| | |
|---|
| | | data_path_and_name_and_type, |
|---|
| | | preprocess=preprocess_fn, |
|---|
| | | dest_sample_rate=dest_sample_rate, |
|---|
| | | speed_perturb=args.speed_perturb if mode=="train" else None, |
|---|
| | | ) |
|---|
| | | |
|---|
| | | # sampler |
|---|
| | | dataset_conf = args.dataset_conf |
|---|
| | | batch_sampler = LengthBatchSampler( |
|---|
| | | batch_bins=dataset_conf["batch_size"], |
|---|
| | | batch_bins=dataset_conf["batch_conf"]["batch_size"] * args.ngpu, |
|---|
| | | shape_files=shape_files, |
|---|
| | | sort_in_batch=dataset_conf["sort_in_batch"] if hasattr(dataset_conf, "sort_in_batch") else "descending", |
|---|
| | | sort_batch=dataset_conf["sort_batch"] if hasattr(dataset_conf, "sort_batch") else "ascending", |
|---|
| | |
|---|
| | | args.max_update = len(bs_list) * args.max_epoch |
|---|
| | | logging.info("Max update: {}".format(args.max_update)) |
|---|
| | | |
|---|
| | | if args.distributed: |
|---|
| | | if args.distributed and mode=="train": |
|---|
| | | world_size = torch.distributed.get_world_size() |
|---|
| | | rank = torch.distributed.get_rank() |
|---|
| | | for batch in batches: |
|---|
| | |
|---|
| | | self.num_iters_per_epoch = None |
|---|
| | | self.shuffle = mode == "train" |
|---|
| | | self.seed = args.seed |
|---|
| | | self.num_workers = args.num_workers |
|---|
| | | self.num_workers = args.dataset_conf.get("num_workers", 8) |
|---|
| | | self.collate_fn = collate_fn |
|---|
| | | self.pin_memory = args.ngpu > 0 |
|---|
| | | |
|---|
