearning/webnn#526 (5 comments by @anssiko, @huningxin, @inexorabletash)
-
Style: Only use partial interfaces for MLGraphBuilder webmachinelearning/webnn#586 (4 comments by @anssiko, @inexorabletash, @zolkis)
-
Bug fix: Make concat() validation and initialization more precise. webmachinelearning/webnn#577 (4 comments by @fdwr, @huningxin, @inexorabletash)
-
Content: Define operand concept, simplify graph connection steps webmachinelearning/webnn#591 (3 comments by @inexorabletash, @zolkis)
-
Bug fix: Correct an unclosed autolink webmachinelearning/webnn#581 (3 comments by @fdwr, @inexorabletash)
-
Wording: Use shorter possessive form for abstract object properties webmachinelearning/webnn#565 (3 comments by @anssiko, @inexorabletash)
-
Wording change: Add types to all internal algorithm members webmachinelearning/webnn#536 (3 comments by @anssiko, @huningxin, @inexorabletash)
-
Bug fix: Validate MLGraph's context in MLContext's compute() webmachinelearning/webnn#584 (2 comments by @anssiko, @fdwr)
-
Style: Turn on "Assume Explicit For" and scope autolinks webmachinelearning/webnn#580 (2 comments by @inexorabletash)
-
Bug fix: Optional sequences shouldn't be defaulted to null webmachinelearning/webnn#539 (2 comments by @fdwr)
-
Wording change: drop links to File API's "object", define platform operand/operator webmachinelearning/webnn#540 (2 comments by @huningxin, @inexorabletash)
-
Apply [EnforceRange] to all integer types webmachinelearning/webnn#592 (1 comment by @inexorabletash)
-
Bug fix: gather() steps shouldn't invoke methods webmachinelearning/webnn#578 (1 comment by @inexorabletash)
-
Remove hard-coded colors which break dark mode webmachinelearning/webnn#557 (1 comment by @reillyeon)
-
TR publications
-
9 documents published.
-
Web Neural Network API published 8 times as a Candidate Recommendation Draft. History of Web Neural Network API
-
Ethical Principles for Web Machine Learning published as a Draft Note. History of Ethical Principles for Web Machine Learning
-
Past meetings
Participants
-
1 participant joined.
-
Geoff Gustafson (Intel Corporation) joined.
-
GitHub
webmachinelearning/webnn
-
17 issues opened.
-
Implement web platform tests for input validation steps webmachinelearning/webnn#545, by @huningxin.
-
[MLBuffer] Support for MLBuffer in graph execution webmachinelearning/webnn#544, by @bbernhar.
-
[MLBuffer] Uploading/downloading tensor data webmachinelearning/webnn#543, by @bbernhar.
-
[MLBuffer] Creation and representing MLBuffer on a XPU devices webmachinelearning/webnn#542, by @bbernhar.
-
Vision transformers webmachinelearning/webnn#537, by @bhack.
-
Drop the support of synchronous execution webmachinelearning/webnn#531, by @huningxin.
-
Specify WebNN timelines webmachinelearning/webnn#529, by @a-sully.
-
Consider removing `MLCommandEncoder` webmachinelearning/webnn#528, by @a-sully.
-
SSMs/Mamba models webmachinelearning/webnn#524, by @bhack.
-
Standardize on × or * for multiplication, and ÷ or / for division in formulas webmachinelearning/webnn#506, by @inexorabletash.
-
Supported operations query mechanism webmachinelearning/webnn#504, by @ibelem.
-
concat() assertions around input shape/datatypes seem invalid webmachinelearning/webnn#503, by @inexorabletash.
-
Convert "If descriptor.dimensions does not exist, then descriptor defines a scalar input." from assertion to note webmachinelearning/webnn#502, by @inexorabletash.
-
Drop "The permissions and context validity have been checked..." notes webmachinelearning/webnn#501, by @inexorabletash.
-
Define "calculating output dimensions" webmachinelearning/webnn#500, by @inexorabletash.
-
Streamline context creation and validation algorithm definitions webmachinelearning/webnn#498, by @inexorabletash.
-
Define power preference and device type in terms of WebIDL enums webmachinelearning/webnn#497, by @inexorabletash.
-
-
13 issues closed.
-
Request for argMax op webmachinelearning/webnn#184.
-
Need for type casting? webmachinelearning/webnn#284.
-
The type of MLTransposeOptions.permutation should be unsigned long webmachinelearning/webnn#317.
-
Add the clamp() algorithm webmachinelearning/webnn#347.
-
Improve batchNorm webmachinelearning/webnn#334.
-
Streamline context creation and validation algorithm definitions webmachinelearning/webnn#498.
-
Clarify/simplify graph execution on GPU / MLCommandEncoder webmachinelearning/webnn#333.
-
CommandBuffer usage clarification: internal, external, both? webmachinelearning/webnn#264.
-
Remove the support of `null` value in `newShape` parameter of `reshape` build method webmachinelearning/webnn#388.
-
Specify the ModelBuilder.createModel and other ModelBuilder members webmachinelearning/webnn#107.
-
MLOperandDataType naming recommendation webmachinelearning/webnn#299.
-
Unsqueeze operator is missing (so add unsqueeze, or better yet, delete squeeze) webmachinelearning/webnn#296.
-
Support dedicated square root operator webmachinelearning/webnn#438.
-
-
37 issues received 91 new comments.
-
Support for device-based tensor storage objects webmachinelearning/webnn#482 (20 comments by @a-sully, @bbernhar, @RafaelCintron, @reillyeon, @wacky6, @wchao1115)
-
Constant sequential filling operation needs output shape parameter webmachinelearning/webnn#492 (9 comments by @fdwr, @Honry, @huningxin, @miaobin, @wacky6, @zolkis)
-
Drop the support of synchronous execution webmachinelearning/webnn#531 (4 comments by @a-sully, @anssiko, @huningxin, @wacky6)
-
Define algorithms for dictionaries with lists as default values webmachinelearning/webnn#211 (4 comments by @inexorabletash, @zolkis)
-
Use modern WebIDL and Infra standard conventions webmachinelearning/webnn#210 (3 comments by @anssiko, @inexorabletash)
-
CommandBuffer usage clarification: internal, external, both? webmachinelearning/webnn#264 (3 comments by @a-sully, @anssiko, @bbernhar)
-
Integration with real-time video processing webmachinelearning/webnn#226 (3 comments by @alvestrand, @anssiko)
-
Implement web platform tests for input validation steps webmachinelearning/webnn#545 (2 comments by @anssiko, @huningxin)
-
Clarify/simplify graph execution on GPU / MLCommandEncoder webmachinelearning/webnn#333 (2 comments by @a-sully, @anssiko)
-
Use deep copy for setting internal slots? webmachinelearning/webnn#395 (2 comments by @inexorabletash)
-
Reserve asserts only for non-obvious things webmachinelearning/webnn#455 (2 comments by @anssiko, @inexorabletash)
-
Streamline context creation and validation algorithm definitions webmachinelearning/webnn#498 (2 comments by @inexorabletash, @zolkis)
-
webmachinelearning/webnn#29 (1 comment by @inexorabletash)
-
The type of MLTransposeOptions.permutation should be unsigned long webmachinelearning/webnn#317 (1 comment by @inexorabletash)
-
Allow checking whether operators/types are supported for a backend before creating a graph webmachinelearning/webnn#463 (1 comment by @reillyeon)
-
webmachinelearning/webnn#33 (1 comment by @a-sully)
-
Subclass MLGraph based on the context that creates it webmachinelearning/webnn#344 (1 comment by @a-sully)
-
The input of `triangular` should be at least 2D tensor webmachinelearning/webnn#494 (1 comment by @BruceDai)
-
Unsqueeze operator is missing (so add unsqueeze, or better yet, delete squeeze) webmachinelearning/webnn#296 (1 comment by @inexorabletash)
-
Clarify the restriction for `minValue` and `maxValue` of `MLClampOptions` webmachinelearning/webnn#396 (1 comment by @huningxin)
-
Define power preference and device type in terms of WebIDL enums webmachinelearning/webnn#497 (1 comment by @inexorabletash)
-
-
34 pull requests opened.
-
Bug fix: Link operator names to the MLGraphBuilder methods webmachinelearning/webnn#547, by @inexorabletash.
-
Remove MLCommandEncoder webmachinelearning/webnn#546, by @a-sully.
-
Add MLBuffer exploration doc webmachinelearning/webnn#541, by @a-sully.
-
Wording change: drop links to File API's "object", define platform operand/operator webmachinelearning/webnn#540, by @inexorabletash.
-
Bug fix: Optional sequences shouldn't be defaulted to null webmachinelearning/webnn#539, by @inexorabletash.
-
Bug fix: Provide default for MLLayerNormalizationOptions axes in steps webmachinelearning/webnn#538, by @inexorabletash.
-
Wording change: Add types to all internal algorithm members webmachinelearning/webnn#536, by @inexorabletash.
-
Wording change: Define power prefs & device type using WebIDL enums webmachinelearning/webnn#535, by @inexorabletash.
-
New content: Add definition for shape broadcasting webmachinelearning/webnn#534, by @inexorabletash.
-
Process: Add documentation for labels, current and proposed webmachinelearning/webnn#533, by @inexorabletash.
-
Update Status of this document for CR Snapshot webmachinelearning/webnn#532, by @anssiko.
-
Meta: Add a gitignore webmachinelearning/webnn#530, by @a-sully.
-
Bug fix: Correct linking for list/range iterations webmachinelearning/webnn#527, by @inexorabletash.
-
Wording change: Dedupe dictionary member type and default definitions webmachinelearning/webnn#526, by @inexorabletash.
-
Document coding conventions and best practices for spec maintainers webmachinelearning/webnn#525, by @inexorabletash.
-
Wording change: Tighten up output shape calculation algorithms webmachinelearning/webnn#523, by @inexorabletash.
-
Bug fix: Correct indentation of an iteration substep webmachinelearning/webnn#522, by @inexorabletash.
-
Wording change: Tidy up more arrays within algorithms to be lists webmachinelearning/webnn#521, by @inexorabletash.
-
Bug fix: resample2d output size calculation algorithm and usage webmachinelearning/webnn#520, by @inexorabletash.
-
Wording change: Streamline context creation and validation algorithms webmachinelearning/webnn#519, by @inexorabletash.
-
Wording change: Don't use asserts for validating argument types webmachinelearning/webnn#518, by @inexorabletash.
-
Wording change: Use Infra's "Set ... to ..." pattern for properties webmachinelearning/webnn#517, by @inexorabletash.
-
Fix PR Preview webmachinelearning/webnn#516, by @anssiko.
-
Wording change: Prefix literal lists with "the list" webmachinelearning/webnn#515, by @inexorabletash.
-
Stylistic change: Linkify MLOperandDataType enum members webmachinelearning/webnn#514, by @inexorabletash.
-
Wording change: Add missing receiver when assigning to internal slot. webmachinelearning/webnn#513, by @inexorabletash.
-
Wording change: New up a specific type, rather than just object webmachinelearning/webnn#512, by @inexorabletash.
-
Wording change: Define power prefs & device type using WebIDL enums webmachinelearning/webnn#511, by @inexorabletash.
-
Wording change: Remove arg type test for internal validation algorithm webmachinelearning/webnn#510, by @inexorabletash.
-
Stylistic change: Fix indentation of steps validating bias option webmachinelearning/webnn#509, by @inexorabletash.
-
Stylistic change: Linkify MLOperandDataType enum members webmachinelearning/webnn#508, by @inexorabletash.
-
Revise use cases with transformers webmachinelearning/webnn#507, by @anssiko.
-
Bug fix: Remove null value support in reshape()'s newShape parameter webmachinelearning/webnn#505, by @inexorabletash.
-
Stylistic change: Replace some explicit use of link target naming with scopes webmachinelearning/webnn#499, by @inexorabletash.
-
-
20 pull requests merged.
-
Bug fix: resample2d output size calculation algorithm and usage webmachinelearning/webnn#520.
-
Bug fix: Correct linking for list/range iterations webmachinelearning/webnn#527.
-
Wording change: Tidy up more arrays within algorithms to be lists webmachinelearning/webnn#521.
-
Wording change: Don't use asserts for validating argument types webmachinelearning/webnn#518.
-
Wording change: Streamline context creation and validation algorithms webmachinelearning/webnn#519.
-
Meta: Add a gitignore webmachinelearning/webnn#530.
-
Revise use cases with transformers webmachinelearning/webnn#507.
-
Stylistic change: Replace some explicit use of link target naming with scopes webmachinelearning/webnn#499.
-
Document coding conventions and best practices for spec maintainers webmachinelearning/webnn#525.
-
Wording change: Use Infra's "Set ... to ..." pattern for properties webmachinelearning/webnn#517.
-
Wording change: Prefix literal lists with "the list" webmachinelearning/webnn#515.
-
Bug fix: Correct indentation of an iteration substep webmachinelearning/webnn#522.
-
Wording change: New up a specific type, rather than just object webmachinelearning/webnn#512.
-
Bug fix: Remove null value support in reshape()'s newShape parameter webmachinelearning/webnn#505.
-
Fix PR Preview webmachinelearning/webnn#516.
-
Wording change: Remove arg type test for internal validation algorithm webmachinelearning/webnn#510.
-
Wording change: Add missing receiver when assigning to internal slot. webmachinelearning/webnn#513.
-
Stylistic change: Linkify MLOperandDataType enum members webmachinelearning/webnn#514.
-
Stylistic change: Fix indentation of steps validating bias option webmachinelearning/webnn#509.
-
Adopt Bikeshed best practices and Infra spec conventions webmachinelearning/webnn#495.
-
-
3 pull requests closed.
-
Simplify MLContext creation webmachinelearning/webnn#322.
-
Stylistic change: Linkify MLOperandDataType enum members webmachinelearning/webnn#508.
-
Wording change: Define power prefs & device type using WebIDL enums webmachinelearning/webnn#511.
-
-
27 pull requests received 123 new comments.
-
Wording change: Tighten up output shape calculation algorithms webmachinelearning/webnn#523 (15 comments by @huningxin, @inexorabletash)
-
Wording change: drop links to File API's "object", define platform operand/operator webmachinelearning/webnn#540 (11 comments by @anssiko, @inexorabletash, @zolkis)
-
Process: Add documentation for labels, current and proposed webmachinelearning/webnn#533 (10 comments by @anssiko, @inexorabletash)
-
Wording change: Streamline context creation and validation algorithms webmachinelearning/webnn#519 (9 comments by @huningxin, @inexorabletash, @wchao1115, @zolkis)
-
Stylistic change: Replace some explicit use of link target naming with scopes webmachinelearning/webnn#499 (9 comments by @inexorabletash, @wchao1115)
-
Revise use cases with transformers webmachinelearning/webnn#507 (8 comments by @anssiko, @bhack, @inexorabletash)
-
Adopt Bikeshed best practices and Infra spec conventions webmachinelearning/webnn#495 (8 comments by @anssiko, @huningxin, @inexorabletash, @zolkis)
-
Wording change: Don't use asserts for validating argument types webmachinelearning/webnn#518 (6 comments by @inexorabletash, @wchao1115, @zolkis)
-
Wording change: Prefix literal lists with "the list" webmachinelearning/webnn#515 (6 comments by @anssiko, @huningxin, @inexorabletash, @zolkis)
-
New content: Add definition for shape broadcasting webmachinelearning/webnn#534 (5 comments by @a-sully, @inexorabletash, @wacky6)
-
Wording change: Tidy up more arrays within algorithms to be lists webmachinelearning/webnn#521 (5 comments by @huningxin, @inexorabletash)
-
Update Status of this document for CR Snapshot webmachinelearning/webnn#532 (4 comments by @anssiko, @wchao1115)
-
Document coding conventions and best practices for spec maintainers webmachinelearning/webnn#525 (3 comments by @anssiko, @inexorabletash)
-
Wording change: Dedupe dictionary member type and default definitions webmachinelearning/webnn#526 (2 comments by @inexorabletash, @zolkis)
-
Bug fix: resample2d output size calculation algorithm and usage webmachinelearning/webnn#520 (2 comments by @huningxin, @inexorabletash)
-
Bug fix: Correct indentation of an iteration substep webmachinelearning/webnn#522 (2 comments by @anssiko, @inexorabletash)
-
Stylistic change: Linkify MLOperandDataType enum members webmachinelearning/webnn#508 (2 comments by @inexorabletash, @zolkis)
-
Wording change: Define power prefs & device type using WebIDL enums webmachinelearning/webnn#511 (2 comments by @inexorabletash, @zolkis)
-
Wording change: Add types to all internal algorithm members webmachinelearning/webnn#536 (1 comment by @inexorabletash)
-
Bug fix: Remove null value support in reshape()'s newShape parameter webmachinelearning/webnn#505 (1 comment by @inexorabletash)
-
Wording change: New up a specific type, rather than just object webmachinelearning/webnn#512 (1 comment by @inexorabletash)
-
Wording change: Add missing receiver when assigning to internal slot. webmachinelearning/webnn#513 (1 comment by @zolkis)
-
webmachinelearning/webmachinelearning-ethics
-
1 pull request opened.
-
Expand transparency and explainability mitigations webmachinelearning/webmachinelearning-ethics#32, by @anssiko.
-
-
1 pull request merged.
-
Expand transparency and explainability mitigations webmachinelearning/webmachinelearning-ethics#32.
-
-
1 pull request received 5 new comments.
-
Expand transparency and explainability mitigations webmachinelearning/webmachinelearning-ethics#32 (5 comments by @anssiko, @dontcallmedom)
-