diff --git a/docs/tutorial/README_zh.md b/docs/tutorial/README_zh.md index 78acb5825..4d5e31079 100644 --- a/docs/tutorial/README_zh.md +++ b/docs/tutorial/README_zh.md @@ -268,7 +268,7 @@ torchrun --nnodes 1 --nproc_per_node ${gpu_num} \ export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` 在从节点上（假设IP为192.168.1.2），你需要确保MASTER_ADDR和MASTER_PORT环境变量与主节点设置的一致，并运行同样的命令： @@ -276,7 +276,7 @@ torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192. export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` diff --git a/examples/README_zh.md b/examples/README_zh.md index 78acb5825..4d5e31079 100644 --- a/examples/README_zh.md +++ b/examples/README_zh.md @@ -268,7 +268,7 @@ torchrun --nnodes 1 --nproc_per_node ${gpu_num} \ export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` 在从节点上（假设IP为192.168.1.2），你需要确保MASTER_ADDR和MASTER_PORT环境变量与主节点设置的一致，并运行同样的命令： @@ -276,7 +276,7 @@ torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192. export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` diff --git a/examples/industrial_data_pretraining/paraformer-zh-spk/README_zh.md b/examples/industrial_data_pretraining/paraformer-zh-spk/README_zh.md index 78acb5825..4d5e31079 100644 --- a/examples/industrial_data_pretraining/paraformer-zh-spk/README_zh.md +++ b/examples/industrial_data_pretraining/paraformer-zh-spk/README_zh.md @@ -268,7 +268,7 @@ torchrun --nnodes 1 --nproc_per_node ${gpu_num} \ export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` 在从节点上（假设IP为192.168.1.2），你需要确保MASTER_ADDR和MASTER_PORT环境变量与主节点设置的一致，并运行同样的命令： @@ -276,7 +276,7 @@ torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192. export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` diff --git a/examples/industrial_data_pretraining/paraformer/README_zh.md b/examples/industrial_data_pretraining/paraformer/README_zh.md index 78acb5825..4d5e31079 100644 --- a/examples/industrial_data_pretraining/paraformer/README_zh.md +++ b/examples/industrial_data_pretraining/paraformer/README_zh.md @@ -268,7 +268,7 @@ torchrun --nnodes 1 --nproc_per_node ${gpu_num} \ export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` 在从节点上（假设IP为192.168.1.2），你需要确保MASTER_ADDR和MASTER_PORT环境变量与主节点设置的一致，并运行同样的命令： @@ -276,7 +276,7 @@ torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192. export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` diff --git a/examples/industrial_data_pretraining/paraformer_streaming/README_zh.md b/examples/industrial_data_pretraining/paraformer_streaming/README_zh.md index 78acb5825..4d5e31079 100644 --- a/examples/industrial_data_pretraining/paraformer_streaming/README_zh.md +++ b/examples/industrial_data_pretraining/paraformer_streaming/README_zh.md @@ -268,7 +268,7 @@ torchrun --nnodes 1 --nproc_per_node ${gpu_num} \ export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` 在从节点上（假设IP为192.168.1.2），你需要确保MASTER_ADDR和MASTER_PORT环境变量与主节点设置的一致，并运行同样的命令： @@ -276,7 +276,7 @@ torchrun --nnodes 2 --node_rank 0 --nproc_per_node ${gpu_num} --master_addr=192. export CUDA_VISIBLE_DEVICES="0,1" gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') -torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr=192.168.1.1 --master_port=12345 \ +torchrun --nnodes 2 --node_rank 1 --nproc_per_node ${gpu_num} --master_addr 192.168.1.1 --master_port 12345 \ ../../../funasr/bin/train.py ${train_args} ``` diff --git a/funasr/datasets/audio_datasets/samplers.py b/funasr/datasets/audio_datasets/samplers.py index a56a980f7..01f5e6a3e 100644 --- a/funasr/datasets/audio_datasets/samplers.py +++ b/funasr/datasets/audio_datasets/samplers.py @@ -23,11 +23,11 @@ def CustomDistributedBatchSampler_fn(dataset, **kwargs): batch_sampler = CustomDistributedBatchSampler(dataset, **kwargs) else: - # if kwargs.get("sort_size", -1) > 0: - # batch_sampler = CustomDistributedBufferDynamicBatchSampler(dataset, **kwargs) - # else: - # batch_sampler = CustomDistributedDynamicBatchSampler(dataset, **kwargs) - batch_sampler = CustomDistributedDynamicBatchSampler(dataset, **kwargs) + if kwargs.get("sort_size", -1) > 0: + batch_sampler = CustomDistributedBufferDynamicBatchSampler(dataset, **kwargs) + else: + batch_sampler = CustomDistributedDynamicBatchSampler(dataset, **kwargs) + # batch_sampler = CustomDistributedDynamicBatchSampler(dataset, **kwargs) dataloader_args["batch_sampler"] = batch_sampler dataloader_args["num_workers"] = kwargs.get("num_workers", 4) @@ -244,6 +244,8 @@ class CustomDistributedDynamicBatchSampler(DistributedSampler): self.total_size = len(self.dataset) # self.num_samples = int(math.ceil(self.total_size / self.num_replicas)) self.epoch = 0 + self.max_token_length = kwargs.get("max_token_length", 2048) + self.length_scale_source = kwargs.get("length_scale_source", 1.0) def __iter__(self): if self.shuffle: @@ -262,6 +264,8 @@ class CustomDistributedDynamicBatchSampler(DistributedSampler): for idx in indices: sample_length = self.dataset.get_source_len(idx) + if sample_length > self.max_token_length: + continue potential_batch_length = (max_len_in_batch if sample_length < max_len_in_batch else sample_length) * ( len(batch) + 1) @@ -269,12 +273,12 @@ class CustomDistributedDynamicBatchSampler(DistributedSampler): batch.append(idx) if sample_length > max_len_in_batch: max_len_in_batch = sample_length - current_batch_length = max_len_in_batch * len(batch) + # current_batch_length = max_len_in_batch * len(batch) else: batches.append(batch) batch = [idx] max_len_in_batch = sample_length - current_batch_length = max_len_in_batch + # current_batch_length = max_len_in_batch # Add the last batch if it's not empty and we're not dropping it if batch and (not self.drop_last or len(batch) * max_len_in_batch == self.batch_size): @@ -293,6 +297,7 @@ class CustomDistributedDynamicBatchSampler(DistributedSampler): class CustomDistributedBufferDynamicBatchSampler(DistributedSampler): def __init__(self, dataset, batch_size, + batch_type="token", num_replicas=None, rank=None, shuffle=True, @@ -312,6 +317,7 @@ class CustomDistributedBufferDynamicBatchSampler(DistributedSampler): self.num_replicas = num_replicas self.dataset = dataset self.batch_size = batch_size + self.batch_type = batch_type self.is_training = is_training self.shuffle = shuffle and is_training self.drop_last = drop_last @@ -319,42 +325,54 @@ class CustomDistributedBufferDynamicBatchSampler(DistributedSampler): self.total_size = len(self.dataset) # self.num_samples = int(math.ceil(self.total_size / self.num_replicas)) self.epoch = 0 - self.sort_size = sort_size - + self.sort_size = sort_size * num_replicas + self.max_token_length = kwargs.get("max_token_length", 2048) + self.length_scale_source = kwargs.get("length_scale_source", 1.0) + def __iter__(self): if self.shuffle: g = torch.Generator() g.manual_seed(self.epoch) - indices = torch.randperm(self.total_size, generator=g).tolist() + indices = torch.randperm(len(self.dataset), generator=g).tolist() else: - indices = list(range(self.total_size)) - - # Distribute indices among replicas - indices = indices[self.rank:self.total_size:self.num_replicas] + indices = list(range(len(self.dataset))) - # Sort indices into buffers - sorted_buffers = [sorted(indices[i:i + self.sort_size], key=lambda idx: self.dataset.get_source_len(idx)) for i in range(0, len(indices), self.sort_size)] - - batches = [] - for buffer in sorted_buffers: + # Create sorted buffers and form batches + buffer_batches = [] + for i in range(0, len(indices), self.sort_size): + buffer = sorted(indices[i:i + self.sort_size], key=lambda idx: self.dataset.get_source_len(idx)) batch = [] max_len_in_batch = 0 for idx in buffer: - sample_length = self.dataset.get_source_len(idx) + original_sample_length = self.dataset.get_source_len(idx) + if original_sample_length > self.max_sample_length: + continue + sample_length = 1 if self.batch_type == "example" else original_sample_length potential_batch_length = max(max_len_in_batch, sample_length) * (len(batch) + 1) if potential_batch_length <= self.batch_size: batch.append(idx) max_len_in_batch = max(max_len_in_batch, sample_length) else: - batches.append(batch) + buffer_batches.append(batch) batch = [idx] max_len_in_batch = sample_length - - # Add the last batch if it's not empty and we're not dropping it - if batch and (not self.drop_last or len(batch) * max_len_in_batch == self.batch_size): - batches.append(batch) + if batch: + buffer_batches.append(batch) - return iter(batches) + # Ensure each rank gets the same number of batches, duplicate data if needed + batches_per_rank = math.ceil(len(buffer_batches) / self.num_replicas) + total_batches_needed = batches_per_rank * self.num_replicas + buffer_batches.extend(buffer_batches[:total_batches_needed - len(buffer_batches)]) + + # Evenly distribute batches from buffer_batches to each rank + rank_batches = [[] for _ in range(self.num_replicas)] + for i, batch in enumerate(buffer_batches): + rank_batches[i % self.num_replicas].append(batch) + + # Assign all batches for the current rank directly + final_batches = rank_batches[self.rank] + + return iter(final_batches) def __len__(self): diff --git a/funasr/train_utils/trainer.py b/funasr/train_utils/trainer.py index 55dce9996..35a266fbf 100644 --- a/funasr/train_utils/trainer.py +++ b/funasr/train_utils/trainer.py @@ -161,17 +161,17 @@ class Trainer: self.best_step_or_epoch = ckpt_name best_ckpt = Path(os.path.join(self.output_dir, f'model.pt.best')) torch.save(state, best_ckpt) - logging.info(f"Update best acc: {self.val_acc_step_or_eoch[self.best_step_or_epoch]}, {best_ckpt}") + logging.info(f"Update best acc: {self.val_acc_step_or_eoch[self.best_step_or_epoch]:.4f}, {best_ckpt}") else: - logging.info(f"No improvement in acc: {self.val_acc_step_or_eoch[ckpt_name]} < {self.val_acc_step_or_eoch[self.best_step_or_epoch]}") + logging.info(f"No improvement in acc: {self.val_acc_step_or_eoch[ckpt_name]:.4f} < {self.val_acc_step_or_eoch[self.best_step_or_epoch]:.4f}") elif self.avg_keep_nbest_models_type == "loss": if self.val_loss_step_or_eoch[ckpt_name] <= self.val_loss_step_or_eoch[self.best_step_or_epoch]: self.best_step_or_epoch = ckpt_name best_ckpt = Path(os.path.join(self.output_dir, f'model.pt.best')) torch.save(state, best_ckpt) - logging.info(f"Update best loss: {self.val_loss_step_or_eoch[self.best_step_or_epoch]}, {best_ckpt}") + logging.info(f"Update best loss: {self.val_loss_step_or_eoch[self.best_step_or_epoch]:.4f}, {best_ckpt}") else: - logging.info(f"No improvement in loss: {self.val_loss_step_or_eoch[ckpt_name]} > {self.val_loss_step_or_eoch[self.best_step_or_epoch]}") + logging.info(f"No improvement in loss: {self.val_loss_step_or_eoch[ckpt_name]:.4f} > {self.val_loss_step_or_eoch[self.best_step_or_epoch]:.4f}") else: print("Undo") self.saved_ckpts[ckpt_name] = getattr(self, f"val_{self.avg_keep_nbest_models_type}_step_or_eoch")[ckpt_name]