从Checkpoint继续微调

CNUIGB commented 5 months ago

System Info / 系統信息

微信截图_20240419122803 微信截图_20240419122752 微信截图_20240419122736 微信截图_20240419115040 从Checkpoint继续微调报错，使用的LoRA微调，但是无法从Checkpoint继续

Who can help? / 谁可以帮助到您？

No response

Information / 问题信息

[ ] The official example scripts / 官方的示例脚本
[ ] My own modified scripts / 我自己修改的脚本和任务

Reproduction / 复现过程

如上图

Expected behavior / 期待表现

想问一下是为什么，怎么解决

qingdengyue commented 5 months ago

训练的TrainingArguments 上有一个 output_dir,运行了之后看对应目录下有没 checkpoint的文件。有的话就可以用 resume_from_checkpoint了。以及如果还有问题的话。能够贴下代码。感谢 :smile:

CNUIGB commented 5 months ago

训练的TrainingArguments 上有一个 output_dir,运行了之后看对应目录下有没 checkpoint的文件。有的话就可以用 resume_from_checkpoint了。以及如果还有问题的话。能够贴下代码。感谢 😄

微信截图_20240419162756 是有checkpoint文件的，代码的话就是官方的那个命令： OMP_NUM_THREADS=1 torchrun --standalone --nnodes=1 --nproc_per_node=6 /home/liuyawei/llm/ChatGLM3/finetune_demo/finetune_hf.py /home/liuyawei/llm/ChatGLM3/finetune_demo/data /home/liuyawei/model/chatglm3_v2/chatglm3-6b /home/liuyawei/llm/ChatGLM3/finetune_demo/configs/lora.yaml 1000

我是想从checkpoint-1000继续，但是resume_from_checkpoint不管是写数字1000还是yes，都会报这个错

CNUIGB commented 5 months ago

训练的TrainingArguments 上有一个 output_dir,运行了之后看对应目录下有没 checkpoint的文件。有的话就可以用 resume_from_checkpoint了。以及如果还有问题的话。能够贴下代码。感谢 😄

微信截图_20240419163130 补充一下lora的配置文件的内容

CNUIGB commented 5 months ago

训练的TrainingArguments 上有一个 output_dir,运行了之后看对应目录下有没 checkpoint的文件。有的话就可以用 resume_from_checkpoint了。以及如果还有问题的话。能够贴下代码。感谢 😄

还有一点就是requirements.txt中的mpi4py那个依赖包，我今天运行继续训练的命令时提示我没有这个module，所以我在网上下了一个，但是版本是3.1.6，官方是>=3.1.5 不知道这个会不会有影响

zRzRzRzRzRzRzR commented 5 months ago

训练的TrainingArguments 上有一个 output_dir,运行了之后看对应目录下有没 checkpoint的文件。有的话就可以用 resume_from_checkpoint了。以及如果还有问题的话。能够贴下代码。感谢 😄

还有一点就是requirements.txt中的mpi4py那个依赖包，我今天运行继续训练的命令时提示我没有这个module，所以我在网上下了一个，但是版本是3.1.6，官方是>=3.1.5 不知道这个会不会有影响

没有影响

qingdengyue commented 5 months ago

训练的TrainingArguments 上有一个 output_dir,运行了之后看对应目录下有没 checkpoint的文件。有的话就可以用 resume_from_checkpoint了。以及如果还有问题的话。能够贴下代码。感谢 😄

是有checkpoint文件的，代码的话就是官方的那个命令： OMP_NUM_THREADS=1 torchrun --standalone --nnodes=1 --nproc_per_node=6 /home/liuyawei/llm/ChatGLM3/finetune_demo/finetune_hf.py /home/liuyawei/llm/ChatGLM3/finetune_demo/data /home/liuyawei/model/chatglm3_v2/chatglm3-6b /home/liuyawei/llm/ChatGLM3/finetune_demo/configs/lora.yaml 1000

我是想从checkpoint-1000继续，但是resume_from_checkpoint不管是写数字1000还是yes，都会报这个错

启动日志有没？控制台输出的那个

CNUIGB commented 5 months ago

训练的TrainingArguments 上有一个 output_dir,运行了之后看对应目录下有没 checkpoint的文件。有的话就可以用 resume_from_checkpoint了。以及如果还有问题的话。能够贴下代码。感谢 😄

是有checkpoint文件的，代码的话就是官方的那个命令： OMP_NUM_THREADS=1 torchrun --standalone --nnodes=1 --nproc_per_node=6 /home/liuyawei/llm/ChatGLM3/finetune_demo/finetune_hf.py /home/liuyawei/llm/ChatGLM3/finetune_demo/data /home/liuyawei/model/chatglm3_v2/chatglm3-6b /home/liuyawei/llm/ChatGLM3/finetune_demo/configs/lora.yaml 1000 我是想从checkpoint-1000继续，但是resume_from_checkpoint不管是写数字1000还是yes，都会报这个错

启动日志有没？控制台输出的那个

有的有的，我给您贴一下：

master_addr is only used for static rdzv_backend and when rdzv_endpoint is not specified. Loading checkpoint shards: 100%|██████████████████| 7/7 [00:06<00:00, 1.05it/s] Loading checkpoint shards: 100%|██████████████████| 7/7 [00:06<00:00, 1.04it/s] Loading checkpoint shards: 100%|██████████████████| 7/7 [00:06<00:00, 1.14it/s] Loading checkpoint shards: 100%|██████████████████| 7/7 [00:06<00:00, 1.04it/s] Loading checkpoint shards: 100%|██████████████████| 7/7 [00:06<00:00, 1.04it/s] trainable params: 1,949,696 || all params: 6,245,533,696 || trainable%: 0.031217444255383614 --> Model

--> model has 1.949696M params

trainable params: 1,949,696 || all params: 6,245,533,696 || trainable%: 0.031217444255383614 --> Model trainable params: 1,949,696 || all params: 6,245,533,696 || trainable%: 0.031217444255383614 --> Model

--> model has 1.949696M params

trainable params: 1,949,696 || all params: 6,245,533,696 || trainable%: 0.031217444255383614 --> Model trainable params: 1,949,696 || all params: 6,245,533,696 || trainable%: 0.031217444255383614 --> Model

--> model has 1.949696M params

Loading checkpoint shards: 100%|██████████████████| 7/7 [00:06<00:00, 1.02it/s] trainable params: 1,949,696 || all params: 6,245,533,696 || trainable%: 0.031217444255383614 --> Model

--> model has 1.949696M params

train_dataset: Dataset({ features: ['input_ids', 'labels'], num_rows: 241 }) val_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) train_dataset: Dataset({ features: ['input_ids', 'labels'], num_rows: 241 }) test_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) resume checkpoint from checkpoint-1000 val_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) train_dataset: Dataset({ features: ['input_ids', 'labels'], num_rows: 241 }) test_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) resume checkpoint from checkpoint-1000 val_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) test_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) resume checkpoint from checkpoint-1000 train_dataset: Dataset({ features: ['input_ids', 'labels'], num_rows: 241 }) val_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) train_dataset: Dataset({ features: ['input_ids', 'labels'], num_rows: 241 }) test_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) resume checkpoint from checkpoint-1000 val_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) test_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) max_steps is given, it will override any value given in num_train_epochs resume checkpoint from checkpoint-1000 train_dataset: Dataset({ features: ['input_ids', 'labels'], num_rows: 241 }) val_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) test_dataset: Dataset({ features: ['input_ids', 'output_ids'], num_rows: 95 }) resume checkpoint from checkpoint-1000 [2024-04-19 16:53:40,505] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-04-19 16:53:40,574] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-04-19 16:53:40,618] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-04-19 16:53:40,782] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-04-19 16:53:40,820] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-04-19 16:53:40,957] [INFO] [real_accelerator.py:191:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-04-19 16:53:43,158] [INFO] [logging.py:96:log_dist] [Rank -1] DeepSpeed info: version=0.13.2, git-hash=unknown, git-branch=unknown [2024-04-19 16:53:43,158] [INFO] [comm.py:637:init_distributed] cdb=None [2024-04-19 16:53:50,096] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Flops Profiler Enabled: False [2024-04-19 16:53:50,098] [INFO] [logging.py:96:log_dist] [Rank 0] Using client Optimizer as basic optimizer [2024-04-19 16:53:50,098] [INFO] [logging.py:96:log_dist] [Rank 0] Removing param_group that has no 'params' in the basic Optimizer [2024-04-19 16:53:50,101] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Basic Optimizer = AdamW [2024-04-19 16:53:50,101] [INFO] [utils.py:56:is_zero_supported_optimizer] Checking ZeRO support for optimizer=AdamW type=<class 'torch.optim.adamw.AdamW'> [2024-04-19 16:53:50,101] [INFO] [logging.py:96:log_dist] [Rank 0] Creating torch.float32 ZeRO stage 2 optimizer [2024-04-19 16:53:50,101] [INFO] [stage_1_and_2.py:149:init] Reduce bucket size 500000000 [2024-04-19 16:53:50,101] [INFO] [stage_1_and_2.py:150:init] Allgather bucket size 500000000 [2024-04-19 16:53:50,101] [INFO] [stage_1_and_2.py:151:init] CPU Offload: False [2024-04-19 16:53:50,101] [INFO] [stage_1_and_2.py:152:init] Round robin gradient partitioning: False [2024-04-19 16:53:50,808] [INFO] [utils.py:800:see_memory_usage] Before initializing optimizer states [2024-04-19 16:53:50,809] [INFO] [utils.py:801:see_memory_usage] MA 11.67 GB Max_MA 11.67 GB CA 11.68 GB Max_CA 12 GB [2024-04-19 16:53:50,809] [INFO] [utils.py:808:see_memory_usage] CPU Virtual Memory: used = 14.79 GB, percent = 5.9% [2024-04-19 16:53:50,904] [INFO] [utils.py:800:see_memory_usage] After initializing optimizer states [2024-04-19 16:53:50,904] [INFO] [utils.py:801:see_memory_usage] MA 11.67 GB Max_MA 11.67 GB CA 11.68 GB Max_CA 12 GB [2024-04-19 16:53:50,904] [INFO] [utils.py:808:see_memory_usage] CPU Virtual Memory: used = 14.8 GB, percent = 5.9% [2024-04-19 16:53:50,905] [INFO] [stage_1_and_2.py:539:init] optimizer state initialized [2024-04-19 16:53:50,997] [INFO] [utils.py:800:see_memory_usage] After initializing ZeRO optimizer [2024-04-19 16:53:50,998] [INFO] [utils.py:801:see_memory_usage] MA 11.67 GB Max_MA 11.67 GB CA 11.68 GB Max_CA 12 GB [2024-04-19 16:53:50,998] [INFO] [utils.py:808:see_memory_usage] CPU Virtual Memory: used = 14.8 GB, percent = 5.9% [2024-04-19 16:53:50,999] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed Final Optimizer = AdamW [2024-04-19 16:53:50,999] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed using client LR scheduler [2024-04-19 16:53:50,999] [INFO] [logging.py:96:log_dist] [Rank 0] DeepSpeed LR Scheduler = None [2024-04-19 16:53:50,999] [INFO] [logging.py:96:log_dist] [Rank 0] step=0, skipped=0, lr=[5e-05], mom=[(0.9, 0.999)] [2024-04-19 16:53:51,000] [INFO] [config.py:987:print] DeepSpeedEngine configuration: [2024-04-19 16:53:51,000] [INFO] [config.py:991:print] activation_checkpointing_config { "partition_activations": false, "contiguous_memory_optimization": false, "cpu_checkpointing": false, "number_checkpoints": null, "synchronize_checkpoint_boundary": false, "profile": false } [2024-04-19 16:53:51,000] [INFO] [config.py:991:print] aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True} [2024-04-19 16:53:51,000] [INFO] [config.py:991:print] amp_enabled .................. False [2024-04-19 16:53:51,000] [INFO] [config.py:991:print] amp_params ................... False [2024-04-19 16:53:51,000] [INFO] [config.py:991: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 } [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] bfloat16_enabled ............. False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] checkpoint_parallel_write_pipeline False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] checkpoint_tag_validation_enabled True [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] checkpoint_tag_validation_fail False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] comms_config ................. <deepspeed.comm.config.DeepSpeedCommsConfig object at 0x7f8505a38b50> [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] communication_data_type ...... None [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] compile_config ............... enabled=False backend='inductor' kwargs={} [2024-04-19 16:53:51,001] [INFO] [config.py:991: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}} [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] curriculum_enabled_legacy .... False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] curriculum_params_legacy ..... False [2024-04-19 16:53:51,001] [INFO] [config.py:991: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}}}} [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] data_efficiency_enabled ...... False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] dataloader_drop_last ......... False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] disable_allgather ............ False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] dump_state ................... False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] dynamic_loss_scale_args ...... None [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_enabled ........... False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_gas_boundary_resolution 1 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_layer_name ........ bert.encoder.layer [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_layer_num ......... 0 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_max_iter .......... 100 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_stability ......... 1e-06 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_tol ............... 0.01 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] eigenvalue_verbose ........... False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] elasticity_enabled ........... False [2024-04-19 16:53:51,001] [INFO] [config.py:991: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 } [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] fp16_auto_cast ............... None [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] fp16_enabled ................. False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] fp16_master_weights_and_gradients False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] global_rank .................. 0 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] grad_accum_dtype ............. None [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] gradient_accumulation_steps .. 1 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] gradient_clipping ............ 1.0 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] gradient_predivide_factor .... 1.0 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] graph_harvesting ............. False [2024-04-19 16:53:51,001] [INFO] [config.py:991: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 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] initial_dynamic_scale ........ 65536 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] load_universal_checkpoint .... False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] loss_scale ................... 0 [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] memory_breakdown ............. False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] mics_hierarchial_params_gather False [2024-04-19 16:53:51,001] [INFO] [config.py:991:print] mics_shard_size .............. -1 [2024-04-19 16:53:51,002] [INFO] [config.py:991: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 [2024-04-19 16:53:51,002] [INFO] [config.py:991: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 } [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] optimizer_legacy_fusion ...... False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] optimizer_name ............... None [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] optimizer_params ............. None [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0, 'pipe_partitioned': True, 'grad_partitioned': True} [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] pld_enabled .................. False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] pld_params ................... False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] prescale_gradients ........... False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] scheduler_name ............... None [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] scheduler_params ............. None [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] seq_parallel_communication_data_type torch.float32 [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] sparse_attention ............. None [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] sparse_gradients_enabled ..... False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] steps_per_print .............. inf [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] train_batch_size ............. 6 [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] train_micro_batch_size_per_gpu 1 [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] use_data_before_expertparallel False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] use_node_local_storage ....... False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] wall_clock_breakdown ......... False [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] weight_quantization_config ... None [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] world_size ................... 6 [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] zero_allow_untested_optimizer True [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] zero_config .................. stage=2 contiguous_gradients=True reduce_scatter=True reduce_bucket_size=500000000 use_multi_rank_bucket_allreduce=True allgather_partitions=True allgather_bucket_size=500000000 overlap_comm=True load_from_fp32_weights=True elastic_checkpoint=False offload_param=None 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=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 [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] zero_enabled ................. True [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] zero_force_ds_cpu_optimizer .. True [2024-04-19 16:53:51,002] [INFO] [config.py:991:print] zero_optimization_stage ...... 2 [2024-04-19 16:53:51,002] [INFO] [config.py:977:print_user_config] json = { "fp16": { "enabled": false, "loss_scale": 0, "loss_scale_window": 1000, "initial_scale_power": 16, "hysteresis": 2, "min_loss_scale": 1 }, "bf16": { "enabled": false }, "zero_optimization": { "stage": 2, "allgather_partitions": true, "allgather_bucket_size": 5.000000e+08, "overlap_comm": true, "reduce_scatter": true, "reduce_bucket_size": 5.000000e+08, "contiguous_gradients": true }, "gradient_accumulation_steps": 1, "gradient_clipping": 1.0, "steps_per_print": inf, "train_batch_size": 6, "train_micro_batch_size_per_gpu": 1, "wall_clock_breakdown": false, "zero_allow_untested_optimizer": true } Attempting to resume from /home/liuyawei/model/finetune_lora/web_continue/checkpoint-1000 [2024-04-19 16:53:51,003] [INFO] [torch_checkpoint_engine.py:27:load] [Torch] Loading checkpoint from /home/liuyawei/model/finetune_lora/web_continue/checkpoint-1000/global_step1000/mp_rank_00_model_states.pt... [2024-04-19 16:53:51,009] [INFO] [torch_checkpoint_engine.py:29:load] [Torch] Loaded checkpoint from /home/liuyawei/model/finetune_lora/web_continue/checkpoint-1000/global_step1000/mp_rank_00_model_states.pt. [2024-04-19 16:53:51,009] [INFO] [torch_checkpoint_engine.py:27:load] [Torch] Loading checkpoint from /home/liuyawei/model/finetune_lora/web_continue/checkpoint-1000/global_step1000/mp_rank_00_model_states.pt... [2024-04-19 16:53:51,012] [INFO] [torch_checkpoint_engine.py:29:load] [Torch] Loaded checkpoint from /home/liuyawei/model/finetune_lora/web_continue/checkpoint-1000/global_step1000/mp_rank_00_model_states.pt. ╭───────────────────── Traceback (most recent call last) ──────────────────────╮ │ /home/liuyawei/llm/ChatGLM3/finetune_demo/finetune_hf.py:547 in main │ │ │ │ 544 │ │ │ │ model.enable_input_require_grads() │ │ 545 │ │ │ │ checkpointdir = output_dir + "/checkpoint-" + str(chec │ │ 546 │ │ │ │ print("resume checkpoint from checkpoint-" + str(chec │ │ ❱ 547 │ │ │ │ trainer.train(resume_from_checkpoint=checkpointdir) │ │ 548 │ │ │ else: │ │ 549 │ │ │ │ # If not, start from scratch │ │ 550 │ │ │ │ trainer.train() │ │ │ │ /home/liuyawei/.local/lib/python3.10/site-packages/transformers/trainer.py:1 │ │ 539 in train │ │ │ │ 1536 │ │ │ finally: │ │ 1537 │ │ │ │ hf_hub_utils.enable_progress_bars() │ │ 1538 │ │ else: │ │ ❱ 1539 │ │ │ return inner_training_loop( │ │ 1540 │ │ │ │ args=args, │ │ 1541 │ │ │ │ resume_from_checkpoint=resume_from_checkpoint, │ │ 1542 │ │ │ │ trial=trial, │ │ │ │ /home/liuyawei/.local/lib/python3.10/site-packages/transformers/trainer.py:1 │ │ 708 in _inner_training_loop │ │ │ │ 1705 │ │ # ckpt loading │ │ 1706 │ │ if resume_from_checkpoint is not None: │ │ 1707 │ │ │ if self.is_deepspeed_enabled: │ │ ❱ 1708 │ │ │ │ deepspeed_load_checkpoint(self.modelwrapped, resume │ │ 1709 │ │ │ elif is_sagemaker_mp_enabled() or self.is_fsdp_enabled: │ │ 1710 │ │ │ │ self._load_from_checkpoint(resume_from_checkpoint, se │ │ 1711 │ │ │ │ /home/liuyawei/.local/lib/python3.10/site-packages/transformers/integrations │ │ /deepspeed.py:402 in deepspeed_load_checkpoint │ │ │ │ 399 │ if len(deepspeed_checkpoint_dirs) > 0: │ │ 400 │ │ logger.info(f"Attempting to resume from {checkpoint_path}") │ │ 401 │ │ # this magically updates self.optimizer and self.lr_scheduler │ │ ❱ 402 │ │ loadpath, = deepspeed_engine.load_checkpoint( │ │ 403 │ │ │ checkpoint_path, load_optimizer_states=True, load_lr_sched │ │ 404 │ │ ) │ │ 405 │ │ if load_path is None: │ │ │ │ /home/liuyawei/.local/lib/python3.10/site-packages/deepspeed/runtime/engine. │ │ py:2750 in load_checkpoint │ │ │ │ 2747 │ │ │ # Prepare for checkpoint load by ensuring all parameters │ │ 2748 │ │ │ self.optimizer.checkpoint_event_prologue() │ │ 2749 │ │ │ │ ❱ 2750 │ │ load_path, client_states = self._load_checkpoint(load_dir, │ │ 2751 │ │ │ │ │ │ │ │ │ │ │ │ │ │ tag, │ │ 2752 │ │ │ │ │ │ │ │ │ │ │ │ │ │ loadmodule │ │ 2753 │ │ │ │ │ │ │ │ │ │ │ │ │ │ load_optimiz │ │ │ │ /home/liuyawei/.local/lib/python3.10/site-packages/deepspeed/runtime/engine. │ │ py:2835 in _load_checkpoint │ │ │ │ 2832 │ │ │ │ │ │ │ │ │ │ │ │ numexperts=self.num │ │ 2833 │ │ │ │ │ │ │ │ │ │ │ │ checkpoint_engine=sel │ │ 2834 │ │ if not self.load_universal_checkpoint(): │ │ ❱ 2835 │ │ │ self.load_module_state_dict(checkpoint=checkpoint, │ │ 2836 │ │ │ │ │ │ │ │ │ │ strict=load_module_strict, │ │ 2837 │ │ │ │ │ │ │ │ │ │ custom_load_fn=custom_load_fn │ │ 2838 │ │ │ │ │ │ │ │ │ │ fetch_z3_params=fetch_z3_para │ │ │ │ /home/liuyawei/.local/lib/python3.10/site-packages/deepspeed/runtime/engine. │ │ py:2613 in load_module_state_dict │ │ │ │ 2610 │ │ │ if custom_load_fn: │ │ 2611 │ │ │ │ custom_load_fn(src=module_state_dict, dst=self.module │ │ 2612 │ │ │ else: │ │ ❱ 2613 │ │ │ │ self.module.load_state_dict( │ │ 2614 │ │ │ │ │ module_state_dict, # TODO │ │ 2615 │ │ │ │ │ strict=strict) │ │ 2616 │ │ │ │ /home/liuyawei/.local/lib/python3.10/site-packages/torch/nn/modules/module.p │ │ y:2041 in load_state_dict │ │ │ │ 2038 │ │ │ │ │ │ ', '.join('"{}"'.format(k) for k in missing_k │ │ 2039 │ │ │ │ 2040 │ │ if len(error_msgs) > 0: │ │ ❱ 2041 │ │ │ raise RuntimeError('Error(s) in loading state_dict for {} │ │ 2042 │ │ │ │ │ │ │ self.class.name, "\n\t".join(e │ │ 2043 │ │ return _IncompatibleKeys(missing_keys, unexpected_keys) │ │ 2044 │ ╰──────────────────────────────────────────────────────────────────────────────╯ RuntimeError: Error(s) in loading state_dict for PeftModelForCausalLM: Missing key(s) in state_dict: "base_model.model.transformer.embedding.word_embeddings.weight", "base_model.model.transformer.encoder.layers.0.input_layernorm.weight", "base_model.model.transformer.encoder.layers.0.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.0.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.0.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.0.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.0.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.0.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.1.input_layernorm.weight", "base_model.model.transformer.encoder.layers.1.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.1.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.1.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.1.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.1.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.1.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.2.input_layernorm.weight", "base_model.model.transformer.encoder.layers.2.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.2.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.2.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.2.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.2.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.2.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.3.input_layernorm.weight", "base_model.model.transformer.encoder.layers.3.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.3.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.3.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.3.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.3.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.3.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.4.input_layernorm.weight", "base_model.model.transformer.encoder.layers.4.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.4.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.4.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.4.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.4.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.4.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.5.input_layernorm.weight", "base_model.model.transformer.encoder.layers.5.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.5.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.5.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.5.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.5.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.5.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.6.input_layernorm.weight", "base_model.model.transformer.encoder.layers.6.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.6.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.6.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.6.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.6.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.6.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.7.input_layernorm.weight", "base_model.model.transformer.encoder.layers.7.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.7.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.7.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.7.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.7.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.7.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.8.input_layernorm.weight", "base_model.model.transformer.encoder.layers.8.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.8.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.8.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.8.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.8.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.8.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.9.input_layernorm.weight", "base_model.model.transformer.encoder.layers.9.self_attention.query_key_value.ba se_layer.weight", "base_model.model.transformer.encoder.layers.9.self_attention.query_key_value.ba se_layer.bias", "base_model.model.transformer.encoder.layers.9.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.9.post_attention_layernorm.weight", "base_model.model.transformer.encoder.layers.9.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.9.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.10.input_layernorm.weight", "base_model.model.transformer.encoder.layers.10.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.10.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.10.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.10.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.10.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.10.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.11.input_layernorm.weight", "base_model.model.transformer.encoder.layers.11.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.11.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.11.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.11.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.11.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.11.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.12.input_layernorm.weight", "base_model.model.transformer.encoder.layers.12.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.12.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.12.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.12.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.12.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.12.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.13.input_layernorm.weight", "base_model.model.transformer.encoder.layers.13.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.13.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.13.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.13.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.13.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.13.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.14.input_layernorm.weight", "base_model.model.transformer.encoder.layers.14.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.14.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.14.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.14.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.14.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.14.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.15.input_layernorm.weight", "base_model.model.transformer.encoder.layers.15.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.15.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.15.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.15.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.15.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.15.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.16.input_layernorm.weight", "base_model.model.transformer.encoder.layers.16.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.16.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.16.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.16.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.16.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.16.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.17.input_layernorm.weight", "base_model.model.transformer.encoder.layers.17.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.17.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.17.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.17.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.17.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.17.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.18.input_layernorm.weight", "base_model.model.transformer.encoder.layers.18.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.18.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.18.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.18.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.18.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.18.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.19.input_layernorm.weight", "base_model.model.transformer.encoder.layers.19.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.19.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.19.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.19.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.19.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.19.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.20.input_layernorm.weight", "base_model.model.transformer.encoder.layers.20.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.20.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.20.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.20.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.20.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.20.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.21.input_layernorm.weight", "base_model.model.transformer.encoder.layers.21.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.21.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.21.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.21.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.21.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.21.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.22.input_layernorm.weight", "base_model.model.transformer.encoder.layers.22.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.22.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.22.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.22.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.22.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.22.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.23.input_layernorm.weight", "base_model.model.transformer.encoder.layers.23.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.23.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.23.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.23.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.23.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.23.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.24.input_layernorm.weight", "base_model.model.transformer.encoder.layers.24.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.24.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.24.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.24.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.24.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.24.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.25.input_layernorm.weight", "base_model.model.transformer.encoder.layers.25.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.25.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.25.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.25.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.25.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.25.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.26.input_layernorm.weight", "base_model.model.transformer.encoder.layers.26.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.26.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.26.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.26.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.26.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.26.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.layers.27.input_layernorm.weight", "base_model.model.transformer.encoder.layers.27.self_attention.query_key_value.b ase_layer.weight", "base_model.model.transformer.encoder.layers.27.self_attention.query_key_value.b ase_layer.bias", "base_model.model.transformer.encoder.layers.27.self_attention.dense.weight", "base_model.model.transformer.encoder.layers.27.post_attention_layernorm.weight" , "base_model.model.transformer.encoder.layers.27.mlp.dense_h_to_4h.weight", "base_model.model.transformer.encoder.layers.27.mlp.dense_4h_to_h.weight", "base_model.model.transformer.encoder.final_layernorm.weight", "base_model.model.transformer.output_layer.weight".

qingdengyue commented 4 months ago

--nproc_per_node=6 中的 6改成1 试下呢?

CNUIGB commented 4 months ago

--nproc_per_node=6 中的 6改成1 试下呢?

刚刚试了一下，还是一样的错误信息，我看错误是Missing key(s) in state_dict，想问一下是不是哪里加载错了训练后的文件夹中adapter_config.json里有base model的路径，但是trainer_state.json里没有

想问一下checkpoint的结构是这样吗微信截图_20240422142814

zRzRzRzRzRzRzR commented 4 months ago

ptuing是，更多微调问题可以在社区咨询 #253

THUDM / ChatGLM3