0.7.0 版本生成的trainer_log.json不完整

[Cucunnber]

Reminder

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

Reproduction

sft训练，记录的trainer_log只能到0.87epoch

训练脚本

deepspeed src/train.py --model_name_or_path=xxxx \
        --stage sft \
        --dataset "sft_dataset_merge" \
        --finetuning_type  full \
        --overwrite_cache \
        --preprocessing_num_workers  32 \
        --template  xxx \
        --flash_attn fa2 \
        --output_dir  xxx \
        --bf16  true  \
        --do_train  true  \
        --do_eval false \
        --seed 42 \
        --gradient_accumulation_steps 2 \
        --learning_rate  1e-05 \
        --warmup_ratio 0.02 \
        --cutoff_len 4096 \
        --tf32 true \
        --logging_steps  10 \
        --logging_strategy  steps \
        --lr_scheduler_type  cosine \
        --max_steps  -1 \
        --num_train_epochs  3 \
        --overwrite_output_dir  true  \
        --per_device_train_batch_size 4 \
        --remove_unused_columns  true \
        --report_to tensorboard \
        --plot_loss \
        --save_steps 2000 \
        --eval_steps 200 \
        --val_size 0.01 \
        --evaluation_strategy steps \
        --load_best_model_at_end \
        --save_total_limit  2 \
        --save_safetensors  true  \
        --deepspeed=ds_z3_lr_schedule.json

trainer_log记录

{"current_steps": 10, "total_steps": 690, "loss": 0.635, "learning_rate": 7.1428571428571436e-06, "epoch": 0.04, "percentage": 1.45, "elapsed_time": "0:01:55", "remaining_time": "2:11:22"}
{"current_steps": 20, "total_steps": 690, "loss": 0.4894, "learning_rate": 9.998056338091415e-06, "epoch": 0.09, "percentage": 2.9, "elapsed_time": "0:03:22", "remaining_time": "1:53:18"}
{"current_steps": 30, "total_steps": 690, "loss": 0.486, "learning_rate": 9.986183876164412e-06, "epoch": 0.13, "percentage": 4.35, "elapsed_time": "0:05:26", "remaining_time": "1:59:35"}
{"current_steps": 40, "total_steps": 690, "loss": 0.473, "learning_rate": 9.96354437049027e-06, "epoch": 0.17, "percentage": 5.8, "elapsed_time": "0:07:04", "remaining_time": "1:54:54"}
{"current_steps": 50, "total_steps": 690, "loss": 0.4353, "learning_rate": 9.930186708264902e-06, "epoch": 0.22, "percentage": 7.25, "elapsed_time": "0:08:47", "remaining_time": "1:52:30"}
{"current_steps": 60, "total_steps": 690, "loss": 0.4402, "learning_rate": 9.88618292120984e-06, "epoch": 0.26, "percentage": 8.7, "elapsed_time": "0:10:45", "remaining_time": "1:52:56"}
{"current_steps": 70, "total_steps": 690, "loss": 0.4109, "learning_rate": 9.831628030028698e-06, "epoch": 0.3, "percentage": 10.14, "elapsed_time": "0:12:39", "remaining_time": "1:52:05"}
{"current_steps": 80, "total_steps": 690, "loss": 0.4154, "learning_rate": 9.76663983922178e-06, "epoch": 0.35, "percentage": 11.59, "elapsed_time": "0:14:33", "remaining_time": "1:51:01"}
{"current_steps": 90, "total_steps": 690, "loss": 0.3968, "learning_rate": 9.691358682701927e-06, "epoch": 0.39, "percentage": 13.04, "elapsed_time": "0:16:23", "remaining_time": "1:49:16"}
{"current_steps": 100, "total_steps": 690, "loss": 0.3841, "learning_rate": 9.605947120760878e-06, "epoch": 0.43, "percentage": 14.49, "elapsed_time": "0:18:04", "remaining_time": "1:46:40"}
{"current_steps": 110, "total_steps": 690, "loss": 0.3916, "learning_rate": 9.510589589040554e-06, "epoch": 0.48, "percentage": 15.94, "elapsed_time": "0:20:04", "remaining_time": "1:45:51"}
{"current_steps": 120, "total_steps": 690, "loss": 0.3583, "learning_rate": 9.405492000267228e-06, "epoch": 0.52, "percentage": 17.39, "elapsed_time": "0:21:59", "remaining_time": "1:44:25"}
{"current_steps": 130, "total_steps": 690, "loss": 0.3739, "learning_rate": 9.29088129960862e-06, "epoch": 0.56, "percentage": 18.84, "elapsed_time": "0:23:53", "remaining_time": "1:42:54"}
{"current_steps": 140, "total_steps": 690, "loss": 0.3806, "learning_rate": 9.16700497461403e-06, "epoch": 0.61, "percentage": 20.29, "elapsed_time": "0:25:50", "remaining_time": "1:41:32"}
{"current_steps": 150, "total_steps": 690, "loss": 0.3589, "learning_rate": 9.034130520795774e-06, "epoch": 0.65, "percentage": 21.74, "elapsed_time": "0:27:37", "remaining_time": "1:39:26"}
{"current_steps": 160, "total_steps": 690, "loss": 0.3358, "learning_rate": 8.892544864005899e-06, "epoch": 0.69, "percentage": 23.19, "elapsed_time": "0:29:16", "remaining_time": "1:36:57"}
{"current_steps": 170, "total_steps": 690, "loss": 0.3258, "learning_rate": 8.742553740855507e-06, "epoch": 0.74, "percentage": 24.64, "elapsed_time": "0:31:16", "remaining_time": "1:35:40"}
{"current_steps": 180, "total_steps": 690, "loss": 0.3344, "learning_rate": 8.584481038514573e-06, "epoch": 0.78, "percentage": 26.09, "elapsed_time": "0:33:10", "remaining_time": "1:33:59"}
{"current_steps": 190, "total_steps": 690, "loss": 0.3343, "learning_rate": 8.418668095317912e-06, "epoch": 0.82, "percentage": 27.54, "elapsed_time": "0:34:33", "remaining_time": "1:30:56"}
{"current_steps": 200, "total_steps": 690, "loss": 0.3376, "learning_rate": 8.245472963687484e-06, "epoch": 0.87, "percentage": 28.99, "elapsed_time": "0:36:13", "remaining_time": "1:28:45"}
{"current_steps": 200, "total_steps": 690, "eval_loss": 0.31446942687034607, "epoch": 0.87, "percentage": 28.99, "elapsed_time": "0:36:21", "remaining_time": "1:29:04"}

Expected behavior

我使用新版本进行了多次sft训练，都遇到了这个问题。

System Info

• transformers version: 4.39.3
• Platform: Linux-5.15.0-25-generic-x86_64-with-glibc2.35
• Python version: 3.10.12
• Huggingface_hub version: 0.20.1
• Safetensors version: 0.4.3
• Accelerate version: 0.27.2
• Accelerate config: - compute_environment: LOCAL_MACHINE
  - distributed_type: MULTI_GPU
  - mixed_precision: bf16
  - use_cpu: False
  - debug: True
  - num_processes: 8
  - machine_rank: 0
  - num_machines: 1
  - gpu_ids: all
  - rdzv_backend: static
  - same_network: True
  - main_training_function: main
  - downcast_bf16: no
  - tpu_use_cluster: False
  - tpu_use_sudo: False
  - tpu_env: []
• PyTorch version (GPU?): 2.1.1+cu121 (True)
• Tensorflow version (GPU?): 2.15.0 (True)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using GPU in script?:
• Using distributed or parallel set-up in script?:

Others

No response

[Jungle728]

same question

[hiyouga]

fixed