search

CVI-SZU / Linly

Chinese-LLaMA 1&2、Chinese-Falcon 基础模型；ChatFlow中文对话模型；中文OpenLLaMA模型；NLP预训练/指令微调数据集
3.03k stars 235 forks source link

Multi machine pre-training hung #111

Open BUPTAnderson opened 1 year ago

BUPTAnderson commented 1 year ago 
deepspeed --hostfile=hostfile pretrain.py --deepspeed --deepspeed_config models/deepspeed_zero3_config.json --enable_zero3 \
                      --pretrained_model_path models/llama-7b.bin \
                      --dataset_path $OUTPUT_DATASET_PATH --spm_model_path $LLaMA_PATH/tokenizer.model \
                      --config_path models/llama/7b_config.json \
                      --output_model_path models/llama_zh_7b \
                      --world_size 8 --data_processor lm  --deepspeed_checkpoint_activations \
                      --total_steps 300000 --save_checkpoint_steps 5000 --batch_size 24

hostfie has 4 v100 machine:

1.1.1.1 slots=8
2.2.2.2 slots=8
3.3.3.3 slots=8
4.4.4.4 slots=8

the master machine 1.1.1.1 print follow log and the pre-training hung:

1.1.1.1: [2023-06-16 15:32:01,673] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: True
1.1.1.1: [2023-06-16 15:32:01,674] [INFO] [logging.py:96:log_dist] [Rank 0] Removing param_group that has no 'params' in the client Optimizer
1.1.1.1: [2023-06-16 15:32:01,674] [INFO] [logging.py:96:log_dist] [Rank 0] Using client Optimizer as basic optimizer
1.1.1.1: [2023-06-16 15:32:01,692] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Basic Optimizer = DeepSpeedCPUAdam
1.1.1.1: [2023-06-16 15:32:01,692] [INFO] [utils.py:51:is_zero_supported_optimizer] Checking ZeRO support for optimizer=DeepSpeedCPUAdam type=<class 'deepspeed.ops.adam.cpu_adam.DeepSpeedCPUAdam'>
1.1.1.1: [2023-06-16 15:32:01,692] [INFO] [logging.py:96:log_dist] [Rank 0] Creating fp16 ZeRO stage 3 optimizer, MiCS is enabled False, Hierarchical params gather False
1.1.1.1: [2023-06-16 15:32:01,692] [INFO] [logging.py:96:log_dist] [Rank 0] Creating torch.float16 ZeRO stage 3 optimizer
1.1.1.1: [2023-06-16 15:32:01,791] [INFO] [utils.py:785:see_memory_usage] Stage 3 initialize beginning
1.1.1.1: [2023-06-16 15:32:01,791] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.73 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:01,792] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 37.82 GB, percent = 15.1%
1.1.1.1: [2023-06-16 15:32:01,794] [INFO] [stage3.py:113:__init__] Reduce bucket size 500,000,000
1.1.1.1: [2023-06-16 15:32:01,794] [INFO] [stage3.py:114:__init__] Prefetch bucket size 50,000,000
1.1.1.1: [2023-06-16 15:32:01,856] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [begin]
1.1.1.1: [2023-06-16 15:32:01,857] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:01,857] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 37.82 GB, percent = 15.1%
1.1.1.1: Parameter Offload: Total persistent parameters: 266240 in 65 params
1.1.1.1: [2023-06-16 15:32:01,941] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [end]
1.1.1.1: [2023-06-16 15:32:01,941] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:01,941] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 37.82 GB, percent = 15.1%
1.1.1.1: [2023-06-16 15:32:02,007] [INFO] [utils.py:785:see_memory_usage] Before creating fp16 partitions
1.1.1.1: [2023-06-16 15:32:02,008] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:02,008] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 37.82 GB, percent = 15.1%
1.1.1.1: [2023-06-16 15:32:10,831] [INFO] [utils.py:785:see_memory_usage] After creating fp16 partitions: 1
1.1.1.1: [2023-06-16 15:32:10,832] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:10,832] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 47.35 GB, percent = 18.9%
1.1.1.1: [2023-06-16 15:32:10,894] [INFO] [utils.py:785:see_memory_usage] Before creating fp32 partitions
1.1.1.1: [2023-06-16 15:32:10,894] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:10,895] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 47.35 GB, percent = 18.9%
1.1.1.1: [2023-06-16 15:32:11,059] [INFO] [utils.py:785:see_memory_usage] After creating fp32 partitions
1.1.1.1: [2023-06-16 15:32:11,059] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:11,059] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 48.14 GB, percent = 19.2%
1.1.1.1: [2023-06-16 15:32:11,247] [INFO] [utils.py:785:see_memory_usage] Before initializing optimizer states
1.1.1.1: [2023-06-16 15:32:11,248] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:11,248] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 58.49 GB, percent = 23.3%
1.1.1.1: [2023-06-16 15:32:13,026] [INFO] [logging.py:96:log_dist] [Rank 0] rank=0 time (ms) | init_optimizer_state: 1731.81
1.1.1.1: [2023-06-16 15:32:13,148] [INFO] [utils.py:785:see_memory_usage] After initializing optimizer states
1.1.1.1: [2023-06-16 15:32:13,149] [INFO] [utils.py:786:see_memory_usage] MA 0.0 GB         Max_MA 0.0 GB         CA 0.73 GB         Max_CA 1 GB
1.1.1.1: [2023-06-16 15:32:13,149] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 68.91 GB, percent = 27.5%
1.1.1.1: [2023-06-16 15:32:13,149] [INFO] [stage3.py:392:_setup_for_real_optimizer] optimizer state initialized
1.1.1.1: [2023-06-16 15:32:15,376] [INFO] [utils.py:785:see_memory_usage] After initializing ZeRO optimizer
1.1.1.1: [2023-06-16 15:32:15,377] [INFO] [utils.py:786:see_memory_usage] MA 0.93 GB         Max_MA 1.42 GB         CA 1.67 GB         Max_CA 2 GB
1.1.1.1: [2023-06-16 15:32:15,377] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory:  used = 73.03 GB, percent = 29.1%
1.1.1.1: [2023-06-16 15:32:15,377] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Final Optimizer = DeepSpeedCPUAdam
1.1.1.1: [2023-06-16 15:32:15,377] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed using client LR scheduler
1.1.1.1: [2023-06-16 15:32:15,377] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed LR Scheduler = <torch.optim.lr_scheduler.LambdaLR object at 0x7f347c617df0>
1.1.1.1: [2023-06-16 15:32:15,378] [INFO] [logging.py:96:log_dist] [Rank 0] step=0, skipped=0, lr=[0.0], mom=[(0.9, 0.999)]
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:955:print] DeepSpeedEngine configuration:
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   activation_checkpointing_config  {
1.1.1.1:     "partition_activations": false,
1.1.1.1:     "contiguous_memory_optimization": false,
1.1.1.1:     "cpu_checkpointing": false,
1.1.1.1:     "number_checkpoints": null,
1.1.1.1:     "synchronize_checkpoint_boundary": false,
1.1.1.1:     "profile": false
1.1.1.1: }
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   amp_enabled .................. False
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   amp_params ................... False
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   autotuning_config ............ {
1.1.1.1:     "enabled": false,
1.1.1.1:     "start_step": null,
1.1.1.1:     "end_step": null,
1.1.1.1:     "metric_path": null,
1.1.1.1:     "arg_mappings": null,
1.1.1.1:     "metric": "throughput",
1.1.1.1:     "model_info": null,
1.1.1.1:     "results_dir": "autotuning_results",
1.1.1.1:     "exps_dir": "autotuning_exps",
1.1.1.1:     "overwrite": true,
1.1.1.1:     "fast": true,
1.1.1.1:     "start_profile_step": 3,
1.1.1.1:     "end_profile_step": 5,
1.1.1.1:     "tuner_type": "gridsearch",
1.1.1.1:     "tuner_early_stopping": 5,
1.1.1.1:     "tuner_num_trials": 50,
1.1.1.1:     "model_info_path": null,
1.1.1.1:     "mp_size": 1,
1.1.1.1:     "max_train_batch_size": null,
1.1.1.1:     "min_train_batch_size": 1,
1.1.1.1:     "max_train_micro_batch_size_per_gpu": 1.024000e+03,
1.1.1.1:     "min_train_micro_batch_size_per_gpu": 1,
1.1.1.1:     "num_tuning_micro_batch_sizes": 3
1.1.1.1: }
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   bfloat16_enabled ............. False
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   checkpoint_parallel_write_pipeline  False
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   checkpoint_tag_validation_enabled  True
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   checkpoint_tag_validation_fail  False
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7f347c6175e0>
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   communication_data_type ...... None
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959: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}}
1.1.1.1: [2023-06-16 15:32:15,379] [INFO] [config.py:959:print]   curriculum_enabled_legacy .... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   curriculum_params_legacy ..... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959: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}}}}
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   data_efficiency_enabled ...... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   dataloader_drop_last ......... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   disable_allgather ............ False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   dump_state ................... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   dynamic_loss_scale_args ...... {'init_scale': 65536, 'scale_window': 1000, 'delayed_shift': 2, 'min_scale': 1}
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_enabled ........... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_gas_boundary_resolution  1
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_layer_name ........ bert.encoder.layer
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_layer_num ......... 0
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_max_iter .......... 100
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_stability ......... 1e-06
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_tol ............... 0.01
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   eigenvalue_verbose ........... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   elasticity_enabled ........... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   flops_profiler_config ........ {
1.1.1.1:     "enabled": true,
1.1.1.1:     "profile_step": 1,
1.1.1.1:     "module_depth": -1,
1.1.1.1:     "top_modules": 3,
1.1.1.1:     "detailed": true,
1.1.1.1:     "output_file": null
1.1.1.1: }
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   fp16_auto_cast ............... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   fp16_enabled ................. True
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   fp16_master_weights_and_gradients  False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   global_rank .................. 0
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   grad_accum_dtype ............. None
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   gradient_accumulation_steps .. 1
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   gradient_clipping ............ 0.0
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   gradient_predivide_factor .... 1.0
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   hybrid_engine ................ enabled=False max_out_tokens=512 inference_tp_size=1 release_inference_cache=False pin_parameters=True tp_gather_partition_size=8
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   initial_dynamic_scale ........ 65536
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   load_universal_checkpoint .... False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   loss_scale ................... 0
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   memory_breakdown ............. False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   mics_hierarchial_params_gather  False
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   mics_shard_size .............. -1
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959: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
1.1.1.1: [2023-06-16 15:32:15,380] [INFO] [config.py:959:print]   nebula_config ................ {
1.1.1.1:     "enabled": false,
1.1.1.1:     "persistent_storage_path": null,
1.1.1.1:     "persistent_time_interval": 100,
1.1.1.1:     "num_of_version_in_retention": 2,
1.1.1.1:     "enable_nebula_load": true,
1.1.1.1:     "load_path": null
1.1.1.1: }
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   optimizer_legacy_fusion ...... False
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   optimizer_name ............... adam
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   optimizer_params ............. {'lr': 1e-05, 'weight_decay': 0.01}
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   pld_enabled .................. False
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   pld_params ................... False
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   prescale_gradients ........... False
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   scheduler_name ............... None
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   scheduler_params ............. None
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   sparse_attention ............. None
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   sparse_gradients_enabled ..... False
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   steps_per_print .............. 100
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   train_batch_size ............. 32
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   train_micro_batch_size_per_gpu  1
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   use_node_local_storage ....... False
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   wall_clock_breakdown ......... True
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   world_size ................... 32
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   zero_allow_untested_optimizer  True
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   zero_config .................. stage=3 contiguous_gradients=True reduce_scatter=True reduce_bucket_size=500,000,000 allgather_partitions=True allgather_bucket_size=500,000,000 overlap_comm=True load_from_fp32_weights=True elastic_checkpoint=False offload_param=DeepSpeedZeroOffloadParamConfig(device='cpu', nvme_path=None, buffer_count=5, buffer_size=100,000,000, max_in_cpu=1,000,000,000, pin_memory=True) offload_optimizer=DeepSpeedZeroOffloadOptimizerConfig(device='cpu', nvme_path=None, buffer_count=4, pin_memory=True, 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 mics_shard_size=-1 mics_hierarchical_params_gather=False memory_efficient_linear=True
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   zero_enabled ................. True
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   zero_force_ds_cpu_optimizer .. True
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:959:print]   zero_optimization_stage ...... 3
1.1.1.1: [2023-06-16 15:32:15,381] [INFO] [config.py:945:print_user_config]   json = {
1.1.1.1:     "gradient_accumulation_steps": 1,
1.1.1.1:     "train_micro_batch_size_per_gpu": 1,
1.1.1.1:     "steps_per_print": 100,
1.1.1.1:     "optimizer": {
1.1.1.1:         "type": "Adam",
1.1.1.1:         "params": {
1.1.1.1:             "lr": 1e-05,
1.1.1.1:             "weight_decay": 0.01
1.1.1.1:         }
1.1.1.1:     },
1.1.1.1:     "flops_profiler": {
1.1.1.1:         "enabled": true,
1.1.1.1:         "profile_step": 1,
1.1.1.1:         "module_depth": -1,
1.1.1.1:         "top_modules": 3,
1.1.1.1:         "detailed": true
1.1.1.1:     },
1.1.1.1:     "fp16": {
1.1.1.1:         "enabled": true,
1.1.1.1:         "loss_scale": 0,
1.1.1.1:         "loss_scale_window": 1000,
1.1.1.1:         "hysteresis": 2,
1.1.1.1:         "min_loss_scale": 1
1.1.1.1:     },
1.1.1.1:     "zero_optimization": {
1.1.1.1:         "stage": 3,
1.1.1.1:         "offload_param": {
1.1.1.1:             "device": "cpu",
1.1.1.1:             "pin_memory": true
1.1.1.1:         },
1.1.1.1:         "offload_optimizer": {
1.1.1.1:             "device": "cpu",
1.1.1.1:             "pin_memory": true
1.1.1.1:         }
1.1.1.1:     },
1.1.1.1:     "activation_checkpointing": {
1.1.1.1:         "partition_activations": false,
1.1.1.1:         "contiguous_memory_optimization": false,
1.1.1.1:         "cpu_checkpointing": false
1.1.1.1:     },
1.1.1.1:     "wall_clock_breakdown": false,
1.1.1.1:     "zero_allow_untested_optimizer": true
1.1.1.1: }
BUPTAnderson commented 1 year ago

Every v100 machine single pre-training for TencentPretrain is ok, and pre-training for FastChat use the same 4 nodes is ok too.