search

KimMeen / Time-LLM

[ICLR 2024] Official implementation of " 🦙 Time-LLM: Time Series Forecasting by Reprogramming Large Language Models"
https://arxiv.org/abs/2310.01728
Apache License 2.0
1.29k stars 221 forks source link

multi GPU error #100

Closed zhangtianhong-1998 closed 3 months ago

zhangtianhong-1998 commented 4 months ago

我用两张A6000 96GB和两张GV100 尝试运行LLama 模型，但是cuda报错 单卡bert是能够正常运行，但是一旦切换到双卡就在soure_embedding前向传播部分开始报错 source embeddings = self.mapping_layer(self.word embeddings.permute(1, 0)).permute(1, 0)

报错如下，请问有碰见过类似情况的吗，求助！

[2024-06-01 22:11:10,870] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) The following values were not passed to accelerate launch and had defaults used instead: --num_machines was set to a value of 1 --dynamo_backend was set to a value of 'no' To avoid this warning pass in values for each of the problematic parameters or run accelerate config. [2024-06-01 22:11:13,842] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-06-01 22:11:14,481] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-06-01 22:11:15,874] [INFO] [comm.py:637:init_distributed] cdb=None [2024-06-01 22:11:16,787] [INFO] [comm.py:637:init_distributed] cdb=None [2024-06-01 22:11:16,787] [INFO] [comm.py:668:init_distributed] Initializing TorchBackend in DeepSpeed with backend nccl Loading checkpoint shards: 100%|████████████████████████████████████| 33/33 [00:13<00:00, 2.48it/s] Loading checkpoint shards: 100%|████████████████████████████████████| 33/33 [00:13<00:00, 2.45it/s] [2024-06-01 22:12:03,981] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed info: version=0.13.0, git-hash=unknown, git-branch=unknown [2024-06-01 22:12:15,085] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: False [2024-06-01 22:12:15,086] [INFO] [logging.py:96:log_dist] [Rank 0] Using client Optimizer as basic optimizer [2024-06-01 22:12:15,086] [INFO] [logging.py:96:log_dist] [Rank 0] Removing param_group that has no 'params' in the basic Optimizer [2024-06-01 22:12:15,087] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Basic Optimizer = Adam [2024-06-01 22:12:15,087] [INFO] [utils.py:56:is_zero_supported_optimizer] Checking ZeRO support for optimizer=Adam type=<class 'torch.optim.adam.Adam'> [2024-06-01 22:12:15,087] [INFO] [logging.py:96:log_dist] [Rank 0] Creating torch.bfloat16 ZeRO stage 2 optimizer [2024-06-01 22:12:15,087] [INFO] [stage_1_and_2.py:143:init] Reduce bucket size 200000000 [2024-06-01 22:12:15,087] [INFO] [stage_1_and_2.py:144:init] Allgather bucket size 200000000 [2024-06-01 22:12:15,087] [INFO] [stage_1_and_2.py:145:init] CPU Offload: False [2024-06-01 22:12:15,087] [INFO] [stage_1_and_2.py:146:init] Round robin gradient partitioning: False 0it [00:00, ?it/s][2024-06-01 22:12:15,837] [INFO] [utils.py:791:see_memory_usage] Before initializing optimizer states [2024-06-01 22:12:15,837] [INFO] [utils.py:792:see_memory_usage] MA 12.51 GB Max_MA 12.55 GB CA 12.55 GB Max_CA 13 GB [2024-06-01 22:12:15,838] [INFO] [utils.py:799:see_memory_usage] CPU Virtual Memory: used = 9.52 GB, percent = 1.9% [2024-06-01 22:12:15,934] [INFO] [utils.py:791:see_memory_usage] After initializing optimizer states [2024-06-01 22:12:15,935] [INFO] [utils.py:792:see_memory_usage] MA 12.68 GB Max_MA 12.94 GB CA 12.98 GB Max_CA 13 GB [2024-06-01 22:12:15,935] [INFO] [utils.py:799:see_memory_usage] CPU Virtual Memory: used = 9.52 GB, percent = 1.9% [2024-06-01 22:12:15,935] [INFO] [stage_1_and_2.py:533:init] optimizer state initialized [2024-06-01 22:12:16,027] [INFO] [utils.py:791:see_memory_usage] After initializing ZeRO optimizer [2024-06-01 22:12:16,027] [INFO] [utils.py:792:see_memory_usage] MA 12.68 GB Max_MA 12.68 GB CA 12.98 GB Max_CA 13 GB [2024-06-01 22:12:16,027] [INFO] [utils.py:799:see_memory_usage] CPU Virtual Memory: used = 9.52 GB, percent = 1.9% [2024-06-01 22:12:16,028] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Final Optimizer = Adam [2024-06-01 22:12:16,028] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed using client LR scheduler [2024-06-01 22:12:16,028] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed LR Scheduler = None [2024-06-01 22:12:16,028] [INFO] [logging.py:96:log_dist] [Rank 0] step=0, skipped=0, lr=[0.0003999999999999993], mom=[(0.95, 0.999)] [2024-06-01 22:12:16,028] [INFO] [config.py:984:print] DeepSpeedEngine configuration: [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] activation_checkpointing_config { "partition_activations": false, "contiguous_memory_optimization": false, "cpu_checkpointing": false, "number_checkpoints": null, "synchronize_checkpoint_boundary": false, "profile": false } [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True} [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] amp_enabled .................. False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] amp_params ................... False [2024-06-01 22:12:16,029] [INFO] [config.py:988: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 } [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] bfloat16_enabled ............. True [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] checkpoint_parallel_write_pipeline False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] checkpoint_tag_validation_enabled True [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] checkpoint_tag_validation_fail False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7bd3dbb9f6d0> [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] communication_data_type ...... None [2024-06-01 22:12:16,029] [INFO] [config.py:988: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}} [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] curriculum_enabled_legacy .... False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] curriculum_params_legacy ..... False [2024-06-01 22:12:16,029] [INFO] [config.py:988: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}}}} [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] data_efficiency_enabled ...... False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] dataloader_drop_last ......... False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] disable_allgather ............ False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] dump_state ................... False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] dynamic_loss_scale_args ...... None [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_enabled ........... False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_gas_boundary_resolution 1 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_layer_name ........ bert.encoder.layer [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_layer_num ......... 0 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_max_iter .......... 100 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_stability ......... 1e-06 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_tol ............... 0.01 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] eigenvalue_verbose ........... False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] elasticity_enabled ........... False [2024-06-01 22:12:16,029] [INFO] [config.py:988: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 } [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] fp16_auto_cast ............... None [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] fp16_enabled ................. False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] fp16_master_weights_and_gradients False [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] global_rank .................. 0 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] grad_accum_dtype ............. None [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] gradient_accumulation_steps .. 1 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] gradient_clipping ............ 0.0 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] gradient_predivide_factor .... 1.0 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] graph_harvesting ............. False [2024-06-01 22:12:16,029] [INFO] [config.py:988: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 [2024-06-01 22:12:16,029] [INFO] [config.py:988:print] initial_dynamic_scale ........ 1 [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] load_universal_checkpoint .... False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] loss_scale ................... 1.0 [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] memory_breakdown ............. False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] mics_hierarchial_params_gather False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] mics_shard_size .............. -1 [2024-06-01 22:12:16,030] [INFO] [config.py:988: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 [2024-06-01 22:12:16,030] [INFO] [config.py:988: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 } [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] optimizer_legacy_fusion ...... False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] optimizer_name ............... None [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] optimizer_params ............. None [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0, 'pipe_partitioned': True, 'grad_partitioned': True} [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] pld_enabled .................. False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] pld_params ................... False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] prescale_gradients ........... False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] scheduler_name ............... None [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] scheduler_params ............. None [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] seq_parallel_communication_data_type torch.float32 [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] sparse_attention ............. None [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] sparse_gradients_enabled ..... False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] steps_per_print .............. inf [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] train_batch_size ............. 48 [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] train_micro_batch_size_per_gpu 24 [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] use_data_before_expertparallel False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] use_node_local_storage ....... False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] wall_clock_breakdown ......... False [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] weight_quantization_config ... None [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] world_size ................... 2 [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] zero_allow_untested_optimizer True [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] zero_config .................. stage=2 contiguous_gradients=True reduce_scatter=True reduce_bucket_size=200000000 use_multi_rank_bucket_allreduce=True allgather_partitions=True allgather_bucket_size=200000000 overlap_comm=True load_from_fp32_weights=True elastic_checkpoint=False offload_param=None offload_optimizer=None sub_group_size=1000000000 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 zero_hpz_partition_size=1 zero_quantized_weights=False zero_quantized_nontrainable_weights=False zero_quantized_gradients=False mics_shard_size=-1 mics_hierarchical_params_gather=False memory_efficient_linear=True pipeline_loading_checkpoint=False override_module_apply=True [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] zero_enabled ................. True [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] zero_force_ds_cpu_optimizer .. True [2024-06-01 22:12:16,030] [INFO] [config.py:988:print] zero_optimization_stage ...... 2 [2024-06-01 22:12:16,030] [INFO] [config.py:974:print_user_config] json = { "bf16": { "enabled": true, "auto_cast": true }, "zero_optimization": { "stage": 2, "allgather_partitions": true, "allgather_bucket_size": 2.000000e+08, "overlap_comm": true, "reduce_scatter": true, "reduce_bucket_size": 2.000000e+08, "contiguous_gradients": true, "sub_group_size": 1.000000e+09 }, "gradient_accumulation_steps": 1, "train_batch_size": 48, "train_micro_batch_size_per_gpu": 24, "steps_per_print": inf, "wall_clock_breakdown": false, "fp16": { "enabled": false }, "zero_allow_untested_optimizer": true } 0it [00:00, ?it/s]../aten/src/ATen/native/cuda/Indexing.cu:1146: indexSelectLargeIndex: block: [183,0,0], thread: [64,0,0] Assertion srcIndex < srcSelectDimSize failed. ../aten/src/ATen/native/cuda/Indexing.cu:1146: indexSelectLargeIndex: block: [183,0,0], thread: [65,0,0] Assertion srcIndex < srcSelectDimSize failed. ... ../aten/src/ATen/native/cuda/Indexing.cu:1146: indexSelectLargeIndex: block: [221,0,0], thread: [127,0,0] Assertion srcIndex < srcSelectDimSize failed. 0it [00:00, ?it/s] ../aten/src/ATen/native/cuda/Indexing.cu:1146: indexSelectLargeIndex: block: [523,0,0], thread: [32,0,0] Assertion srcIndex < srcSelectDimSize failed. .... ../aten/src/ATen/native/cuda/Indexing.cu:1146: indexSelectLargeIndex: block: [523,0,0], thread: [63,0,0] Assertion srcIndex < srcSelectDimSize failed. 0it [00:00, ?it/s] Traceback (most recent call last): File "/media/lenovo/DATA/zth/Time-LLM-main/run_main.py", line 211, in outputs = model(batch_x, batch_x_mark, dec_inp, batch_y_mark) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 15, in wrapped_fn ret_val = func(*args, *kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/deepspeed/runtime/engine.py", line 1842, in forward loss = self.module(inputs, kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, kwargs) File "/media/lenovo/DATA/zth/Time-LLM-main/models/TimeLLM.py", line 197, in forward dec_out = self.forecast(x_enc, x_mark_enc, x_dec, x_mark_dec) File "/media/lenovo/DATA/zth/Time-LLM-main/models/TimeLLM.py", line 238, in forecast source_embeddings = self.mapping_layer(self.word_embeddings.permute(1, 0)).permute(1, 0) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, *kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/linear.py", line 114, in forward return F.linear(input, self.weight, self.bias) RuntimeError: CUDA error: CUBLAS_STATUS_NOT_INITIALIZED when calling cublasCreate(handle) Traceback (most recent call last): File "/media/lenovo/DATA/zth/Time-LLM-main/run_main.py", line 211, in outputs = model(batch_x, batch_x_mark, dec_inp, batch_y_mark) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(args, kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 15, in wrapped_fn ret_val = func(*args, kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/deepspeed/runtime/engine.py", line 1842, in forward loss = self.module(*inputs, *kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(args, kwargs) File "/media/lenovo/DATA/zth/Time-LLM-main/models/TimeLLM.py", line 197, in forward dec_out = self.forecast(x_enc, x_mark_enc, x_dec, x_mark_dec) File "/media/lenovo/DATA/zth/Time-LLM-main/models/TimeLLM.py", line 238, in forecast source_embeddings = self.mapping_layer(self.word_embeddings.permute(1, 0)).permute(1, 0) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/nn/modules/linear.py", line 114, in forward return F.linear(input, self.weight, self.bias) RuntimeError: CUDA error: CUBLAS_STATUS_NOT_INITIALIZED when calling cublasCreate(handle) terminate called after throwing an instance of 'c10::Error' what(): CUDA error: device-side assert triggered CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Exception raised from c10_cuda_check_implementation at ../c10/cuda/CUDAException.cpp:44 (most recent call first): frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x70cf3ddaf4d7 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libc10.so) frame #1: c10::detail::torchCheckFail(char const, char const, unsigned int, std::string const&) + 0x64 (0x70cf3dd7936b in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libc10.so) frame #2: c10::cuda::c10_cuda_check_implementation(int, char const, char const, int, bool) + 0x118 (0x70cf52d42b58 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libc10_cuda.so) frame #3: c10d::ProcessGroupNCCL::WorkNCCL::finishedGPUExecutionInternal() const + 0x80 (0x70cee4777450 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so) frame #4: c10d::ProcessGroupNCCL::WorkNCCL::isCompleted() + 0x58 (0x70cee477aa28 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so) frame #5: c10d::ProcessGroupNCCL::workCleanupLoop() + 0x227 (0x70cee477bf77 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so) frame #6: + 0xdc253 (0x70cf3d2dc253 in /lib/x86_64-linux-gnu/libstdc++.so.6) frame #7: + 0x94ac3 (0x70cf5d494ac3 in /lib/x86_64-linux-gnu/libc.so.6) frame #8: + 0x126850 (0x70cf5d526850 in /lib/x86_64-linux-gnu/libc.so.6)

terminate called after throwing an instance of 'c10::Error' what(): CUDA error: device-side assert triggered CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Exception raised from c10_cuda_check_implementation at ../c10/cuda/CUDAException.cpp:44 (most recent call first): frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7bd4877af4d7 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libc10.so) frame #1: c10::detail::torchCheckFail(char const, char const, unsigned int, std::string const&) + 0x64 (0x7bd48777936b in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libc10.so) frame #2: c10::cuda::c10_cuda_check_implementation(int, char const, char const, int, bool) + 0x118 (0x7bd491003b58 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libc10_cuda.so) frame #3: c10d::ProcessGroupNCCL::WorkNCCL::finishedGPUExecutionInternal() const + 0x80 (0x7bd419177450 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so) frame #4: c10d::ProcessGroupNCCL::WorkNCCL::isCompleted() + 0x58 (0x7bd41917aa28 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so) frame #5: c10d::ProcessGroupNCCL::workCleanupLoop() + 0x227 (0x7bd41917bf77 in /home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so) frame #6: + 0xdc253 (0x7bd471cdc253 in /lib/x86_64-linux-gnu/libstdc++.so.6) frame #7: + 0x94ac3 (0x7bd491e94ac3 in /lib/x86_64-linux-gnu/libc.so.6) frame #8: + 0x126850 (0x7bd491f26850 in /lib/x86_64-linux-gnu/libc.so.6) ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: -6) local_rank: 0 (pid: 5552) of binary: /home/lenovo/anaconda3/envs/timellm/bin/python Traceback (most recent call last): File "/home/lenovo/anaconda3/envs/timellm/bin/accelerate", line 8, in sys.exit(main()) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/accelerate/commands/accelerate_cli.py", line 45, in main args.func(args) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/accelerate/commands/launch.py", line 932, in launch_command multi_gpu_launcher(args) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/accelerate/commands/launch.py", line 627, in multi_gpu_launcher distrib_run.run(args) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/distributed/run.py", line 785, in run elastic_launch( File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 134, in call return launch_agent(self._config, self._entrypoint, list(args)) File "/home/lenovo/anaconda3/envs/timellm/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 250, in launch_agent raise ChildFailedError( torch.distributed.elastic.multiprocessing.errors.ChildFailedError:

run_main.py FAILED

Failures: [1]: time : 2024-06-01_22:12:17 host : lenovo-ThinkStation-P7 rank : 1 (local_rank: 1) exitcode : -6 (pid: 5553) error_file: <N/A> traceback : Signal 6 (SIGABRT) received by PID 5553

Root Cause (first observed failure): [0]: time : 2024-06-01_22:12:17 host : lenovo-ThinkStation-P7 rank : 0 (local_rank: 0) exitcode : -6 (pid: 5552) error_file: <N/A> traceback : Signal 6 (SIGABRT) received by PID 5552

zhangtianhong-1998 commented 4 months ago

我的cuda是11.7

zhangtianhong-1998 commented 4 months ago

尝试降低python至3.8.5仍然出现类似错误

zhangtianhong-1998 commented 4 months ago

1717333321595 I used custom data during Debug to find the above error appearing in LLama embeddings，but the issue is still unresolved （我使用自定义数据在Debug过程中发现上述错误出现在LLama embeddings，但是问题仍未解决）

zhangtianhong-1998 commented 4 months ago

code: prompt_embeddings = self.llm_model.get_input_embeddings()(prompt.to(x_enc.device)) # (batch, prompt_token, dim)

When I switched the custom data back to ETTh2, everything seemed normal, but once I used the custom data, this unknown error occurred, which I thought was amazing. （当我将自定义数据切换回ETTh2时，一切趋于正常，然而一但使用自定义的数据，变发生了这个未知的错误，我觉得很神奇） Currently I am using weights from modelscope, this weight shows the above error, I am switching to author recommended weights for further testing (目前我是用的权重来自于modelscope，在这个llama7b的权重显示出上述错误，我正在切换到作者推荐权重进行进一步测试)

这仅仅是语言模型的嵌入编码，我在ETTh2和自定义数据中测试发现，最大token id 为32000（使用了[PAD]），最小token id 为 0 但是不会出现越界的问题，由于词表为32000，tokenid应该位于0~31999之间，多余的[PAD]带来了额外的32000，所以我能理解越界的问题，然而这个问题却只出现在了自定义的数据，这是令人疑惑的

zhangtianhong-1998 commented 4 months ago

这个越界问题确实源于不正确的权重配置，[PAD]特殊token带来了额外的token id ，但是然而这个问题却只出现在了自定义的数据，这仍然无法解释

kwuking commented 4 months ago

你好，请为你的自定义数据集是什么样子的呢？目前出现的问题是使用自定义数据集会出现token id越界的现象吗？如果尝试将[PAD]去掉是否可以解决你的问题呢？

zhangtianhong-1998 commented 4 months ago

你好，请为你的自定义数据集是什么样子的呢？目前出现的问题是使用自定义数据集会出现token id越界的现象吗？如果尝试将[PAD]去掉是否可以解决你的问题呢？

自定义数据就是一般的多元时序数据，目前出现的问题是，我从modelscope下载的权重缺少eostoken，然后导致自定义数据失败，但是这个在ETTh2上却不会出现问题，然后我目前使用了您项目中的权重地址下载了相关权重，发现可以正常训练了。我预计是由于越界，我同时打印了最小和最大的token id 发现最小是0，最大是32000（这样就有32001个token id），词表只有32000，但是如果越界应该同时发生，然而ETTh2上却没有出现，所以我比较困惑。删去[PAD] token我目前尚未尝试这一步骤，我将会在后续实验中尝试，进一步反馈