problem loading a model

[vshirokov89]

here's a problem loading a model. The download freezes, and then the app says it's downloaded. However, when switching to the model, an error occurs:
The model failed to load within 60 seconds. This usually means the model files are corrupted or incompatible. To fix this: 1. Go to Settings → Models. 2. Delete the problematic model. 3. Redownload the model. If the problem persists, try a smaller model (Tiny or Base).
Не удалось переключить модель: Model load failed: Model loading timed out after 120s. This can happen on first launch after an update. Try restarting the app, or delete and re-download the model from Settings.
I tried installing Pindrop from scratch and rolling back to earlier versions. It didn't help.
Please tell me how to submit an error?

[yhay81]

You have already done enough reinstall/re-download testing; the next useful step is a bug report with one clean reproduction and the Pindrop log.

1. In Settings → Advanced, set Log level to Debug (if that option is present in your build).
2. Quit and reopen Pindrop, try to select/load the problematic model once, wait for the timeout, then quit Pindrop.
3. In Finder use Go → Go to Folder… and open:

~/Library/Application Support/Pindrop/Logs

Or from Terminal:

open "$HOME/Library/Application Support/Pindrop/Logs"

Attach the newest pindrop-*.log to a new GitHub Issue. The project README explicitly routes bugs to Issues, while Discussions are for questions/ideas (Community section).

Also include:

• Pindrop version and build number;
• macOS version, Mac model/chip, RAM, and free disk space;
• exact model/provider and size (WhisperKit Tiny/Base, Parakeet, etc.);
• whether Tiny also fails, and whether the failure happens immediately after “downloaded” or only during first load;
• a screenshot of Settings → Models after the failed download.

The log is especially valuable here because current Pindrop records model-load stages under Boot and Transcription, including engine creation, the underlying engine.loadModel start/finish, and the 120-second watchdog (TranscriptionService.swift). The on-disk location and rotation are defined in Logger.swift.

Before uploading, skim the file for anything you do not want public. Pindrop’s normal log path applies redaction to paths, URLs, tokens, and transcription text, but reviewing an attachment is still good practice.

The two different messages (60 seconds and 120 seconds) suggest the UI/download state and the actual engine load are reaching separate timeout paths; the log will show whether the model download is incomplete/corrupt or whether a fully downloaded model is hanging during Core ML/engine initialization.