wjustus01/dana-voice-fixed

这是一个针对Unslotch微调优化的 Dana 语音数据集的修复和过滤版本。数据集对音频数据的numpy数组格式进行了修复，并过滤掉了超过1900个token的样本，以避免在微调过程中出现输入ID长度超过模型最大序列长度的错误。This is a fixed and filtered version of the Dana voice dataset optimized for Unsloth fine-tuning. The dataset has been fixed for numpy array format issues in audio data and filtered out examples exceeding the token limit of 1900 to prevent the Input IDs length > models max sequence length error during fine-tuning.