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373fbd15e0af
diff --git a/funasr/datasets/openai_datasets/datasets.py b/funasr/datasets/openai_datasets/datasets.pyindex 9b36078f3..07a7c319a 100644--- a/funasr/datasets/openai_datasets/datasets.py+++ b/funasr/datasets/openai_datasets/datasets.py@@ -63,6 +63,7 @@ class OpenAIDataset(torch.utils.data.Dataset):# self.kwargs = kwargsself.max_token_length = kwargs.get("max_token_length", 1024)self.batch_size_scale_ratio_max = kwargs.get("batch_size_scale_ratio_max", 1.5)+ self.batch_size_token_max = kwargs.get("batch_size_token_max", 2500)def get_source_len(self, index):item = self.index_ds[index]@@ -213,7 +214,7 @@ class OpenAIDataset(torch.utils.data.Dataset):b, t = outputs["input_ids"].shapeif b > 1 and b * t > self.batch_size * self.batch_size_scale_ratio_max:logging.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 last data"+ f"Warning, {idx}th, b*t: {b}*{t}={b * t} > batch_size_sample_max: {self.batch_size_token_max}, drop last data")samples = samples[:-1]continue