(inputs, tokenizer)
| 44 | |
| 45 | |
| 46 | def collate_fn(inputs, tokenizer): |
| 47 | |
| 48 | input_texts = [_['input_text'] for _ in inputs] |
| 49 | source = [_['source'] for _ in inputs] |
| 50 | gt = [_['gt'] for _ in inputs] |
| 51 | audio_path = [_['audio_path'] for _ in inputs] |
| 52 | audio_info = [tokenizer.process_audio(_['input_text']) for _ in inputs ] |
| 53 | input_tokens = tokenizer(input_texts, |
| 54 | return_tensors='pt', |
| 55 | padding='longest', |
| 56 | audio_info= audio_info) |
| 57 | |
| 58 | return input_tokens.input_ids, input_tokens.attention_mask, source, gt,audio_path,audio_info |
| 59 | |
| 60 | |
| 61 | class InferenceSampler(torch.utils.data.sampler.Sampler): |
nothing calls this directly
no test coverage detected