Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
decoding
游雁dce53b268b55
diff --git a/funasr/bin/train_ds.py b/funasr/bin/train_ds.pyindex da99adccc..c55f9504e 100644--- a/funasr/bin/train_ds.py+++ b/funasr/bin/train_ds.py@@ -66,6 +66,7 @@ def main(**kwargs):# open tf32torch.backends.cuda.matmul.allow_tf32 = kwargs.get("enable_tf32", True)+ rank = int(os.environ.get("RANK", 0))local_rank = int(os.environ.get("LOCAL_RANK", 0))world_size = int(os.environ.get("WORLD_SIZE", 1))@@ -83,7 +84,7 @@ def main(**kwargs):dist.init_process_group(backend=kwargs.get("backend", "nccl"), init_method="env://")torch.cuda.set_device(local_rank)- rank = dist.get_rank()+ # rank = dist.get_rank()logging.info("Build model, frontend, tokenizer")device = kwargs.get("device", "cuda")diff --git a/funasr/train_utils/trainer_ds.py b/funasr/train_utils/trainer_ds.pyindex 0fbac966f..e6b6284bc 100644--- a/funasr/train_utils/trainer_ds.py+++ b/funasr/train_utils/trainer_ds.py@@ -78,7 +78,7 @@ class Trainer:output_dir (str): The directory where model checkpoints will be saved. Default is './'.resume (str, optional): The file path to a checkpoint to resume training from."""- self.rank = kwargs.get("rank", 0)+ self.rank = rankself.local_rank = local_rankself.world_size = world_sizeself.use_ddp = use_ddp