J4R3LL

RuntimeError with DirectML backend (torch.cat concat in tokenizer)

3

I’m running VibeVoice on an AMD GPU with torch-directml as the backend, and I hit a runtime error during inference: ```bash \vibevoice\modular\modular_vibevoice_tokenizer.py", line 495, in _forward_streaming full_input = torch.cat([cached_input, x],...