Whisper model download recovery and manual cache

Oyma downloads Whisper speech model weights on demand when you first select a model in settings. If a network interruption or VPN disconnect corrupts a download, follow these steps to recover.

Model file locations and checksums

Downloaded speech models reside in your user Application Support directory:

~/Library/Application Support/Oyma/models/
├── whisper-base/
│   └── model.bin (~148 MB)
├── whisper-small/
│   └── model.bin (~488 MB)
└── whisper-large-v3-turbo/
    └── model.bin (~1.6 GB)

Step 1: Clearing corrupted partial downloads

If a download hangs repeatedly:

1. Quit Oyma with ⌘Q.
2. Open Terminal and remove incomplete model artifacts:

   rm -rf ~/Library/Application\ Support/Oyma/models/*.tmp

3. Relaunch Oyma and retry the download in Settings › Transcription.

Step 2: Manual model placement (Offline recovery)

If your corporate firewall or VPN blocks direct model downloads:

1. Download the required model weights on an unrestricted network.
2. Create the target models directory on your Mac:

   mkdir -p ~/Library/Application\ Support/Oyma/models/whisper-small

3. Copy model.bin into that directory.
4. Launch Oyma. The application will recognize the model as installed and display a green checkmark.

Related guides

• Model sizing in Transcription models.
• Benchmarks in Whisper benchmarks on Apple silicon.
• Air-gapped setup in Air-gapped Mac deployment protocol.