同一个数据集和模型，相同参数设置，训练两次，0.5epoch时会因为模型见到数据顺序不同的原因导致很大效果差异吗？

[tiphaineeee]

Reminder

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

System Info

model

model_name_or_path: qwen/Qwen2-1.5B-Instruct

method

stage: sft
do_train: true
finetuning_type: full

lora_target: all

deepspeed: examples/deepspeed/ds_z3_config.json

dataset

dataset: xxx
template: qwen
cutoff_len: 1024
max_samples: 1000000000 #1000
overwrite_cache: true
preprocessing_num_workers: 16

output

output_dir: xxx

output_dir: xxx
logging_steps: 10
save_steps: 200
plot_loss: true
overwrite_output_dir: true
report_to: tensorboard
logging_dir: xxx

train

per_device_train_batch_size: 4
gradient_accumulation_steps: 16
learning_rate: 1.0e-4
num_train_epochs: 5
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000

eval

val_size: 0.002
per_device_eval_batch_size: 4
eval_strategy: steps
eval_steps: 200 #500

Reproduction

Put your message here.

Others

No response