here are my questions,I have more than 4 gpus to run the train.py,but it still out of memory,I check the usage of memory and find that one of them overflows and produce the bug,how can I solve it?

[z1968357787]

WARNING:torch.distributed.run:

---

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.

---

Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a configuration with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a configuration with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a configuration with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a configuration with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Explicitly passing a revision is encouraged when loading a model with custom code to ensure no malicious code has been contributed in a newer revision.
Traceback (most recent call last):
File "", line 1, in
FileNotFoundError: [Errno 2] No such file or directory: '/home/cike/.cache/huggingface/modules/transformers_modules/THUDM/chatglm-6b/551a50efec3acc5a9b94de8ec46d33d0f81919f7/modeling_chatglm.py'
Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:15<00:00, 1.92s/it]
Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:15<00:00, 1.95s/it]
Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:15<00:00, 1.96s/it]
Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:15<00:00, 1.97s/it]
trainable_params:22020096 (0.35%), non_trainable_params:6255206400
trainable_params:22020096 (0.35%), non_trainable_params:6255206400
trainable_params:22020096 (0.35%), non_trainable_params:6255206400
trainable_params:22020096 (0.35%), non_trainable_params:6255206400
[2023-04-06 08:33:02,712] [INFO] [logging.py:93:log_dist] [Rank -1] DeepSpeed info: version=0.8.3, git-hash=unknown, git-branch=unknown
[2023-04-06 08:33:02,747] [INFO] [logging.py:93:log_dist] [Rank -1] DeepSpeed info: version=0.8.3, git-hash=unknown, git-branch=unknown
[2023-04-06 08:33:02,756] [INFO] [logging.py:93:log_dist] [Rank -1] DeepSpeed info: version=0.8.3, git-hash=unknown, git-branch=unknown
[2023-04-06 08:33:02,874] [INFO] [logging.py:93:log_dist] [Rank -1] DeepSpeed info: version=0.8.3, git-hash=unknown, git-branch=unknown
[2023-04-06 08:33:24,442] [INFO] [logging.py:93:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: False
[2023-04-06 08:33:24,445] [INFO] [logging.py:93:log_dist] [Rank 0] Removing param_group that has no 'params' in the client Optimizer
[2023-04-06 08:33:24,445] [INFO] [logging.py:93:log_dist] [Rank 0] Using client Optimizer as basic optimizer
[2023-04-06 08:33:24,509] [INFO] [logging.py:93:log_dist] [Rank 0] DeepSpeed Basic Optimizer = AdamW
[2023-04-06 08:33:24,509] [INFO] [utils.py:55:is_zero_supported_optimizer] Checking ZeRO support for optimizer=AdamW type=<class 'torch.optim.adamw.AdamW'>
[2023-04-06 08:33:24,509] [INFO] [logging.py:93:log_dist] [Rank 0] Creating torch.bfloat16 ZeRO stage 2 optimizer
[2023-04-06 08:33:24,510] [INFO] [stage_1_and_2.py:144:init] Reduce bucket size 500,000,000
[2023-04-06 08:33:24,510] [INFO] [stage_1_and_2.py:145:init] Allgather bucket size 500,000,000
[2023-04-06 08:33:24,510] [INFO] [stage_1_and_2.py:146:init] CPU Offload: False
[2023-04-06 08:33:24,510] [INFO] [stage_1_and_2.py:147:init] Round robin gradient partitioning: False
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
Emitting ninja build file /home/cike/.cache/torch_extensions/py39_cu117/utils/build.ninja...
Building extension module utils...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
ninja: no work to do.
Loading extension module utils...
Time to load utils op: 0.35059165954589844 seconds
Loading extension module utils...
Time to load utils op: 0.40517687797546387 seconds
Loading extension module utils...
Time to load utils op: 0.40523695945739746 seconds
Loading extension module utils...
Time to load utils op: 0.40430521965026855 seconds
Rank: 3 partition count [4] and sizes[(5505024, False)]
Rank: 2 partition count [4] and sizes[(5505024, False)]
Rank: 0 partition count [4] and sizes[(5505024, False)]
Rank: 1 partition count [4] and sizes[(5505024, False)]
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module utils, skipping build step...
Loading extension module utils...
Time to load utils op: 0.00042438507080078125 seconds
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module utils, skipping build step...
Loading extension module utils...
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
Time to load utils op: 0.00040459632873535156 seconds
No modifications detected for re-loaded extension module utils, skipping build step...
Loading extension module utils...
Time to load utils op: 0.00035071372985839844 seconds
0%| | 0/12241 [00:00<?, ?it/s][2023-04-06 08:33:25,838] [INFO] [utils.py:829:see_memory_usage] Before initializing optimizer states
[2023-04-06 08:33:25,839] [INFO] [utils.py:830:see_memory_usage] MA 11.71 GB Max_MA 11.72 GB CA 11.75 GB Max_CA 12 GB
[2023-04-06 08:33:25,839] [INFO] [utils.py:838:see_memory_usage] CPU Virtual Memory: used = 12.0 GB, percent = 4.8%
[2023-04-06 08:33:26,025] [INFO] [utils.py:829:see_memory_usage] After initializing optimizer states
[2023-04-06 08:33:26,025] [INFO] [utils.py:830:see_memory_usage] MA 11.76 GB Max_MA 11.82 GB CA 11.85 GB Max_CA 12 GB
[2023-04-06 08:33:26,026] [INFO] [utils.py:838:see_memory_usage] CPU Virtual Memory: used = 12.0 GB, percent = 4.8%
[2023-04-06 08:33:26,026] [INFO] [stage_1_and_2.py:520:init] optimizer state initialized
[2023-04-06 08:33:26,092] [INFO] [utils.py:829:see_memory_usage] After initializing ZeRO optimizer
[2023-04-06 08:33:26,093] [INFO] [utils.py:830:see_memory_usage] MA 11.76 GB Max_MA 11.76 GB CA 11.85 GB Max_CA 12 GB
[2023-04-06 08:33:26,093] [INFO] [utils.py:838:see_memory_usage] CPU Virtual Memory: used = 12.0 GB, percent = 4.8%
[2023-04-06 08:33:26,094] [INFO] [logging.py:93:log_dist] [Rank 0] DeepSpeed Final Optimizer = AdamW
[2023-04-06 08:33:26,095] [INFO] [logging.py:93:log_dist] [Rank 0] DeepSpeed using client LR scheduler
[2023-04-06 08:33:26,095] [INFO] [logging.py:93:log_dist] [Rank 0] DeepSpeed LR Scheduler = None
[2023-04-06 08:33:26,095] [INFO] [logging.py:93:log_dist] [Rank 0] step=0, skipped=0, lr=[0.0], mom=[(0.9, 0.999)]
[2023-04-06 08:33:26,096] [INFO] [config.py:1018:print] DeepSpeedEngine configuration:
[2023-04-06 08:33:26,096] [INFO] [config.py:1022:print] activation_checkpointing_config {
"partition_activations": false,
"contiguous_memory_optimization": false,
"cpu_checkpointing": false,
"number_checkpoints": null,
"synchronize_checkpoint_boundary": false,
"profile": false
}
[2023-04-06 08:33:26,096] [INFO] [config.py:1022:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2023-04-06 08:33:26,096] [INFO] [config.py:1022:print] amp_enabled .................. False
[2023-04-06 08:33:26,096] [INFO] [config.py:1022:print] amp_params ................... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] autotuning_config ............ {
"enabled": false,
"start_step": null,
"end_step": null,
"metric_path": null,
"arg_mappings": null,
"metric": "throughput",
"model_info": null,
"results_dir": "autotuning_results",
"exps_dir": "autotuning_exps",
"overwrite": true,
"fast": true,
"start_profile_step": 3,
"end_profile_step": 5,
"tuner_type": "gridsearch",
"tuner_early_stopping": 5,
"tuner_num_trials": 50,
"model_info_path": null,
"mp_size": 1,
"max_train_batch_size": null,
"min_train_batch_size": 1,
"max_train_micro_batch_size_per_gpu": 1.024000e+03,
"min_train_micro_batch_size_per_gpu": 1,
"num_tuning_micro_batch_sizes": 3
}
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] bfloat16_enabled ............. True
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] checkpoint_parallel_write_pipeline False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] checkpoint_tag_validation_enabled True
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] checkpoint_tag_validation_fail False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7f15140266a0>
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] communication_data_type ...... None
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] compression_config ........... {'weight_quantization': {'shared_parameters': {'enabled': False, 'quantizer_kernel': False, 'schedule_offset': 0, 'quantize_groups': 1, 'quantize_verbose': False, 'quantization_type': 'symmetric', 'quantize_weight_in_forward': False, 'rounding': 'nearest', 'fp16_mixed_quantize': False, 'quantize_change_ratio': 0.001}, 'different_groups': {}}, 'activation_quantization': {'shared_parameters': {'enabled': False, 'quantization_type': 'symmetric', 'range_calibration': 'dynamic', 'schedule_offset': 1000}, 'different_groups': {}}, 'sparse_pruning': {'shared_parameters': {'enabled': False, 'method': 'l1', 'schedule_offset': 1000}, 'different_groups': {}}, 'row_pruning': {'shared_parameters': {'enabled': False, 'method': 'l1', 'schedule_offset': 1000}, 'different_groups': {}}, 'head_pruning': {'shared_parameters': {'enabled': False, 'method': 'topk', 'schedule_offset': 1000}, 'different_groups': {}}, 'channel_pruning': {'shared_parameters': {'enabled': False, 'method': 'l1', 'schedule_offset': 1000}, 'different_groups': {}}, 'layer_reduction': {'enabled': False}}
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] curriculum_enabled_legacy .... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] curriculum_params_legacy ..... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] data_efficiency_config ....... {'enabled': False, 'seed': 1234, 'data_sampling': {'enabled': False, 'num_epochs': 1000, 'num_workers': 0, 'curriculum_learning': {'enabled': False}}, 'data_routing': {'enabled': False, 'random_ltd': {'enabled': False, 'layer_token_lr_schedule': {'enabled': False}}}}
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] data_efficiency_enabled ...... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] dataloader_drop_last ......... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] disable_allgather ............ False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] dump_state ................... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] dynamic_loss_scale_args ...... None
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_enabled ........... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_gas_boundary_resolution 1
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_layer_name ........ bert.encoder.layer
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_layer_num ......... 0
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_max_iter .......... 100
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_stability ......... 1e-06
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_tol ............... 0.01
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] eigenvalue_verbose ........... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] elasticity_enabled ........... False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] flops_profiler_config ........ {
"enabled": false,
"profile_step": 1,
"module_depth": -1,
"top_modules": 1,
"detailed": true,
"output_file": null
}
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] fp16_auto_cast ............... None
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] fp16_enabled ................. False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] fp16_master_weights_and_gradients False
[2023-04-06 08:33:26,097] [INFO] [config.py:1022:print] global_rank .................. 0
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] grad_accum_dtype ............. None
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] gradient_accumulation_steps .. 8
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] gradient_clipping ............ 0.0
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] gradient_predivide_factor .... 1.0
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] initial_dynamic_scale ........ 1
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] load_universal_checkpoint .... False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] loss_scale ................... 1.0
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] memory_breakdown ............. False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] monitor_config ............... tensorboard=TensorBoardConfig(enabled=False, output_path='', job_name='DeepSpeedJobName') wandb=WandbConfig(enabled=False, group=None, team=None, project='deepspeed') csv_monitor=CSVConfig(enabled=False, output_path='', job_name='DeepSpeedJobName') enabled=False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] nebula_config ................ {
"enabled": false,
"persistent_storage_path": null,
"persistent_time_interval": 100,
"num_of_version_in_retention": 2,
"enable_nebula_load": true,
"load_path": null
}
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] optimizer_legacy_fusion ...... False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] optimizer_name ............... None
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] optimizer_params ............. None
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] pld_enabled .................. False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] pld_params ................... False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] prescale_gradients ........... False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] scheduler_name ............... None
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] scheduler_params ............. None
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] sparse_attention ............. None
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] sparse_gradients_enabled ..... False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] steps_per_print .............. inf
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] train_batch_size ............. 32
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] train_micro_batch_size_per_gpu 1
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] use_node_local_storage ....... False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] wall_clock_breakdown ......... False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] world_size ................... 4
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] zero_allow_untested_optimizer True
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] zero_config .................. stage=2 contiguous_gradients=True reduce_scatter=True reduce_bucket_size=500,000,000 allgather_partitions=True allgather_bucket_size=500,000,000 overlap_comm=False load_from_fp32_weights=True elastic_checkpoint=False offload_param=DeepSpeedZeroOffloadParamConfig(device='none', nvme_path=None, buffer_count=5, buffer_size=100,000,000, max_in_cpu=1,000,000,000, pin_memory=False) offload_optimizer=DeepSpeedZeroOffloadOptimizerConfig(device='none', nvme_path=None, buffer_count=4, pin_memory=False, pipeline=False, pipeline_read=False, pipeline_write=False, fast_init=False) sub_group_size=1,000,000,000 cpu_offload_param=None cpu_offload_use_pin_memory=None cpu_offload=None prefetch_bucket_size=50,000,000 param_persistence_threshold=100,000 model_persistence_threshold=sys.maxsize max_live_parameters=1,000,000,000 max_reuse_distance=1,000,000,000 gather_16bit_weights_on_model_save=False stage3_gather_fp16_weights_on_model_save=False ignore_unused_parameters=True legacy_stage1=False round_robin_gradients=False
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] zero_enabled ................. True
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] zero_force_ds_cpu_optimizer .. True
[2023-04-06 08:33:26,098] [INFO] [config.py:1022:print] zero_optimization_stage ...... 2
[2023-04-06 08:33:26,099] [INFO] [config.py:1007:print_user_config] json = {
"train_batch_size": 32,
"train_micro_batch_size_per_gpu": 1,
"gradient_accumulation_steps": 8,
"zero_optimization": {
"stage": 2,
"offload_optimizer": {
"device": "none"
},
"offload_param": {
"device": "none"
},
"stage3_gather_16bit_weights_on_model_save": false
},
"steps_per_print": inf,
"bf16": {
"enabled": true
},
"fp16": {
"enabled": false
},
"zero_allow_untested_optimizer": true
}
Using /home/cike/.cache/torch_extensions/py39_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module utils, skipping build step...
Loading extension module utils...
Time to load utils op: 0.00032019615173339844 seconds
loss: 2.640625: 0%| | 1/12241 [00:04<14:19:20, 4.21s/it]
Traceback (most recent call last):
File "/home/cike/zzp/LoRA/ChatGLM-finetune-LoRA/train.py", line 220, in
accelerator.backward(loss)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/accelerate/accelerator.py", line 1677, in backward
self.deepspeed_engine_wrapped.backward(loss, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/accelerate/utils/deepspeed.py", line 167, in backward
self.engine.backward(loss, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/utils/nvtx.py", line 11, in wrapped_fn
ret_val = func(*args, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/engine.py", line 2008, in backward
self.allreduce_gradients()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/utils/nvtx.py", line 11, in wrapped_fn
ret_val = func(*args, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/engine.py", line 1918, in allreduce_gradients
self.optimizer.overlapping_partition_gradients_reduce_epilogue()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 834, in overlapping_partition_gradients_reduce_epilogue
self.independent_gradient_partition_epilogue()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 720, in independent_gradient_partition_epilogue
self.reduce_ipg_grads()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 1287, in reduce_ipg_grads
self.average_tensor(self.ipg_buffer[self.ipg_index])
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 1018, in average_tensor
tensor_to_reduce = tensor.to(self.communication_data_type)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 954.00 MiB (GPU 1; 15.90 GiB total capacity; 12.84 GiB already allocated; 927.75 MiB free; 14.17 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
loss: 2.75: 0%| | 1/12241 [00:04<15:35:54, 4.59s/it]
Traceback (most recent call last):
File "/home/cike/zzp/LoRA/ChatGLM-finetune-LoRA/train.py", line 220, in
accelerator.backward(loss)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/accelerate/accelerator.py", line 1677, in backward
self.deepspeed_engine_wrapped.backward(loss, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/accelerate/utils/deepspeed.py", line 167, in backward
self.engine.backward(loss, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/utils/nvtx.py", line 11, in wrapped_fn
ret_val = func(*args, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/engine.py", line 2008, in backward
self.allreduce_gradients()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/utils/nvtx.py", line 11, in wrapped_fn
ret_val = func(*args, **kwargs)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/engine.py", line 1918, in allreduce_gradients
self.optimizer.overlapping_partition_gradients_reduce_epilogue()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 834, in overlapping_partition_gradients_reduce_epilogue
self.independent_gradient_partition_epilogue()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 720, in independent_gradient_partition_epilogue
self.reduce_ipg_grads()
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 1287, in reduce_ipg_grads
self.average_tensor(self.ipg_buffer[self.ipg_index])
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 1018, in average_tensor
tensor_to_reduce = tensor.to(self.communication_data_type)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 954.00 MiB (GPU 3; 15.90 GiB total capacity; 12.88 GiB already allocated; 903.75 MiB free; 14.23 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
WARNING:torch.distributed.elastic.multiprocessing.api:Sending process 107844 closing signal SIGTERM
WARNING:torch.distributed.elastic.multiprocessing.api:Sending process 107846 closing signal SIGTERM
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 1 (pid: 107845) of binary: /home/cike/anaconda/envs/lora/bin/python
Traceback (most recent call last):
File "/home/cike/anaconda/envs/lora/bin/accelerate", line 8, in
sys.exit(main())
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/accelerate/commands/accelerate_cli.py", line 45, in main
args.func(args)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/accelerate/commands/launch.py", line 908, in launch_command
deepspeed_launcher(args)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/accelerate/commands/launch.py", line 647, in deepspeed_launcher
distrib_run.run(args)
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/torch/distributed/run.py", line 785, in run
elastic_launch(
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/torch/distributed/launcher/api.py", line 134, in call
return launch_agent(self._config, self._entrypoint, list(args))
File "/home/cike/anaconda/envs/lora/lib/python3.9/site-packages/torch/distributed/launcher/api.py", line 250, in launch_agent
raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:

train.py FAILED

Failures:
[1]:
time : 2023-04-06_08:33:33
host : 4d9275d5570f
rank : 3 (local_rank: 3)
exitcode : 1 (pid: 107847)
error_file: <N/A>
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html

Root Cause (first observed failure):
[0]:
time : 2023-04-06_08:33:33
host : 4d9275d5570f
rank : 1 (local_rank: 1)
exitcode : 1 (pid: 107845)
error_file: <N/A>
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html