funasr/models/llm_asr_nar/model.py
@@ -315,7 +315,7 @@ model_outputs = self.llm(inputs_embeds=inputs_embeds, attention_mask=attention_mask, labels=None) preds = torch.argmax(model_outputs.logits, -1) text = tokenizer.batch_decode(preds, add_special_tokens=False, skip_special_tokens=True) text = text.split(': "\n')[-1] text = text.split(': \n')[-1] # preds = torch.argmax(model_outputs.logits, -1) ibest_writer = None