Web Machine Learning Working Group - Activity
No matches.
GitHub
webmachinelearning/webnn
-
2 issues opened.
-
Explore W3C Open Source Program as a way to support GPU-backed WPT runners webmachinelearning/webnn#972, by @anssiko.
-
Investigate how to integrate WebNN caching with Cross-Origin Storage webmachinelearning/webnn#971, by @anssiko.
-
-
14 issues received 15 new comments.
-
Proposal: Support build-time metadata to describe how a graph will be driven webmachinelearning/webnn#965 (2 comments by @anssiko, @wangw-1991)
-
Explore W3C Open Source Program as a way to support GPU-backed WPT runners webmachinelearning/webnn#972 (1 comment by @ianbjacobs)
-
Consider specify lower limit for conv2d/pool2d kernel sizes, dilations, strides webmachinelearning/webnn#928 (1 comment by @anssiko)
-
Expand the expand operator to support blockwise broadcasting webmachinelearning/webnn#903 (1 comment by @anssiko)
-
Make zeroPoint optional for quantizeLinear and dequantizeLinear operations webmachinelearning/webnn#943 (1 comment by @anssiko)
-
-
1 pull request received 23 new comments.
-
Add explainer: Support dynamic language models on NPUs via WebNN webmachinelearning/webnn#969 (23 comments by @matthargett, @mklimenko-nv, @wangw-1991)
-
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 joined.
-
Dave Raggett (W3C) joined.
-
-
5 participants left.
-
Lingyan Zhao (Microsoft Corporation) left.
-
Chai Chaoweeraprasit (Microsoft Corporation) left.
-
Josi Rosenfeld (CleverThis, Inc.) left.
-
Adnaan Nazir (CleverThis, Inc.) left.
-
Aditya Chhabra (CleverThis, Inc.) left.
-
GitHub
webmachinelearning/webnn
-
9 issues opened.
-
Consider dropping negative indices support in gather*/scatter* operators webmachinelearning/webnn#970, by @shiyi9801.
-
Add `topK` operator (a sliced `sort`) webmachinelearning/webnn#968, by @fdwr.
-
Do subnormal results have to be preserved when the reference formula overflows in float32? webmachinelearning/webnn#967, by @matthargett.
-
Proposal: Support stateful model for WebNN webmachinelearning/webnn#966, by @wangw-1991.
-
Proposal: Support build-time metadata to describe how a graph will be driven webmachinelearning/webnn#965, by @wangw-1991.
-
What is the current testing/shipping status of WebNN in Chrome and Edge, and what should developers use today? webmachinelearning/webnn#964, by @webmaxru.
-
Expose engine-defined compute timing for WebNN (`MLComputeTiming`) webmachinelearning/webnn#963, by @shubhamg13.
-
Direct Video Plane Access for WebNN webmachinelearning/webnn#961, by @bbernhar.
-
Clarify finite/infinite classification at numerical tolerance boundaries webmachinelearning/webnn#960, by @matthargett.
-
-
3 issues closed.
-
conv2d: infer groups from input and filter shapes when not specified webmachinelearning/webnn#959.
-
Validate LSTM/GRU full-sequence output regardless of the `returnSequence` flag webmachinelearning/webnn#949.
-
Proposal: STC-1.0 Tensor Container for Zero-Copy WebGPU/Wasm Loading and Cryptographic Provenance webmachinelearning/webnn#952.
-
-
19 issues received 36 new comments.
-
Proposal: Support build-time metadata to describe how a graph will be driven webmachinelearning/webnn#965 (6 comments by @fdwr, @matthargett, @reillyeon, @wangw-1991)
-
Proposal: Support stateful model for WebNN webmachinelearning/webnn#966 (5 comments by @fdwr, @matthargett, @sushraja-msft, @wangw-1991)
-
Expose engine-defined compute timing for WebNN (`MLComputeTiming`) webmachinelearning/webnn#963 (4 comments by @shubhamg13, @vmukhachev)
-
(de)quantization behaviors on CoreML webmachinelearning/webnn#822 (2 comments by @fdwr, @matthargett)
-
Add `topK` operator (a sliced `sort`) webmachinelearning/webnn#968 (2 comments by @fdwr, @matthargett)
-
Effective MLComputePolicy exposure webmachinelearning/webnn#934 (2 comments by @anssiko, @shubhamg13)
-
conv2d: infer groups from input and filter shapes when not specified webmachinelearning/webnn#959 (2 comments by @fdwr, @ibelem)
-
Proposal: STC-1.0 Tensor Container for Zero-Copy WebGPU/Wasm Loading and Cryptographic Provenance webmachinelearning/webnn#952 (2 comments by @anssiko, @sovereign-ecosystem)
-
Consider dropping negative indices support in gather*/scatter* operators webmachinelearning/webnn#970 (1 comment by @fdwr)
-
Expand the expand operator to support blockwise broadcasting webmachinelearning/webnn#903 (1 comment by @matthargett)
-
Clarify finite/infinite classification at numerical tolerance boundaries webmachinelearning/webnn#960 (1 comment by @fdwr)
-
Proposal: API to Separate Graph Building from Weight Loading to Reduce Peak Memory Usage webmachinelearning/webnn#901 (1 comment by @RafaelCintron)
-
Device selection criteria for usecase-driven scenarios webmachinelearning/webnn#902 (1 comment by @RattanPriya)
-
How to support scatterElements on TFLite backend when indices operand is not constant webmachinelearning/webnn#878 (1 comment by @vmukhachev)
-
What is the current testing/shipping status of WebNN in Chrome and Edge, and what should developers use today? webmachinelearning/webnn#964 (1 comment by @anssiko)
-
reduceLogSumExp reference decomposition contradicts conformance test; accuracy requirements missing webmachinelearning/webnn#947 (1 comment by @fdwr)
-
-
1 pull request opened.
-
Add explainer: Support dynamic language models on NPUs via WebNN webmachinelearning/webnn#969, by @wangw-1991.
-
-
5 pull requests merged.
-
Add `bfloat16` and `float8e4m3` to the spec webmachinelearning/webnn#956.
-
Add dynamic shape explainer webmachinelearning/webnn#945.
-
Add low-precision floating point data types explainer webmachinelearning/webnn#938.
-
Add stable decomposition clarification webmachinelearning/webnn#957.
-
Enforce validation when creating an MLOperandDescriptor webmachinelearning/webnn#955.
-
-
7 pull requests received 28 new comments.
-
Add dynamic shape explainer webmachinelearning/webnn#945 (12 comments by @anssiko, @huningxin, @miaobin, @reillyeon, @sushraja-msft)
-
Add `bfloat16` and `float8e4m3` to the spec webmachinelearning/webnn#956 (11 comments by @fdwr, @mklimenko-nv)
-
Add explainer: Support dynamic language models on NPUs via WebNN webmachinelearning/webnn#969 (1 comment by @wangw-1991)
-
Refactor device selection: Rename to computePolicy, remove accelerated, and add fallback webmachinelearning/webnn#923 (1 comment by @mingmingtasd)
-
Add low-precision floating point data types explainer webmachinelearning/webnn#938 (1 comment by @huningxin)
-
Enforce validation when creating an MLOperandDescriptor webmachinelearning/webnn#955 (1 comment by @shiyi9801)
-
TR publications
-
2 documents published.
-
Web Neural Network API published twice as a Candidate Recommendation Draft. History of Web Neural Network API
-
Past meetings
Participants
-
1 participant joined.
-
Rattan Priya Bhasin (Microsoft Corporation) joined.
-
-
3 participants left.
-
Kaushik Satpathy (Yahoo Holdings Inc.) left.
-
ChangSeok Oh (ByteDance) left.
-
Stephan Morris (See Me Please) left.
-
-
1 organization left.
-
Yahoo Holdings Inc. left.
-
GitHub
webmachinelearning/webnn
-
2 issues opened.
-
conv2d: infer groups from input and filter shapes when not specified webmachinelearning/webnn#959, by @ibelem.
-
MLTensor export to GPUBuffer with custom usages? webmachinelearning/webnn#954, by @mwyrzykowski.
-
-
3 issues closed.
-
A conformance suite with disallowed intra-op examples would be helpful for hardening. webmachinelearning/webnn#244.
-
Mandate `output_channels % groups == 0` validation for conv2d webmachinelearning/webnn#925.
-
`gather` should allow 0D output webmachinelearning/webnn#946.
-
-
13 issues received 31 new comments.
-
Composite operators / subgraphs webmachinelearning/webnn#907 (9 comments by @anssiko, @huningxin, @mwyrzykowski, @OlehZhyhinas, @snnn, @Yuhengwe1)
-
Accumulation precision is unspecified for matmul/gemm and reduction operations webmachinelearning/webnn#948 (3 comments by @anssiko, @fdwr, @matthargett)
-
MLTensor export to GPUBuffer with custom usages? webmachinelearning/webnn#954 (3 comments by @bbernhar, @fdwr, @mwyrzykowski)
-
Validate LSTM/GRU full-sequence output regardless of the `returnSequence` flag webmachinelearning/webnn#949 (3 comments by @anssiko, @huningxin, @shiyi9801)
-
Consider specify lower limit for conv2d/pool2d kernel sizes, dilations, strides webmachinelearning/webnn#928 (2 comments by @anssiko)
-
Discrepancy in actual and expected output for subgraph test : batchNormalization options.axis=0 + gelu webmachinelearning/webnn#950 (2 comments by @fdwr, @matthargett)
-
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 @anssiko, @matthargett)
-
reduceLogSumExp reference decomposition contradicts conformance test; accuracy requirements missing webmachinelearning/webnn#947 (2 comments by @anssiko, @matthargett)
-
A conformance suite with disallowed intra-op examples would be helpful for hardening. webmachinelearning/webnn#244 (1 comment by @anssiko)
-
Mandate `output_channels % groups == 0` validation for conv2d webmachinelearning/webnn#925 (1 comment by @anssiko)
-
-
4 pull requests opened.
-
Add stable decomposition clarification webmachinelearning/webnn#957, by @mklimenko-nv.
-
Add `bfloat16` and `float8e4m3` to the spec webmachinelearning/webnn#956, by @mklimenko-nv.
-
Enforce validation when creating an MLOperandDescriptor webmachinelearning/webnn#955, by @shiyi9801.
-
Bugfix: Allow 0-D output for gather() webmachinelearning/webnn#953, by @huningxin.
-
-
2 pull requests merged.
-
Add initial WebNN-WebGPU Interop specification webmachinelearning/webnn#942.
-
Bugfix: Allow 0-D output for gather() webmachinelearning/webnn#953.
-
-
5 pull requests received 62 new comments.
-
Add dynamic shape explainer webmachinelearning/webnn#945 (45 comments by @anssiko, @fdwr, @matthargett, @miaobin, @reillyeon, @theHamsta)
-
Add initial WebNN-WebGPU Interop specification webmachinelearning/webnn#942 (10 comments by @anssiko, @bbernhar, @fdwr)
-
Enforce validation when creating an MLOperandDescriptor webmachinelearning/webnn#955 (4 comments by @anssiko, @huningxin, @shiyi9801)
-
Add low-precision floating point data types explainer webmachinelearning/webnn#938 (2 comments by @anssiko, @mklimenko-nv)
-