RuntimeError: The size of tensor a (768) must match the size of tensor b (3072) at non-singleton dimension 0

root@mz32-01:/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat#
root@mz32-01:/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat# python3 train.py --step 3 --deployment-type multi_node
---=== Running Step 3 ===---
Running:
bash /home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/training_scripts/multi_node/run_125m.sh /home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/output/actor-models/125m /home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/output/reward-models/125m '' '' /home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/output/step3-models/125m
[2023-07-16 15:08:13,237] [INFO] [real_accelerator.py:110:get_accelerator] Setting ds_accelerator to cuda (auto detect)
........
[2023-07-16 15:08:33,417] [INFO] [config.py:960:print] DeepSpeedEngine configuration:
[2023-07-16 15:08:33,417] [INFO] [config.py:964:print] activation_checkpointing_config {
"partition_activations": false,
"contiguous_memory_optimization": false,
"cpu_checkpointing": false,
"number_checkpoints": null,
"synchronize_checkpoint_boundary": false,
"profile": false
}
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] amp_enabled .................. False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] amp_params ................... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964: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
}
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] bfloat16_enabled ............. False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] checkpoint_parallel_write_pipeline False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] checkpoint_tag_validation_enabled True
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] checkpoint_tag_validation_fail False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7f676c057070>
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] communication_data_type ...... None
[2023-07-16 15:08:33,418] [INFO] [config.py:964: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}}
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] curriculum_enabled_legacy .... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] curriculum_params_legacy ..... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964: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}}}}
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] data_efficiency_enabled ...... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] dataloader_drop_last ......... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] disable_allgather ............ False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] dump_state ................... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] dynamic_loss_scale_args ...... {'init_scale': 65536, 'scale_window': 100, 'delayed_shift': 2, 'consecutive_hysteresis': False, 'min_scale': 1}
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_enabled ........... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_gas_boundary_resolution 1
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_layer_name ........ bert.encoder.layer
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_layer_num ......... 0
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_max_iter .......... 100
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_stability ......... 1e-06
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_tol ............... 0.01
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] eigenvalue_verbose ........... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] elasticity_enabled ........... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] flops_profiler_config ........ {
"enabled": false,
"recompute_fwd_factor": 0.0,
"profile_step": 1,
"module_depth": -1,
"top_modules": 1,
"detailed": true,
"output_file": null
}
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] fp16_auto_cast ............... False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] fp16_enabled ................. True
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] fp16_master_weights_and_gradients False
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] global_rank .................. 0
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] grad_accum_dtype ............. None
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] gradient_accumulation_steps .. 1
[2023-07-16 15:08:33,418] [INFO] [config.py:964:print] gradient_clipping ............ 1.0
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] gradient_predivide_factor .... 1.0
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] hybrid_engine ................ enabled=True max_out_tokens=512 inference_tp_size=2 release_inference_cache=False pin_parameters=True tp_gather_partition_size=128
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] initial_dynamic_scale ........ 65536
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] load_universal_checkpoint .... False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] loss_scale ................... 0
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] memory_breakdown ............. False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] mics_hierarchial_params_gather False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] mics_shard_size .............. -1
[2023-07-16 15:08:33,419] [INFO] [config.py:964: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
[2023-07-16 15:08:33,419] [INFO] [config.py:964: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
}
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] optimizer_legacy_fusion ...... False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] optimizer_name ............... None
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] optimizer_params ............. None
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] pld_enabled .................. False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] pld_params ................... False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] prescale_gradients ........... False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] scheduler_name ............... None
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] scheduler_params ............. None
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] sparse_attention ............. None
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] sparse_gradients_enabled ..... False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] steps_per_print .............. 10
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] train_batch_size ............. 256
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] train_micro_batch_size_per_gpu 128
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] use_node_local_storage ....... False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] wall_clock_breakdown ......... False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] world_size ................... 2
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] zero_allow_untested_optimizer False
[2023-07-16 15:08:33,419] [INFO] [config.py:964: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='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=30000000 param_persistence_threshold=10000 model_persistence_threshold=sys.maxsize max_live_parameters=30000000 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=False
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] zero_enabled ................. True
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] zero_force_ds_cpu_optimizer .. True
[2023-07-16 15:08:33,419] [INFO] [config.py:964:print] zero_optimization_stage ...... 3
[2023-07-16 15:08:33,419] [INFO] [config.py:950:print_user_config] json = {
"train_batch_size": 256,
"train_micro_batch_size_per_gpu": 128,
"steps_per_print": 10,
"zero_optimization": {
"stage": 3,
"offload_param": {
"device": "none"
},
"offload_optimizer": {
"device": "none"
},
"stage3_param_persistence_threshold": 1.000000e+04,
"stage3_max_live_parameters": 3.000000e+07,
"stage3_prefetch_bucket_size": 3.000000e+07,
"memory_efficient_linear": false
},
"fp16": {
"enabled": true,
"loss_scale_window": 100
},
"gradient_clipping": 1.0,
"prescale_gradients": false,
"wall_clock_breakdown": false,
"hybrid_engine": {
"enabled": true,
"max_out_tokens": 512,
"inference_tp_size": 2,
"release_inference_cache": false,
"pin_parameters": true,
"tp_gather_partition_size": 128
}
}
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
Detected CUDA files, patching ldflags
Emitting ninja build file /root/.cache/torch_extensions/py38_cu117/transformer_inference/build.ninja...
Building extension module transformer_inference...
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 transformer_inference...
Time to load transformer_inference op: 0.4970717430114746 seconds
Loading extension module transformer_inference...
Time to load transformer_inference op: 0.5848901271820068 seconds
[2023-07-16 15:08:34,199] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed-Inference config: {'layer_id': 0, 'hidden_size': 768, 'intermediate_size': 3072, 'heads': 12, 'num_hidden_layers': -1, 'dtype': torch.float16, 'pre_layer_norm': True, 'norm_type': <NormType.LayerNorm: 1>, 'local_rank': -1, 'stochastic_mode': False, 'epsilon': 1e-05, 'mp_size': 2, 'scale_attention': True, 'triangular_masking': True, 'local_attention': False, 'window_size': 1, 'rotary_dim': -1, 'rotate_half': False, 'rotate_every_two': True, 'return_tuple': True, 'mlp_after_attn': True, 'mlp_act_func_type': <ActivationFuncType.ReLU: 2>, 'specialized_mode': False, 'training_mp_size': 1, 'bigscience_bloom': False, 'max_out_tokens': 512, 'min_out_tokens': 512, 'scale_attn_by_inverse_layer_idx': False, 'enable_qkv_quantization': False, 'use_mup': False, 'return_single_tuple': False, 'set_empty_params': True, 'transposed_mode': True}
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module transformer_inference, skipping build step...
Loading extension module transformer_inference...
Time to load transformer_inference op: 0.08850884437561035 seconds
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module transformer_inference, skipping build step...
Loading extension module transformer_inference...
Time to load transformer_inference op: 0.09563326835632324 seconds
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module transformer_inference, skipping build step...
Loading extension module transformer_inference...
Time to load transformer_inference op: 0.08705496788024902 seconds
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module transformer_inference, skipping build step...
Loading extension module transformer_inference...
Time to load transformer_inference op: 0.08372807502746582 seconds
[end] Initialized Actor Model [end] (duration: 7.58s)*
************************[start] Initializing Ref Model [start] **************************
[2023-07-16 15:08:34,910] [INFO] [partition_parameters.py:453:exit] finished initializing model with 0.33B parameters
[2023-07-16 15:08:35,158] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed info: version=0.9.5, git-hash=unknown, git-branch=unknown
[2023-07-16 15:08:35,164] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: False
[2023-07-16 15:08:35,164] [INFO] [logging.py:96:log_dist] [Rank 0] Creating ZeRO Offload
[2023-07-16 15:08:35,478] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [begin]
[2023-07-16 15:08:35,479] [INFO] [utils.py:786:see_memory_usage] MA 1.67 GB Max_MA 1.85 GB CA 2.14 GB Max_CA 2 GB
[2023-07-16 15:08:35,479] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.52 GB, percent = 3.4%
Parameter Offload: Total persistent parameters: 121344 in 122 params
[2023-07-16 15:08:35,790] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [end]
[2023-07-16 15:08:35,790] [INFO] [utils.py:786:see_memory_usage] MA 1.67 GB Max_MA 1.67 GB CA 2.14 GB Max_CA 2 GB
[2023-07-16 15:08:35,791] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.7 GB, percent = 3.5%
[2023-07-16 15:08:35,791] [INFO] [config.py:960:print] DeepSpeedEngine configuration:
[2023-07-16 15:08:35,791] [INFO] [config.py:964:print] activation_checkpointing_config {
"partition_activations": false,
"contiguous_memory_optimization": false,
"cpu_checkpointing": false,
"number_checkpoints": null,
"synchronize_checkpoint_boundary": false,
"profile": false
}
[2023-07-16 15:08:35,791] [INFO] [config.py:964:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2023-07-16 15:08:35,791] [INFO] [config.py:964:print] amp_enabled .................. False
[2023-07-16 15:08:35,791] [INFO] [config.py:964:print] amp_params ................... False
[2023-07-16 15:08:35,791] [INFO] [config.py:964: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
}
[2023-07-16 15:08:35,791] [INFO] [config.py:964:print] bfloat16_enabled ............. False
[2023-07-16 15:08:35,791] [INFO] [config.py:964:print] checkpoint_parallel_write_pipeline False
[2023-07-16 15:08:35,791] [INFO] [config.py:964:print] checkpoint_tag_validation_enabled True
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] checkpoint_tag_validation_fail False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7f6742b42190>
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] communication_data_type ...... None
[2023-07-16 15:08:35,792] [INFO] [config.py:964: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}}
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] curriculum_enabled_legacy .... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] curriculum_params_legacy ..... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964: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}}}}
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] data_efficiency_enabled ...... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] dataloader_drop_last ......... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] disable_allgather ............ False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] dump_state ................... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] dynamic_loss_scale_args ...... None
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_enabled ........... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_gas_boundary_resolution 1
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_layer_name ........ bert.encoder.layer
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_layer_num ......... 0
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_max_iter .......... 100
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_stability ......... 1e-06
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_tol ............... 0.01
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] eigenvalue_verbose ........... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] elasticity_enabled ........... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] flops_profiler_config ........ {
"enabled": false,
"recompute_fwd_factor": 0.0,
"profile_step": 1,
"module_depth": -1,
"top_modules": 1,
"detailed": true,
"output_file": null
}
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] fp16_auto_cast ............... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] fp16_enabled ................. True
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] fp16_master_weights_and_gradients False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] global_rank .................. 0
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] grad_accum_dtype ............. None
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] gradient_accumulation_steps .. 1
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] gradient_clipping ............ 1.0
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] gradient_predivide_factor .... 1.0
[2023-07-16 15:08:35,792] [INFO] [config.py:964: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
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] initial_dynamic_scale ........ 65536
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] load_universal_checkpoint .... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] loss_scale ................... 0
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] memory_breakdown ............. False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] mics_hierarchial_params_gather False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] mics_shard_size .............. -1
[2023-07-16 15:08:35,792] [INFO] [config.py:964: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
[2023-07-16 15:08:35,792] [INFO] [config.py:964: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
}
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] optimizer_legacy_fusion ...... False
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] optimizer_name ............... None
[2023-07-16 15:08:35,792] [INFO] [config.py:964:print] optimizer_params ............. None
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] pld_enabled .................. False
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] pld_params ................... False
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] prescale_gradients ........... False
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] scheduler_name ............... None
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] scheduler_params ............. None
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] sparse_attention ............. None
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] sparse_gradients_enabled ..... False
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] steps_per_print .............. 10
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] train_batch_size ............. 256
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] train_micro_batch_size_per_gpu 128
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] use_node_local_storage ....... False
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] wall_clock_breakdown ......... False
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] world_size ................... 2
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] zero_allow_untested_optimizer False
[2023-07-16 15:08:35,793] [INFO] [config.py:964: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='none', nvme_path=None, buffer_count=5, buffer_size=100,000,000, max_in_cpu=1,000,000,000, pin_memory=False) offload_optimizer=None 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=10000 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=False
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] zero_enabled ................. True
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] zero_force_ds_cpu_optimizer .. True
[2023-07-16 15:08:35,793] [INFO] [config.py:964:print] zero_optimization_stage ...... 3
[2023-07-16 15:08:35,793] [INFO] [config.py:950:print_user_config] json = {
"train_batch_size": 256,
"train_micro_batch_size_per_gpu": 128,
"steps_per_print": 10,
"zero_optimization": {
"stage": 3,
"stage3_param_persistence_threshold": 1.000000e+04,
"offload_param": {
"device": "none"
},
"memory_efficient_linear": false
},
"fp16": {
"enabled": true
},
"gradient_clipping": 1.0,
"prescale_gradients": false,
"wall_clock_breakdown": false
}
[end] Initialized Ref Model [end] (duration: 1.05s)
************************[start] Initializing Critic Model [start] ************************
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module fused_adam, skipping build step...
Loading extension module fused_adam...
Time to load fused_adam op: 0.0006570816040039062 seconds
Using /root/.cache/torch_extensions/py38_cu117 as PyTorch extensions root...
No modifications detected for re-loaded extension module fused_adam, skipping build step...
Loading extension module fused_adam...
Time to load fused_adam op: 0.0010995864868164062 seconds
[2023-07-16 15:08:37,364] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed info: version=0.9.5, git-hash=unknown, git-branch=unknown
[2023-07-16 15:08:37,438] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: False
[2023-07-16 15:08:37,439] [INFO] [logging.py:96:log_dist] [Rank 0] Removing param_group that has no 'params' in the client Optimizer
[2023-07-16 15:08:37,439] [INFO] [logging.py:96:log_dist] [Rank 0] Using client Optimizer as basic optimizer
[2023-07-16 15:08:37,443] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Basic Optimizer = FusedAdam
[2023-07-16 15:08:37,443] [INFO] [utils.py:54:is_zero_supported_optimizer] Checking ZeRO support for optimizer=FusedAdam type=<class 'deepspeed.ops.adam.fused_adam.FusedAdam'>
[2023-07-16 15:08:37,443] [INFO] [logging.py:96:log_dist] [Rank 0] Creating fp16 ZeRO stage 3 optimizer, MiCS is enabled False, Hierarchical params gather False
[2023-07-16 15:08:37,443] [INFO] [logging.py:96:log_dist] [Rank 0] Creating torch.float16 ZeRO stage 3 optimizer
[2023-07-16 15:08:37,776] [INFO] [utils.py:785:see_memory_usage] Stage 3 initialize beginning
[2023-07-16 15:08:37,777] [INFO] [utils.py:786:see_memory_usage] MA 1.9 GB Max_MA 1.9 GB CA 2.14 GB Max_CA 2 GB
[2023-07-16 15:08:37,777] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:37,779] [INFO] [stage3.py:113:init] Reduce bucket size 500,000,000
[2023-07-16 15:08:37,779] [INFO] [stage3.py:114:init] Prefetch bucket size 30000000
[2023-07-16 15:08:38,115] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [begin]
[2023-07-16 15:08:38,116] [INFO] [utils.py:786:see_memory_usage] MA 1.9 GB Max_MA 1.9 GB CA 2.14 GB Max_CA 2 GB
[2023-07-16 15:08:38,116] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
Parameter Offload: Total persistent parameters: 122112 in 123 params
[2023-07-16 15:08:38,491] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [end]
[2023-07-16 15:08:38,491] [INFO] [utils.py:786:see_memory_usage] MA 1.78 GB Max_MA 1.94 GB CA 2.17 GB Max_CA 2 GB
[2023-07-16 15:08:38,492] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:38,834] [INFO] [utils.py:785:see_memory_usage] Before creating fp16 partitions
[2023-07-16 15:08:38,834] [INFO] [utils.py:786:see_memory_usage] MA 1.78 GB Max_MA 1.78 GB CA 2.17 GB Max_CA 2 GB
[2023-07-16 15:08:38,835] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:39,554] [INFO] [utils.py:785:see_memory_usage] After creating fp16 partitions: 2
[2023-07-16 15:08:39,555] [INFO] [utils.py:786:see_memory_usage] MA 1.78 GB Max_MA 1.78 GB CA 1.92 GB Max_CA 2 GB
[2023-07-16 15:08:39,556] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:39,896] [INFO] [utils.py:785:see_memory_usage] Before creating fp32 partitions
[2023-07-16 15:08:39,897] [INFO] [utils.py:786:see_memory_usage] MA 1.78 GB Max_MA 1.78 GB CA 1.92 GB Max_CA 2 GB
[2023-07-16 15:08:39,897] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:40,240] [INFO] [utils.py:785:see_memory_usage] After creating fp32 partitions
[2023-07-16 15:08:40,241] [INFO] [utils.py:786:see_memory_usage] MA 2.02 GB Max_MA 2.13 GB CA 2.27 GB Max_CA 2 GB
[2023-07-16 15:08:40,241] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:40,582] [INFO] [utils.py:785:see_memory_usage] Before initializing optimizer states
[2023-07-16 15:08:40,583] [INFO] [utils.py:786:see_memory_usage] MA 2.02 GB Max_MA 2.02 GB CA 2.27 GB Max_CA 2 GB
[2023-07-16 15:08:40,583] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:40,927] [INFO] [utils.py:785:see_memory_usage] After initializing optimizer states
[2023-07-16 15:08:40,928] [INFO] [utils.py:786:see_memory_usage] MA 2.48 GB Max_MA 2.72 GB CA 2.98 GB Max_CA 3 GB
[2023-07-16 15:08:40,928] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 8.89 GB, percent = 3.5%
[2023-07-16 15:08:40,928] [INFO] [stage3.py:387:_setup_for_real_optimizer] optimizer state initialized
[2023-07-16 15:08:41,277] [INFO] [utils.py:785:see_memory_usage] After initializing ZeRO optimizer
[2023-07-16 15:08:41,277] [INFO] [utils.py:786:see_memory_usage] MA 3.53 GB Max_MA 3.68 GB CA 3.95 GB Max_CA 4 GB
[2023-07-16 15:08:41,277] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 9.05 GB, percent = 3.6%
[2023-07-16 15:08:41,278] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Final Optimizer = FusedAdam
[2023-07-16 15:08:41,278] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed using client LR scheduler
[2023-07-16 15:08:41,278] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed LR Scheduler = <torch.optim.lr_scheduler.LambdaLR object at 0x7f6742500880>
[2023-07-16 15:08:41,278] [INFO] [logging.py:96:log_dist] [Rank 0] step=0, skipped=0, lr=[0.0, 0.0], mom=[(0.9, 0.95), (0.9, 0.95)]
[2023-07-16 15:08:41,278] [INFO] [config.py:960:print] DeepSpeedEngine configuration:
[2023-07-16 15:08:41,278] [INFO] [config.py:964:print] activation_checkpointing_config {
"partition_activations": false,
"contiguous_memory_optimization": false,
"cpu_checkpointing": false,
"number_checkpoints": null,
"synchronize_checkpoint_boundary": false,
"profile": false
}
[2023-07-16 15:08:41,278] [INFO] [config.py:964:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2023-07-16 15:08:41,278] [INFO] [config.py:964:print] amp_enabled .................. False
[2023-07-16 15:08:41,278] [INFO] [config.py:964:print] amp_params ................... False
[2023-07-16 15:08:41,278] [INFO] [config.py:964: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
}
[2023-07-16 15:08:41,278] [INFO] [config.py:964:print] bfloat16_enabled ............. False
[2023-07-16 15:08:41,278] [INFO] [config.py:964:print] checkpoint_parallel_write_pipeline False
[2023-07-16 15:08:41,278] [INFO] [config.py:964:print] checkpoint_tag_validation_enabled True
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] checkpoint_tag_validation_fail False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7f67427aa430>
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] communication_data_type ...... None
[2023-07-16 15:08:41,279] [INFO] [config.py:964: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}}
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] curriculum_enabled_legacy .... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] curriculum_params_legacy ..... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964: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}}}}
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] data_efficiency_enabled ...... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] dataloader_drop_last ......... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] disable_allgather ............ False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] dump_state ................... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] dynamic_loss_scale_args ...... {'init_scale': 65536, 'scale_window': 100, 'delayed_shift': 2, 'consecutive_hysteresis': False, 'min_scale': 1}
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_enabled ........... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_gas_boundary_resolution 1
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_layer_name ........ bert.encoder.layer
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_layer_num ......... 0
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_max_iter .......... 100
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_stability ......... 1e-06
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_tol ............... 0.01
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] eigenvalue_verbose ........... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] elasticity_enabled ........... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] flops_profiler_config ........ {
"enabled": false,
"recompute_fwd_factor": 0.0,
"profile_step": 1,
"module_depth": -1,
"top_modules": 1,
"detailed": true,
"output_file": null
}
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] fp16_auto_cast ............... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] fp16_enabled ................. True
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] fp16_master_weights_and_gradients False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] global_rank .................. 0
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] grad_accum_dtype ............. None
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] gradient_accumulation_steps .. 1
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] gradient_clipping ............ 1.0
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] gradient_predivide_factor .... 1.0
[2023-07-16 15:08:41,279] [INFO] [config.py:964: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
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] initial_dynamic_scale ........ 65536
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] load_universal_checkpoint .... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] loss_scale ................... 0
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] memory_breakdown ............. False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] mics_hierarchial_params_gather False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] mics_shard_size .............. -1
[2023-07-16 15:08:41,279] [INFO] [config.py:964: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
[2023-07-16 15:08:41,279] [INFO] [config.py:964: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
}
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] optimizer_legacy_fusion ...... False
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] optimizer_name ............... None
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] optimizer_params ............. None
[2023-07-16 15:08:41,279] [INFO] [config.py:964:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] pld_enabled .................. False
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] pld_params ................... False
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] prescale_gradients ........... False
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] scheduler_name ............... None
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] scheduler_params ............. None
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] sparse_attention ............. None
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] sparse_gradients_enabled ..... False
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] steps_per_print .............. 10
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] train_batch_size ............. 256
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] train_micro_batch_size_per_gpu 128
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] use_node_local_storage ....... False
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] wall_clock_breakdown ......... False
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] world_size ................... 2
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] zero_allow_untested_optimizer False
[2023-07-16 15:08:41,280] [INFO] [config.py:964: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='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=30000000 param_persistence_threshold=10000 model_persistence_threshold=sys.maxsize max_live_parameters=30000000 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=False
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] zero_enabled ................. True
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] zero_force_ds_cpu_optimizer .. True
[2023-07-16 15:08:41,280] [INFO] [config.py:964:print] zero_optimization_stage ...... 3
[2023-07-16 15:08:41,280] [INFO] [config.py:950:print_user_config] json = {
"train_batch_size": 256,
"train_micro_batch_size_per_gpu": 128,
"steps_per_print": 10,
"zero_optimization": {
"stage": 3,
"offload_param": {
"device": "none"
},
"offload_optimizer": {
"device": "none"
},
"stage3_param_persistence_threshold": 1.000000e+04,
"stage3_max_live_parameters": 3.000000e+07,
"stage3_prefetch_bucket_size": 3.000000e+07,
"memory_efficient_linear": false
},
"fp16": {
"enabled": true,
"loss_scale_window": 100
},
"gradient_clipping": 1.0,
"prescale_gradients": false,
"wall_clock_breakdown": false,
"hybrid_engine": {
"enabled": false,
"max_out_tokens": 512,
"inference_tp_size": 1,
"release_inference_cache": false,
"pin_parameters": true,
"tp_gather_partition_size": 8
}
}
[end] Initialized Critic Model [end] (duration: 5.49s)
************************[start] Initializing Reward Model [start] ************************
[2023-07-16 15:08:42,782] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed info: version=0.9.5, git-hash=unknown, git-branch=unknown
[2023-07-16 15:08:42,847] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: False
[2023-07-16 15:08:42,848] [INFO] [logging.py:96:log_dist] [Rank 0] Creating ZeRO Offload
[2023-07-16 15:08:43,181] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [begin]
[2023-07-16 15:08:43,181] [INFO] [utils.py:786:see_memory_usage] MA 3.77 GB Max_MA 3.77 GB CA 3.95 GB Max_CA 4 GB
[2023-07-16 15:08:43,182] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 9.3 GB, percent = 3.7%
Parameter Offload: Total persistent parameters: 122112 in 123 params
[2023-07-16 15:08:43,563] [INFO] [utils.py:785:see_memory_usage] DeepSpeedZeRoOffload initialize [end]
[2023-07-16 15:08:43,563] [INFO] [utils.py:786:see_memory_usage] MA 3.65 GB Max_MA 3.81 GB CA 3.98 GB Max_CA 4 GB
[2023-07-16 15:08:43,564] [INFO] [utils.py:793:see_memory_usage] CPU Virtual Memory: used = 9.3 GB, percent = 3.7%
[2023-07-16 15:08:43,564] [INFO] [config.py:960:print] DeepSpeedEngine configuration:
[2023-07-16 15:08:43,564] [INFO] [config.py:964:print] activation_checkpointing_config {
"partition_activations": false,
"contiguous_memory_optimization": false,
"cpu_checkpointing": false,
"number_checkpoints": null,
"synchronize_checkpoint_boundary": false,
"profile": false
}
[2023-07-16 15:08:43,564] [INFO] [config.py:964:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2023-07-16 15:08:43,564] [INFO] [config.py:964:print] amp_enabled .................. False
[2023-07-16 15:08:43,564] [INFO] [config.py:964:print] amp_params ................... False
[2023-07-16 15:08:43,564] [INFO] [config.py:964: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
}
[2023-07-16 15:08:43,564] [INFO] [config.py:964:print] bfloat16_enabled ............. False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] checkpoint_parallel_write_pipeline False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] checkpoint_tag_validation_enabled True
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] checkpoint_tag_validation_fail False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7f67424502b0>
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] communication_data_type ...... None
[2023-07-16 15:08:43,565] [INFO] [config.py:964: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}}
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] curriculum_enabled_legacy .... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] curriculum_params_legacy ..... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964: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}}}}
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] data_efficiency_enabled ...... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] dataloader_drop_last ......... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] disable_allgather ............ False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] dump_state ................... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] dynamic_loss_scale_args ...... None
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_enabled ........... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_gas_boundary_resolution 1
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_layer_name ........ bert.encoder.layer
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_layer_num ......... 0
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_max_iter .......... 100
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_stability ......... 1e-06
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_tol ............... 0.01
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] eigenvalue_verbose ........... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] elasticity_enabled ........... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] flops_profiler_config ........ {
"enabled": false,
"recompute_fwd_factor": 0.0,
"profile_step": 1,
"module_depth": -1,
"top_modules": 1,
"detailed": true,
"output_file": null
}
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] fp16_auto_cast ............... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] fp16_enabled ................. True
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] fp16_master_weights_and_gradients False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] global_rank .................. 0
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] grad_accum_dtype ............. None
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] gradient_accumulation_steps .. 1
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] gradient_clipping ............ 1.0
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] gradient_predivide_factor .... 1.0
[2023-07-16 15:08:43,565] [INFO] [config.py:964: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
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] initial_dynamic_scale ........ 65536
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] load_universal_checkpoint .... False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] loss_scale ................... 0
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] memory_breakdown ............. False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] mics_hierarchial_params_gather False
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] mics_shard_size .............. -1
[2023-07-16 15:08:43,565] [INFO] [config.py:964: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
[2023-07-16 15:08:43,565] [INFO] [config.py:964: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
}
[2023-07-16 15:08:43,565] [INFO] [config.py:964:print] optimizer_legacy_fusion ...... False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] optimizer_name ............... None
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] optimizer_params ............. None
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] pld_enabled .................. False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] pld_params ................... False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] prescale_gradients ........... False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] scheduler_name ............... None
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] scheduler_params ............. None
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] sparse_attention ............. None
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] sparse_gradients_enabled ..... False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] steps_per_print .............. 10
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] train_batch_size ............. 256
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] train_micro_batch_size_per_gpu 128
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] use_node_local_storage ....... False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] wall_clock_breakdown ......... False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] world_size ................... 2
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] zero_allow_untested_optimizer False
[2023-07-16 15:08:43,566] [INFO] [config.py:964: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='none', nvme_path=None, buffer_count=5, buffer_size=100,000,000, max_in_cpu=1,000,000,000, pin_memory=False) offload_optimizer=None 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=10000 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=False
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] zero_enabled ................. True
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] zero_force_ds_cpu_optimizer .. True
[2023-07-16 15:08:43,566] [INFO] [config.py:964:print] zero_optimization_stage ...... 3
[2023-07-16 15:08:43,566] [INFO] [config.py:950:print_user_config] json = {
"train_batch_size": 256,
"train_micro_batch_size_per_gpu": 128,
"steps_per_print": 10,
"zero_optimization": {
"stage": 3,
"stage3_param_persistence_threshold": 1.000000e+04,
"offload_param": {
"device": "none"
},
"memory_efficient_linear": false
},
"fp16": {
"enabled": true
},
"gradient_clipping": 1.0,
"prescale_gradients": false,
"wall_clock_breakdown": false
}
[end] Initialized Reward Model [end] (duration: 2.29s)
***** Running training *****
Beginning of Epoch 1/1, Total Generation Batches 120
Traceback (most recent call last):
File "main.py", line 520, in
main()
File "main.py", line 428, in main
out = trainer.generate_experience(batch_prompt['prompt'],
File "/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py", line 98, in generate_experience
seq = self._generate_sequence(prompts, mask)
File "/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py", line 73, in _generate_sequence
seq = self.actor_model.module.generate(prompts,
File "/usr/local/lib/python3.8/dist-packages/deepspeed/runtime/hybrid_engine.py", line 207, in generate
self._fuse_lora(self.layer_params[layer_id], self.lora_params[layer_id])
File "/usr/local/lib/python3.8/dist-packages/deepspeed/runtime/hybrid_engine.py", line 137, in _fuse_lora
weight.data += lora_scaling * torch.matmul(lora_left_weight.t(), lora_right_weight.t())
RuntimeError: The size of tensor a (768) must match the size of tensor b (3072) at non-singleton dimension 0
Traceback (most recent call last):
File "main.py", line 520, in
main()
File "main.py", line 428, in main
out = trainer.generate_experience(batch_prompt['prompt'],
File "/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py", line 98, in generate_experience
seq = self._generate_sequence(prompts, mask)
File "/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/training/step3_rlhf_finetuning/ppo_trainer.py", line 73, in _generate_sequence
seq = self.actor_model.module.generate(prompts,
File "/usr/local/lib/python3.8/dist-packages/deepspeed/runtime/hybrid_engine.py", line 207, in generate
self._fuse_lora(self.layer_params[layer_id], self.lora_params[layer_id])
File "/usr/local/lib/python3.8/dist-packages/deepspeed/runtime/hybrid_engine.py", line 137, in _fuse_lora
weight.data += lora_scaling * torch.matmul(lora_left_weight.t(), lora_right_weight.t())
RuntimeError: The size of tensor a (768) must match the size of tensor b (3072) at non-singleton dimension 0
[2023-07-16 15:08:46,492] [INFO] [launch.py:315:sigkill_handler] Killing subprocess 7239
[2023-07-16 15:08:46,502] [INFO] [launch.py:315:sigkill_handler] Killing subprocess 7240
[2023-07-16 15:08:46,502] [ERROR] [launch.py:321:sigkill_handler] ['/usr/bin/python3', '-u', 'main.py', '--local_rank=1', '--data_path', 'Dahoas/rm-static', '--data_split', '2,4,4', '--actor_model_name_or_path', '/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/output/actor-models/125m', '--critic_model_name_or_path', '/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/output/reward-models/125m', '--num_padding_at_beginning', '1', '--per_device_train_batch_size', '128', '--per_device_mini_train_batch_size', '128', '--generation_batch_numbers', '1', '--ppo_epochs', '1', '--max_answer_seq_len', '256', '--max_prompt_seq_len', '256', '--actor_learning_rate', '5e-4', '--critic_learning_rate', '5e-6', '--actor_weight_decay', '0.1', '--critic_weight_decay', '0.1', '--num_train_epochs', '1', '--lr_scheduler_type', 'cosine', '--gradient_accumulation_steps', '1', '--num_warmup_steps', '100', '--deepspeed', '--seed', '1234', '--enable_hybrid_engine', '--inference_tp_size', '2', '--tp_gather_partition_size', '128', '--actor_zero_stage', '3', '--critic_zero_stage', '3', '--actor_gradient_checkpointing', '--disable_actor_dropout', '--actor_lora_dim', '128', '--actor_lora_module_name', 'decoder.layers.', '--output_dir', '/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat/output/step3-models/125m'] exits with return code = 1
---=== Finished Step 3 in 0:00:34 ===---
root@mz32-01:/home/amd01/llm_dev/DeepSpeedExamples/applications/DeepSpeed-Chat#

Start with train.py --step 3 --deployment-type multi_node and the referenced training/step3_rlhf_finetuning/training_scripts/multi_node/run_125m.sh entry point. Reproduce the Step 3 run and trace the reported 768/3072 tensor-shape mismatch. Done means identifying the failing configuration or input and making the multi-node command complete successfully.