| import torch | |
| import warnings | |
| warnings.filterwarnings("ignore") | |
| from transformers import WhisperForConditionalGeneration, WhisperProcessor | |
| model_name_or_path = "openai/whisper-large-v3" | |
| processor = WhisperProcessor.from_pretrained(model_name_or_path) | |
| model = WhisperForConditionalGeneration.from_pretrained(model_name_or_path, torch_dtype=torch.float16).to("cpu") | |
| inputs = torch.zeros(1, 128, 3000, dtype=torch.float16, device="cpu") | |
| out = model.generate(inputs, return_timestamps=True, num_beams=1) | |
| print("generate output shape:", out.shape) | |