Open xelibrion opened 1 year ago
The issue is that burn-wgpu doesn't currently use the maximum available device memory limits so larger models may fail to run. I'm hoping to resolve this within the next day or two. The slow model loading speed should be resolved by the latest updates.
Not sure if this is related to loading the model, or the transcription process. Also it seems restoring the checkpoint into VRAM takes much longer compared to Python version.
RUST_BACKTRACE=1 cargo run --release audio.wav large-v2