NLG评估DPO，不输出结果

[sunxiaoyu12]

Reminder

• I have read the README and searched the existing issues.

System Info

• llamafactory version: 0.9.1.dev0
• Platform: Linux-6.2.0-26-generic-x86_64-with-glibc2.35
• Python version: 3.10.12
• PyTorch version: 2.3.0a0+ebedce2 (GPU)
• Transformers version: 4.46.1
• Datasets version: 3.1.0
• Accelerate version: 1.0.1
• PEFT version: 0.12.0
• TRL version: 0.9.6
• GPU type: NVIDIA H100 80GB HBM3
• DeepSpeed version: 0.15.4
• Bitsandbytes version: 0.44.1

Reproduction

CUDA_VISIBLE_DEVICES=0,1,2,3 llamafactory-cli train \
--stage dpo \
--do_predict True \
--finetuning_type lora \
--model_name_or_path  model/Meta-Llama-3-8B-Instruct \
--adapter_name_or_path  saves/Meta-Llama-3-8B-Instruct/lora/dpo_train_2024-12-12-10-01-38 \
--flash_attn auto \
--dataset_dir data \
--eval_dataset dpo_zh_demo \
--template llama3 \
--cutoff_len 1024 \
--max_samples 1000 \
--preprocessing_num_workers 16 \
--predict_with_generate False \
--per_device_eval_batch_size 2 \
--max_new_tokens 512 \
--top_p 0.7 \
--temperature 0.95 \
--output_dir saves/Meta-Llama-3-8B-Instruct/lora/eval_nlg_dpo_2024-12-12-09 \
--overwrite_output_dir True

Expected behavior

训练阶段是DPO，微调方法是lora，训练完成之后，使用以上代码进行模型NLG评估，运行期间并未出错，但是却没有结果输出，请问这是怎么回事？是否是少了什么参数？

Others

No response