[2023-12-31 20:06:51,176] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2023-12-31 20:06:55,387] [WARNING] [runner.py:202:fetch_hostfile] Unable to find hostfile, will proceed with training with local resources only.
Detected CUDA_VISIBLE_DEVICES=0,1,2,3 but ignoring it because one or several of --include/--exclude/--num_gpus/--num_nodes cl args were used. If you want to use CUDA_VISIBLE_DEVICES don't pass any of these arguments to deepspeed.
[2023-12-31 20:06:55,387] [INFO] [runner.py:571:main] cmd = /data/jiongxiao_wang/anaconda3/envs/safe-rlhf/bin/python -u -m deepspeed.launcher.launch --world_info=eyJsb2NhbGhvc3QiOiBbMCwgMSwgMiwgM119 --master_addr=127.0.0.1 --master_port=56337 --module --enable_each_rank_log=None safe_rlhf.finetune --train_datasets alpaca --model_name_or_path huggyllama/llama-7b --max_length 512 --trust_remote_code True --epochs 3 --per_device_train_batch_size 4 --per_device_eval_batch_size 4 --gradient_accumulation_steps 16 --gradient_checkpointing --learning_rate 2e-5 --lr_scheduler_type cosine --lr_warmup_ratio 0.03 --weight_decay 0.0 --seed 42 --output_dir /data/jiongxiao_wang/rlhf_attack/safe-rlhf/output/sft --log_type wandb --log_project Safe-RLHF-SFT --zero_stage 3 --bf16 True --tf32 True
[2023-12-31 20:06:57,487] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2023-12-31 20:07:00,478] [INFO] [launch.py:145:main] WORLD INFO DICT: {'localhost': [0, 1, 2, 3]}
[2023-12-31 20:07:00,478] [INFO] [launch.py:151:main] nnodes=1, num_local_procs=4, node_rank=0
[2023-12-31 20:07:00,478] [INFO] [launch.py:162:main] global_rank_mapping=defaultdict(<class 'list'>, {'localhost': [0, 1, 2, 3]})
[2023-12-31 20:07:00,478] [INFO] [launch.py:163:main] dist_world_size=4
[2023-12-31 20:07:00,478] [INFO] [launch.py:165:main] Setting CUDA_VISIBLE_DEVICES=0,1,2,3
[2023-12-31 20:07:02,815] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2023-12-31 20:07:02,856] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2023-12-31 20:07:03,011] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2023-12-31 20:07:03,040] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2023-12-31 20:07:10,639] [INFO] [comm.py:637:init_distributed] cdb=None
[2023-12-31 20:07:10,640] [INFO] [comm.py:637:init_distributed] cdb=None
[2023-12-31 20:07:10,640] [INFO] [comm.py:668:init_distributed] Initializing TorchBackend in DeepSpeed with backend nccl
[2023-12-31 20:07:10,670] [INFO] [comm.py:637:init_distributed] cdb=None
[2023-12-31 20:07:10,675] [INFO] [comm.py:637:init_distributed] cdb=None
Set logger level to WARNING.
ninja: no work to do.
Time to load fused_adam op: 0.14865803718566895 seconds
Time to load fused_adam op: 0.2057504653930664 seconds
Time to load fused_adam op: 0.20213913917541504 seconds
Time to load fused_adam op: 0.2022261619567871 seconds
Parameter Offload: Total persistent parameters: 266240 in 65 params
***** Running training *****
Saving model to "/data/jiongxiao_wang/rlhf_attack/safe-rlhf/output/sft" ...
Saving DeepSpeed Checkpoints...
Converting DeepSpeed Checkpoints to Hugging Face format...
[2023-12-31 21:51:42,560] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Processing zero checkpoint './global_step609'
Detected checkpoint of type zero stage 3, world_size: 4
Parsing checkpoint created by deepspeed==0.12.6
Reconstructed Trainable fp32 state dict with 291 params 6738423808 elements
Saving fp32 state dict to pytorch_model.bin
Model saved!
[2023-12-31 21:52:50,198] [INFO] [launch.py:347:main] Process 189883 exits successfully.
[2023-12-31 21:52:50,198] [INFO] [launch.py:347:main] Process 189885 exits successfully.
[2023-12-31 21:52:50,198] [INFO] [launch.py:347:main] Process 189884 exits successfully.
[2023-12-31 21:52:58,206] [INFO] [launch.py:347:main] Process 189882 exits successfully.