search

ahmetoner / whisper-asr-webservice

OpenAI Whisper ASR Webservice API
https://ahmetoner.github.io/whisper-asr-webservice
MIT License
1.85k stars 331 forks source link

Latest docker container OOM error on tiny model #195

Closed crocodisle closed 4 months ago

crocodisle commented 4 months ago

Hello there!

I've been running the docker container for probably over a year now and recently upgraded. After fixing the swagger version mismatch error, I've run into a few more.

I'm trying to transcribe 1 to 2GB .mp4 video files on a local NAS using the cpu version of whisper and the tiny model. I haven't had issues previously, but now I'm getting new errors.

[2024-02-15 02:53:01 +0000] [1] [ERROR] Worker (pid:7) was sent SIGKILL! Perhaps out of memory?
[2024-02-15 02:53:01 +0000] [26] [INFO] Booting worker with pid: 26
[2024-02-15 02:53:03 +0000] [26] [INFO] Started server process [26]
[2024-02-15 02:53:03 +0000] [26] [INFO] Waiting for application startup.
[2024-02-15 02:53:03 +0000] [26] [INFO] Application startup complete.
[2024-02-15 02:53:24 +0000] [1] [ERROR] Worker (pid:26) was sent SIGKILL! Perhaps out of memory?
[2024-02-15 02:53:24 +0000] [43] [INFO] Booting worker with pid: 43
[2024-02-15 02:53:26 +0000] [43] [INFO] Started server process [43]
[2024-02-15 02:53:26 +0000] [43] [INFO] Waiting for application startup.
[2024-02-15 02:53:26 +0000] [43] [INFO] Application startup complete.

The swagger UI returns TypeError: NetworkError when attempting to fetch resource. The NAS has 8GB RAM. Once again, I haven't had issues previously but now it doesn't seem like it wants to work. I tried both the openai and faster whisper models.

crocodisle commented 4 months ago

Reencoding to a smaller audio file did the trick. The video files had multiple audio tracks so that could have been why it used so much RAM. I'll mark this as closed.