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
total_time/accum_grad
游雁64bf6dd8a1e8
diff --git a/funasr/datasets/audio_datasets/espnet_samplers.py b/funasr/datasets/audio_datasets/espnet_samplers.pyindex e155cd7fc..528f59333 100644--- a/funasr/datasets/audio_datasets/espnet_samplers.py+++ b/funasr/datasets/audio_datasets/espnet_samplers.py@@ -146,10 +146,9 @@ class EspnetStyleBatchSampler(DistributedSampler):start_idx = self.rank * batches_per_rankend_idx = start_idx + batches_per_rankrank_batches = buffer_batches[start_idx + self.start_step : end_idx]- if self.start_step > 0:- logging.info(- f"Warning, rank: {self.rank}, dataloader start from step: {self.start_step}, batch_num_before: {end_idx-start_idx}, now: {len(rank_batches)}"- )+ logging.info(+ f"rank: {self.rank}, dataloader start from step: {self.start_step}, batch_num: {end_idx-start_idx}, batch_num_after_step: {len(rank_batches)}"+ )# Return an iterator over the batches for the current rankreturn iter(rank_batches)