The meaning of the eval parameter in qwen-full-sft.yml in the llama factory library

[yuanllong]

Reminder

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

System Info

Description
train
per_device_train_batch_size: 1
gradient_accumulation_steps: 2
learning_rate: 1.0e-5
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000

eval
val_size: 0.1
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 500

What does the eval parameter mean? Will the verification set be put back into the training set to continue training after verification?

Reproduction

Description
train
per_device_train_batch_size: 1
gradient_accumulation_steps: 2
learning_rate: 1.0e-5
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000

eval
val_size: 0.1
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 500

What does the eval parameter mean? Will the verification set be put back into the training set to continue training after verification?

Others

Description
train
per_device_train_batch_size: 1
gradient_accumulation_steps: 2
learning_rate: 1.0e-5
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000

eval
val_size: 0.1
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 500

What does the eval parameter mean? Will the verification set be put back into the training set to continue training after verification?

[yuanllong]

How do I turn off the options for the validation set?

[hiyouga]

Remove

val_size: 0.1
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 500