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caf70826a693
diff --git a/funasr/bin/train_ds.py b/funasr/bin/train_ds.pyindex c55f9504e..6c0f174c2 100644--- a/funasr/bin/train_ds.py+++ b/funasr/bin/train_ds.py@@ -146,7 +146,7 @@ def main(**kwargs):dataloader = dataloader_class(**kwargs)# dataloader_tr, dataloader_val = dataloader_class(**kwargs)- scaler = GradScaler(enabled=trainer.use_fp16) if trainer.use_fp16 else None+ scaler = GradScaler(enabled=True) if trainer.use_fp16 or trainer.use_bf16 else Nonescaler = ShardedGradScaler(enabled=trainer.use_fp16) if trainer.use_fsdp else scalertrainer.resume_checkpoint(