ValueError on w2v-bert-v2[train]

[allandclive]

On google colab; ValueError: Label values must be <= vocab_size: 29

[allandclive]

`--------------------------------------------------------------------------- ValueError Traceback (most recent call last) in <cell line: 1>() ----> 1 TRAINER.train() # resume_from_checkpoint=True # only if resume

9 frames /usr/local/lib/python3.10/dist-packages/transformers/models/wav2vec2_bert/modeling_wav2vec2_bert.py in forward(self, input_features, attention_mask, output_attentions, output_hidden_states, return_dict, labels) 1245 if labels is not None: 1246 if labels.max() >= self.config.vocab_size: -> 1247 raise ValueError(f"Label values must be <= vocab_size: {self.config.vocab_size}") 1248 1249 # retrieve loss input_lengths from attention_mask

ValueError: Label values must be <= vocab_size: 29`

[phineas-pta]

seem like u tried to fine tune for uganda language

in that case i think u should follow the official guide: https://huggingface.co/blog/fine-tune-w2v2-bert

my use case (vietnamese language) is a bit more mainstream so my script is simplified a lot comparing to official guide