1924499 - Add an API to check if we can run inference

Status: RESOLVED FIXED

(Core :: Machine Learning: General, enhancement, P2)

Description

[Tarek Ziadé (:tarek)]

Two elements:

• a C++ helper to get the info
• a JS function to loop on that info and block if there's not enough memory

Comment 1

[Tarek Ziadé (:tarek)]

GlobalMemoryStatusEx already provides it and is used in nsAvailableMemoryWatcher::IsCommitSpaceLow()

Comment 2

[Tarek Ziadé (:tarek)]

Attachment: Bug 1924499 - Add an API to check if we can run inference r?padenot,gregtatum

Comment 3

[Tarek Ziadé (:tarek)]

It's too hard to find a formula,

I tried based on config.json, on the onnx file sizes, but there are too many factors and unknowns (ion, etc).
we want instead to run in the CI some benchmarks to collect some data depending on the number of threads, the model etc

for now we'll keep a hardcoded list of models with a value

Comment 4

[Tarek Ziadé (:tarek)]

That list can be in RS

Comment 5

[Tarek Ziadé (:tarek)]

Landing this now -- the memory checker is preffed off. I will continue in follow up bugs

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/5f9b82e165a2
Add an API to check if we can run inference r=padenot,gregtatum,frontend-codestyle-reviewers,Standard8

Comment 7

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/5f9b82e165a2