Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

数据并行可能导致的模型训练报错 (#2139)

pointerhacker70645e48072b
diff --git a/funasr/models/sense_voice/model.py b/funasr/models/sense_voice/model.py
index 25e9fafd3..ba82091e9 100644
--- a/funasr/models/sense_voice/model.py
+++ b/funasr/models/sense_voice/model.py
@@ -555,7 +555,8 @@ class SenseVoiceEncoderSmall(nn.Module):
ilens: torch.Tensor,
):
"""Embed positions in tensor."""
- masks = sequence_mask(ilens, device=ilens.device)[:, None, :]
+ maxlen = xs_pad.shape[1]
+ masks = sequence_mask(ilens, maxlen = maxlen, device=ilens.device)[:, None, :]
xs_pad *= self.output_size() ** 0.5