0.1.34 OCR hangs with 100% CPU usage

[nodesocket]

Few things I noticed @DonTizi upgrading to the latest version today 0.1.34.

On startup, appear to be a foreign language notification?

⚠️ Avertissement: la bibliothèque FlagEmbedding n'est pas installée. Exécutez 'rlama install-dependencies' pour l'installer
Pour installer les dépendances, exécutez: rlama install-dependencies
⚠️ Avertissement: la bibliothèque FlagEmbedding n'est pas installée. Exécutez 'rlama install-dependencies' pour l'installer
Pour installer les dépendances, exécutez: rlama install-dependencies

After rlama setup and just a few OCR conversations all my CPU cores go to 100% and seems to hang. No further processing happens, just CPU at 100%.

[nodesocket]

@DonTizi here is a screenshot, not sure if it's helpful. rlama seems to be stuck doing OCR using 100% of my cpus. Look at the system load. 😆

[nodesocket]

After waiting around 15 minutes, it finally moved on to the next PDF file. I'm thinking tesseract using n(cpu cores) + 1 might have been a bit aggressive and actually massively slowing down the process due to resource starvation. Perhaps n(cpu cores) - 1 is a better default for concurrency for tesseract and OCR operations?

[nodesocket]

Confirmed, resource starvation indeed seems to be the problem. A document with 2,3,4 pages processes almost instantly, but as soon as documents that uses all 9 threads starts grinds to a halt. Just curious, why is each worker process of tesseract spawning 3 threads each itself?