minicpm_2_6o全量微调验证集eval_loss不计算不打印，也不绘制eval_loss图

[Maflyflyy]

Reminder

• I have read the above rules and searched the existing issues.

System Info

Reminder

I have read the README and searched the existing issues.
System Info
llamafactory version: 0.9.2.dev0
Python version: 3.10.
PyTorch version: 2.2.0 (NPU)
Transformers version: 4.44.2
NPU type: 910b
Reproduction
model
model_name_or_path: minicpm-o
template: minicpm_o
trust_remote_code: true

method
stage: sft
do_train: true
finetuning_type: sft
xxxxxxx

dataset
dataset: train_dataset
template: xxxxx
cutoff_len: 2048
max_samples: 180000000
overwrite_cache: true
preprocessing_num_workers: 16

output
output_dir: saves/train_output
logging_steps: 10
save_steps: 100
plot_loss: true
overwrite_output_dir: true

train
per_device_train_batch_size: 2
gradient_accumulation_steps: 4
learning_rate: 0.0001
num_train_epochs: 10.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
fp16: true

eval
val_size: 0.1
per_device_eval_batch_size: 2
evaluation_strategy: steps
eval_steps: 100

Reproduction

运行后会进行验证操作，但没有输出eval_loss，以及最后的eval_loss变化图像。训练日志中，仅有{'eval_runtime': 226.5617, 'eval_samples_per_second': 4.423, 'eval_steps_per_second': 0.737, 'epoch': 1.0}，里面不包含eval_loss。并且训练日志中有提示 WARNING - llamafactory.extras.ploting - No metric eval_loss to plot.
同样的config，qwen2正常输出eval_loss和绘图，但是minicpm不行

Others

No response

[Maflyflyy]

同样的yaml改成qwen2的，qwen2正常输出eval_loss和绘图，但是minicpm不行