[Question]: LlamaForCausalLMPipe 不支持eval_batch？

[happyfc1]

请提出你的问题

自己尝试llama做predict，遇到问题 python -u -m paddle.distributed.launch \ --gpus "6,7" \ --log_dir "output/$task_name""_log" \ run_pretrain.py \ --model_type "llama" \ --model_name_or_path "facebook/llama-13b" \ --tokenizer_name_or_path "facebook/llama-13b" \ --input_dir "./data" \ --output_dir "output/$task_name" \ --split 949,50,1 \ --max_seq_length 2048 \ --per_device_train_batch_size 1 \ --gradient_accumulation_steps 4 \ --per_device_eval_batch_size 4 \ --scale_loss 512 \ --tensor_parallel_degree 1 \ --pipeline_parallel_degree 2 \ --virtual_pp_degree 1 \ --sequence_parallel 0 \ --learning_rate 0.00001 \ --min_learning_rate 0.000001 \ --max_steps 10000 \ --save_steps 5000 \ --weight_decay 0.01 \ --warmup_ratio 0.01 \ --max_grad_norm 1.0 \ --logging_steps 10 \ --dataloader_num_workers 1 \ --eval_steps 1000 \ --report_to "visualdl" \ --sharding "stage1" \ --disable_tqdm true \ --continue_training 1\ --recompute 1 \ --do_predict \ --device "gpu"

Traceback (most recent call last): File "/home/PaddleNLP/llm/llama/run_pretrain_fc.py", line 549, in main() File "/home/PaddleNLP/llm/llama/run_pretrain_fc.py", line 544, in main test_ret = trainer.predict(test_dataset) File "/home/PaddleNLP/paddlenlp/trainer/trainer.py", line 2139, in predict output = eval_loop( File "/home/PaddleNLP/paddlenlp/trainer/trainer.py", line 2011, in evaluation_loop loss, logits, labels = self.prediction_step(model, inputs, prediction_loss_only, ignore_keys=ignore_keys) File "/home/PaddleNLP/paddlenlp/trainer/trainer.py", line 2229, in prediction_step return self.prediction_pipeline_step(model, inputs, prediction_loss_only, ignore_keys) File "/home/PaddleNLP/paddlenlp/trainer/trainer.py", line 2188, in prediction_pipeline_step loss = model.eval_batch([inputs, labels], compute_loss=True) File "/root/anaconda3/lib/python3.9/site-packages/paddle/nn/layer/layers.py", line 1474, in getattr return object.getattribute(self, name) AttributeError: 'LlamaForCausalLMPipe' object has no attribute 'eval_batch'

[w5688414]

请问您的paddle和paddlenlp版本是多少？