This message was deleted.
# ask-for-help
s
This message was deleted.
👀 1
🚀 1
j
The container which was produced using
bentoctl build
is giving me latencies around 4000ms.
The one using
bentoml containerize
is more like 40ms
b
Can you tell me which platform you are deploying to?
j
Google Cloud Run - but I’m running both containers locally and using
curl
on 127.0.0.1
The two container images are called:
model:dbx2jmr6qcggqdsv
and
google-cloud-run-model:dbx2jmr6qcggqdsv
So they’re definitely based on the same Bento
One is just 100x slower than the other
b
Got it.
I saw in the previous thread that you are able to update main.tf and get it working. And now it is the very long latency problem. Is that right?
j
Well, I was trying to disable AdaptiveBatching because I was seeing latencies around the 10s mark when deployed to Google Cloud Run so thought it might be the
max_latency_ms
.
Once I discovered how difficult it is to override this globally I just disabled batching at the Model level (retrained and saved with
batching: False
)
Then decided to test locally
And still have incredibly high latency on the
google-cloud-run-
version of the image
So the high latency problem is what prompted the other problem 😛
Or at least the debugging of the high latency
b
Got it. Do you have time today for office hour?
j
I’m in the UK so was just about to sign off for the day almost 6pm here — so unfortunately not 😞
Could fit something in early your time / late in the day my time tomorrow if that would be possible.
b
yeah that should work. And you can find a time on my calendly too https://calendly.com/bo-bentoml/office-hours
What’s the time requirement you have right now?
is this for production or for development and poc?
j
is this for production or for development and poc?
We’re aiming to deploy a model as an internal tool. So production but not customer facing.
What’s the time requirement you have right now?
Aiming to deploy this week.
yeah that should work. And you can find a time on my calendly too https://calendly.com/bo-bentoml/office-hours
Unfortunately, all of those times are between midnight and 5am my time 😞
b
I see. Let’s find a time works better for you. Do you have time tmw 4pm london time/8am PST?
j
That absolutely works for me. Could even do a bit later if it’s better for you (5pm london time?)
b
yeah, that would be amazing!
Really appreciated it
could you dm me your email and I will sent out invite
I appreciate your patience and support on this project.
s
@Judah Rand would you be comfortable sharing the images with us?