Salvage from a widely-heard AI

2025-03-27

When I put large data in the Azure environment's broadband AI, it ran for a while and then gave me an error.

I want to retrieve intermediate data in Azure Container App to find out what caused the failure.

https://gyazo.com/7f58f3e0d85f1b677e529026b56f4eb8

Note: The

This time we are talking about extracting intermediate data, a task that is not necessary for most people.

While this is being done, the admin and results display screens will of course all stop working.

It's something that wouldn't be possible for a normal 24-hour web service operation, but in this case we're doing it because we're not in a situation where we're providing a 24-hour service for our customers.

code::

az containerapp exec \

--name <container app name> \c

--resource-group <resource group name> }

--container <container name> \cHTML

--command "/bin/bash"

$ apt update

$ apt install zip

$ cd broadlistening/pipeline

$ zip -r outputs.zip outputs

$ python -mhttp.server 8001

This is Azure forwarding the request to an HTTP server running on 8000

Temporarily change this setting to 8001.

Now when you access the API server, you will see the results.zip file you just created and be able to download it.

https://gyazo.com/0b3066753a9f09a68b6acddbaef31a5e

Don't forget to put it back when you're done downloading.

A normal successful report would look like the one on the left, but this one is on the right

25000~ data of 5.2MB of extraction results are output to args.csv.

I think it might be "Out of Memory", but I think it might be so.

I'm using text-embedding-3-small with embedding and it's a float32 with 1536 dimensions per case.

Python's float is 8 bytes in double precision float64, so total = 1536 × 8 = 12,288 bytes (about 12 KB)

25,000 records would be roughly 300 MB.

I guess a container with 1GB of memory means there's not that much room for applications.

2GB should be more than enough.

---