Hi, I'm new to runpod and try to debug this error
Is there any way to fetch more log details than this? I learned that the
/logs
endpoint is only for pods.
Also, it would help tremendously if I could debug the Docker image exactly as is locally. So far I tested it by running rp_handler.py
, which obviously did not trigger all issues.3 Replies
Some more info on this - it is a modified version of https://github.com/runpod-workers/worker-faster_whisper/tree/main.
I now tried to create a vanilla Docker image from that repo (just reduced the number of models) and deployed it as a new serverless endpoint. Same result:
Could it be that I'm using podman for this (but with
--format docker
) from Mac (but with --platform linux/amd64
)?maybe, you could try it
or there is some problems with the whisper repo's / its deps
if you try, using the nomal image that is attached in the readme.md (the one they've buildd) is it working?
No, the normal repo as of now actually does not work. I've seen that faster-whisper has been downgraded from 1.1.0 to 0.10.0 - I guess because of the CuDNN issues. However, I managed to fix it: https://github.com/runpod-workers/worker-faster_whisper/pull/53 -- with this I built
fivetwosix/worker-faster_whisper:0.0.0
which works fine as a serverless worker.GitHub
Upgrading faster_whisperer to 1.1.1 and a bunch of other fixes to t...
Switches to latest faster-whisperer, which in turn requires nvidia/cuda:12.3.2-cudnn9-runtime-ubuntu22.04 as the Docker base.
Also pulled in a fix from runpod-python, which is not yet merged - othe...