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1163110135c6
diff --git a/funasr/datasets/openai_datasets/datasets.py b/funasr/datasets/openai_datasets/datasets.pyindex e13c15618..9b36078f3 100644--- a/funasr/datasets/openai_datasets/datasets.py+++ b/funasr/datasets/openai_datasets/datasets.py@@ -51,7 +51,7 @@ class OpenAIDataset(torch.utils.data.Dataset):self.batch_size = kwargs.get("batch_size")self.batch_type = kwargs.get("batch_type")self.prompt_ids_len = 0- self.retry = kwargs.get("retry", 10)+ self.retry = kwargs.get("retry", 100)self.permute = Falsefrom funasr.frontends.whisper_frontend import WhisperFrontend@@ -212,13 +212,9 @@ class OpenAIDataset(torch.utils.data.Dataset):if self.batch_type != "example":b, t = outputs["input_ids"].shapeif b > 1 and b * t > self.batch_size * self.batch_size_scale_ratio_max:- # beg = torch.randint(0, 2, ()).item()- # if b < 2:- # beg = 0logging.info(- f"Warning, b*t: {b}*{t}={b * t} > batch_size*relax: {self.batch_size_scale_ratio_max}*{self.batch_size}={self.batch_size_scale_ratio_max*self.batch_size}, drop half data {idx}th, beg:{beg}"+ f"Warning, b*t: {b}*{t}={b * t} > batch_size*relax: {self.batch_size_scale_ratio_max}*{self.batch_size}={self.batch_size_scale_ratio_max*self.batch_size}, drop last data")- # samples = samples[beg : beg + b : 2]samples = samples[:-1]continue