Conversation
|
Another "TODO" - the new ops need "constraints" tables |
|
Note from discussion w/ @a-sully - CoreML has restrictions on the dequantize op that we'll need to think about.
Re-emphasizing that dequantizing (u)int4 in CoreML is extremely limited (input must be const). @mwyrzykowski - any thoughts about how we can handle the proposed ops efficiently? |
Added data type tables.
Thanks - will address more |
index.bs
Outdated
| </details> | ||
|
|
||
| ### scatterElements ### {#api-mlgraphbuilder-scatterelements} | ||
| Scatter values from the updates tensor along an axis according to the indices in place of the input tensor. |
There was a problem hiding this comment.
🤔 What about "atop" in-place of inplace? Or in-place of a copy of?
inexorabletash
left a comment
There was a problem hiding this comment.
Another batch of feedback; I didn't make it all the way through the PR though.
…ements examples, fix flatten on edge conditions
There was a problem hiding this comment.
🌊 Huge thank you to @fdwr for this significant op set update and @inexorabletash @huningxin, other participants for contributions and comments, in total close to 150 over 200!
As discussed, we'll periodically seek TAG review for significant changes to the spec. This spec update complemented with a demonstration of implementation experience across multiple backends and OSes will be brought to the TAG's attention alongside other significant changes.
inexorabletash
left a comment
There was a problem hiding this comment.
Sorry, noted a few more places where an arg reference should be linkified rather than just styled.
👏 Co-authored-by: Joshua Bell <[email protected]> Co-authored-by: Ningxin Hu <[email protected]>
fdwr
left a comment
There was a problem hiding this comment.
That should address all of Ningxin and Joshua's last feedback. 🤞
|
@inexorabletash Want to take one more pass? |
|
I've been following along, still LGTM. Merge away! |
SHA: 6e19654 Reason: push, by fdwr Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
This CL implements the WebNN spec change [1] that requires the scale and zeroPoint for quantizeLinear and dequantizeLinear ops have the the same rank as input. [1]: webmachinelearning/webnn#805 (comment) Bug: 396176047 Change-Id: Ia310e0c254ad90967b4e3577e54498fa28532f84 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6773953 Commit-Queue: ningxin hu <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Cr-Commit-Position: refs/heads/main@{#1491196}
This CL implements the WebNN spec change [1] that requires the scale and zeroPoint for quantizeLinear and dequantizeLinear ops have the the same rank as input. [1]: webmachinelearning/webnn#805 (comment) Bug: 396176047 Change-Id: Ia310e0c254ad90967b4e3577e54498fa28532f84 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6773953 Commit-Queue: ningxin hu <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Cr-Commit-Position: refs/heads/main@{#1491196}
This CL implements the WebNN spec change [1] that requires the scale and zeroPoint for quantizeLinear and dequantizeLinear ops have the the same rank as input. [1]: webmachinelearning/webnn#805 (comment) Bug: 396176047 Change-Id: Ia310e0c254ad90967b4e3577e54498fa28532f84 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6773953 Commit-Queue: ningxin hu <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Cr-Commit-Position: refs/heads/main@{#1491196}
…s equal to input rank, a=testonly Automatic update from web-platform-tests WebNN: Ensure scale and zeroPoint rank is equal to input rank This CL implements the WebNN spec change [1] that requires the scale and zeroPoint for quantizeLinear and dequantizeLinear ops have the the same rank as input. [1]: webmachinelearning/webnn#805 (comment) Bug: 396176047 Change-Id: Ia310e0c254ad90967b4e3577e54498fa28532f84 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6773953 Commit-Queue: ningxin hu <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Cr-Commit-Position: refs/heads/main@{#1491196} -- wpt-commits: 03b1ef622ac6f93394a64d4213d80a0e8881254f wpt-pr: 53947
This CL implements the WebNN spec change [1] that requires the scale and zeroPoint for quantizeLinear and dequantizeLinear ops have the the same rank as input. [1]: webmachinelearning/webnn#805 (comment) Bug: 396176047 Change-Id: Ia310e0c254ad90967b4e3577e54498fa28532f84 Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6773953 Commit-Queue: ningxin hu <[email protected]> Reviewed-by: Phillis Tang <[email protected]> Cr-Commit-Position: refs/heads/main@{#1491196}
Adds the following operators, per #375 (comment) Support for transformers:
sliceoperator #772reverseoperator #773Todos
RemainingEither done now, or deferred to separate CR/issue:expandto support blockwiseAPI
Preview | Diff