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
fix bug
游雁b75d1e89bb2f
diff --git a/funasr/bin/train.py b/funasr/bin/train.pyindex c3556d1b9..1feb416bd 100644--- a/funasr/bin/train.py+++ b/funasr/bin/train.py@@ -202,6 +202,7 @@ def main(**kwargs):time1 = time.perf_counter()for data_split_i in range(trainer.start_data_split_i, dataloader.data_split_num):+ time_slice_i = time.perf_counter()dataloader_tr, dataloader_val = dataloader.build_iter(epoch, data_split_i=data_split_i, start_step=trainer.start_step)@@ -223,6 +224,14 @@ def main(**kwargs):torch.cuda.empty_cache()+ time_escaped = (time.perf_counter() - time_slice_i) / 3600.0+ logging.info(+ f"rank: {local_rank}, "+ f"time_escaped_epoch: {time_escaped:.3f} hours, "+ f"estimated to finish {dataloader.data_split_num} data_slices, remaining: {(dataloader.data_split_num-data_split_i)*time_escaped:.3f} hours"+ f"epoch: {((trainer.max_epoch - epoch - 1)*dataloader.data_split_num + dataloader.data_split_num-data_split_i)*time_escaped:.3f} hours\n"+ )+trainer.start_data_split_i = 0trainer.validate_epoch(model=model, dataloader_val=dataloader_val, epoch=epoch + 1, writer=writerdiff --git a/funasr/bin/train_ds.py b/funasr/bin/train_ds.pyindex 21d747a59..d9b767915 100644--- a/funasr/bin/train_ds.py+++ b/funasr/bin/train_ds.py@@ -180,7 +180,7 @@ def main(**kwargs):torch.cuda.empty_cache()- time_escaped = (time.perf_counter() - time1) / 3600.0+ time_escaped = (time.perf_counter() - time_slice_i) / 3600.0logging.info(f"rank: {local_rank}, "f"time_escaped_epoch: {time_escaped:.3f} hours, "