- From: Kostiainen, Anssi <anssi.kostiainen@intel.com>
- Date: Thu, 25 Apr 2024 11:11:51 +0000
- To: "public-webmachinelearning-wg@w3.org" <public-webmachinelearning-wg@w3.org>
- Message-ID: <D0B4F069-C9B2-4093-8B34-0DF805BFD6A4@intel.com>
Latest version: https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md
WebML WG Teleconference – 2 May 2024 - 14:00-15:00 UTC
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#webml-wg-teleconference--2-may-2024---1400-1500-utc>
See the timezone table ...
Logistics
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#logistics>
* Chair: Anssi
* Scribe: ? (see howto<https://github.com/webmachinelearning/meetings/blob/main/scribe-howto.md>, volunteers welcome!)
* IRC: irc://irc.w3.org:6667/#webmachinelearning
* IRC web client: https://irc.w3.org/?channels=#webmachinelearning
* Joining instructions: https://lists.w3.org/Archives/Member/internal-webmachinelearning/2023Jun/0000.html
* Minutes: https://www.w3.org/2024/05/02-webmachinelearning-minutes.html
Agenda
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#agenda>
ℹ️ NPU support
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%E2%84%B9%EF%B8%8F-npu-support>
Follow up on the NPU support discussion.
Please review the 3 possible options for extending the MLDeviceType presented and the error handling and fallback considerations. Please continue provide your feedback via:
* ⨀ WebNN should support NPU and QDQ operations webmachinelearning/webnn#623<https://github.com/webmachinelearning/webnn/issues/623>
Related:
* ⨀ WebNN should support int8 quantized models webmachinelearning/webnn#128<https://github.com/webmachinelearning/webnn/issues/128>
* ⨀ API simplification: context types, context options, createContext() webmachinelearning/webnn#302<https://github.com/webmachinelearning/webnn/issues/302>
* ⨀ Add NPU option for WebNN samples webmachinelearning/webnn-samples#220<https://github.com/webmachinelearning/webnn-samples/issues/220>
* ⛭ Chromium implementation (NPU device type) https://chromium-review.googlesource.com/c/chromium/src/+/5330647
ℹ️ MLBuffer
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%E2%84%B9%EF%B8%8F-mlbuffer>
Discuss the recent implementation experience and findings in supporting MLBuffer on more platforms, including createBuffer() specific design considerations for WebGPU usage flag and MLBufferDescriptor.
* ⨀ webmachinelearning/webnn#542<https://github.com/webmachinelearning/webnn/issues/542>
ℹ️ Open issues and PRs
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%E2%84%B9%EF%B8%8F-open-issues-and-prs>
Discuss open issues and review PRs based on your feedback:
* ⨀ All open issues<https://github.com/webmachinelearning/webnn/issues>
* ⛙ All open pull requests<https://github.com/webmachinelearning/webnn/pulls>
A curated shortlist of recently discussed issues below. If any are addressed prior to the meeting, we'll debrief the group on the changes to keep everyone up to date with significant developments. Please refer to the triage guidance<https://github.com/webmachinelearning/webnn/blob/main/docs/IssueTriage.md> for more on our triage practices.
🏷️ Debrief on PRs merged recently
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%EF%B8%8F-debrief-on-prs-merged-recently>
The editors and PR authors will debrief the group on substantive PRs that got merged in the last few weeks, answer question from the group.
* ⛙ Recently merged PRs<https://github.com/webmachinelearning/webnn/pulls?q=is%3Apr+is%3Amerged>
🏷️ bug<https://github.com/webmachinelearning/webnn/labels/bug>
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%EF%B8%8F-bug>
* ArgMax/Min selectLastIndex is not supported on CoreML
* ⨀ webmachinelearning/webnn#652<https://github.com/webmachinelearning/webnn/issues/652>
* ☰ Proposal to remove selectLastIndex parameter
* Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type
* ⨀ webmachinelearning/webnn#653<https://github.com/webmachinelearning/webnn/issues/653>
* ☰ Review presented options: : 1) output int32 2) allow passing output data type as a parameter
* Consider dropping the support of uint64/int64 data type for some operators
* ⨀ webmachinelearning/webnn#654<https://github.com/webmachinelearning/webnn/issues/654>
* ☰ Review the affected list of ops
* Synchronously validate input operands/activations
* ⨀ webmachinelearning/webnn#572<https://github.com/webmachinelearning/webnn/issues/572>
* ☰ A gentle nudge for contributions for the identified remaining work
🏷️ question<https://github.com/webmachinelearning/webnn/labels/question>
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%EF%B8%8F-question>
* Can an MLGraphBuilder be reused?
* ⨀ webmachinelearning/webnn#567<https://github.com/webmachinelearning/webnn/issues/567>
* ☰ Discuss implementation experience, performance implications of MLBuffer constant, ideation around MLBuffer timeline
* Behavior for numeric overflows/underflows/etc in algorithms
* ⨀ webmachinelearning/webnn#636<https://github.com/webmachinelearning/webnn/issues/636>
* ☰ Proposed alternatives: do nothing, reference Infra spec, add helpers, add steps to explicitly validate
* Specify the operand data type constraints of operation
* ⨀ webmachinelearning/webnn#283<https://github.com/webmachinelearning/webnn/issues/283>
* ⛙ webmachinelearning/webnn#646<https://github.com/webmachinelearning/webnn/pull/646>
* ☰ Review PR
🏷️ feature request<https://github.com/webmachinelearning/webnn/labels/feature%20request>
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%EF%B8%8F-feature-request>
* Constant sequential filling operation needs output shape parameter
* ⨀ webmachinelearning/webnn#492<https://github.com/webmachinelearning/webnn/issues/492>
* ☰ Review new information from a deep dive into deducing tensors shapes
* MLNumber / Type of some parameters should match the input data type
* ⨀ webmachinelearning/webnn#442<https://github.com/webmachinelearning/webnn/issues/442>
* ⛙ webmachinelearning/webnn#647<https://github.com/webmachinelearning/webnn/pull/647>
* ☰ Review (draft) PR
🏷️ operator specific<https://github.com/webmachinelearning/webnn/labels/operator%20specific>
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-05-02-wg-agenda.md#%EF%B8%8F-operator-specific>
* Should softmax be a fused activation
* ⨀ webmachinelearning/webnn#471<https://github.com/webmachinelearning/webnn/issues/471>
* ☰ Provide feedback on the proposal
* argMax/Min only support scalar axis in TFLite runtime
* ⨀ webmachinelearning/webnn#629<https://github.com/webmachinelearning/webnn/issues/629>
* ☰ Review other backends and frameworks
* Clarify the restriction for minValue and maxValue of MLClampOptions
* ⨀ webmachinelearning/webnn#396<https://github.com/webmachinelearning/webnn/issues/396>
* ☰ Proposal to remove "Not all implementations support minValue equal to maxValue." note from the spec.
* Remove builder.constant(value, type) variant
* ⨀ webmachinelearning/webnn#475<https://github.com/webmachinelearning/webnn/issues/475>
* ⛙ webmachinelearning/webnn#650<https://github.com/webmachinelearning/webnn/pull/650>
* ☰ Review PR
* Need clarify the maximum limit of the hidden size of the Gru operator
* ⨀ webmachinelearning/webnn#625<https://github.com/webmachinelearning/webnn/issues/625>
* ⛙ webmachinelearning/webnn#644<https://github.com/webmachinelearning/webnn/pull/644>
* ☰ Review PR
Received on Thursday, 25 April 2024 11:12:27 UTC