华为计算微信公众号
昇腾AI开发者公众号
华为计算微博
华为计算今日头条
使用ms-swift框架Lora微调Qwen2.5-7B-Instruct模型,报下面的错误:
[rank1]: Traceback (most recent call last): [rank1]: File "/data1/guihonghao/ms-swift/swift/cli/sft.py", line 5, in <module> [rank1]: sft_main() [rank1]: File "/data1/guihonghao/ms-swift/swift/utils/run_utils.py", line 22, in x_main [rank1]: args, remaining_argv = parse_args(args_class, argv) [rank1]: File "/data1/guihonghao/ms-swift/swift/utils/utils.py", line 131, in parse_args [rank1]: args, remaining_args = parser.parse_args_into_dataclasses(argv, return_remaining_strings=True) [rank1]: File "/home/guihonghao/.local/lib/python3.10/site-packages/transformers/hf_argparser.py", line 352, in parse_args_into_dataclasses [rank1]: obj = dtype(**inputs) [rank1]: File "<string>", line 215, in __init__ [rank1]: File "/data1/guihonghao/ms-swift/swift/llm/utils/argument.py", line 1165, in __post_init__ [rank1]: self.output_dir = add_version_to_work_dir(self.output_dir) [rank1]: File "/data1/guihonghao/ms-swift/swift/utils/utils.py", line 113, in add_version_to_work_dir [rank1]: sub_folder = broadcast_string(sub_folder) [rank1]: File "/data1/guihonghao/ms-swift/swift/utils/torch_utils.py", line 211, in broadcast_string [rank1]: tensor = torch.zeros(buffer_size, dtype=torch.int64, device=device) [rank1]: RuntimeError: map:build/CMakeFiles/torch_npu.dir/compiler_depend.ts:373 NPU function error: aclrtMallocPhysical, error code is 507899 [rank1]: [ERROR] 2024-11-27-00:33:19 (PID:2941779, Device:1, RankID:1) ERR00100 PTA call acl api failed [rank1]: [Error]: An internal error occurs in the Driver module. [rank1]: Rectify the fault based on the error information in the ascend log. [rank1]: EL9999: Inner Error! [rank1]: EL9999 [drv api]halMemCreate failed. drvRet=17.[FUNC:MallocPhysical][FILE:npu_driver.cc][LINE:5439] [rank1]: TraceBack (most recent call last): [rank1]: rtMallocPhysical execute failed, reason=[driver error:internal error][FUNC:FuncErrorReason][FILE:error_message_manage.cc][LINE:50] [rank1]: malloc physical memory failed, runtime result = 507899[FUNC:ReportCallError][FILE:log_inner.cpp][LINE:161]
运行脚本:
source /usr/local/Ascend/ascend-toolkit/set_env.sh export MODELSCOPE_CACHE='/data1/guihonghao/.cache/modelscope' export HF_DATASETS_CACHE='/data1/guihonghao/.cache/huggingface' export NPROC_PER_NODE=4 export ASCEND_RT_VISIBLE_DEVICES=2,3,4,5 export ASCEND_LAUNCH_BLOCKING=1 swift sft \ --model_type 'qwen2_5-7b-instruct' \ --model_id_or_path /data1/guihonghao/models/Qwen2.5-7B-Instruct \ --sft_type 'lora' \ --tuner_backend 'peft' \ --template_type 'AUTO' \ --dtype 'bf16' \ --output_dir /data1/guihonghao/lora_outputs/qwen2_5-7b-instruct-ie-v3 \ --custom_train_dataset_path /data2/guihonghao/data/train_data/IE-v1/res/train.jsonl \ --custom_val_dataset_path /data2/guihonghao/data/train_data/IE-v1/res/dev.jsonl \ --num_train_epochs 5 \ --max_length 2048 \ --check_dataset_strategy 'warning' \ --lora_rank 64 \ --lora_alpha 64 \ --lora_dropout_p 0.05 \ --lora_target_modules 'DEFAULT' \ --gradient_checkpointing true \ --batch_size 6 \ --eval_batch_size 2 \ --weight_decay 0.1 \ --learning_rate '5e-5' \ --gradient_accumulation_steps 4 \ --max_grad_norm 0.5 \ --warmup_ratio 0.03 \ --eval_strategy 'no' \ --save_strategy 'epoch' \ --save_total_limit 10 \ --logging_steps 10 \ --preprocess_num_proc 32 \ --lazy_tokenize True \ --deepspeed 'default-zero2'
我要发帖子
使用ms-swift框架Lora微调Qwen2.5-7B-Instruct模型,报下面的错误:
运行脚本:
source /usr/local/Ascend/ascend-toolkit/set_env.sh export MODELSCOPE_CACHE='/data1/guihonghao/.cache/modelscope' export HF_DATASETS_CACHE='/data1/guihonghao/.cache/huggingface' export NPROC_PER_NODE=4 export ASCEND_RT_VISIBLE_DEVICES=2,3,4,5 export ASCEND_LAUNCH_BLOCKING=1 swift sft \ --model_type 'qwen2_5-7b-instruct' \ --model_id_or_path /data1/guihonghao/models/Qwen2.5-7B-Instruct \ --sft_type 'lora' \ --tuner_backend 'peft' \ --template_type 'AUTO' \ --dtype 'bf16' \ --output_dir /data1/guihonghao/lora_outputs/qwen2_5-7b-instruct-ie-v3 \ --custom_train_dataset_path /data2/guihonghao/data/train_data/IE-v1/res/train.jsonl \ --custom_val_dataset_path /data2/guihonghao/data/train_data/IE-v1/res/dev.jsonl \ --num_train_epochs 5 \ --max_length 2048 \ --check_dataset_strategy 'warning' \ --lora_rank 64 \ --lora_alpha 64 \ --lora_dropout_p 0.05 \ --lora_target_modules 'DEFAULT' \ --gradient_checkpointing true \ --batch_size 6 \ --eval_batch_size 2 \ --weight_decay 0.1 \ --learning_rate '5e-5' \ --gradient_accumulation_steps 4 \ --max_grad_norm 0.5 \ --warmup_ratio 0.03 \ --eval_strategy 'no' \ --save_strategy 'epoch' \ --save_total_limit 10 \ --logging_steps 10 \ --preprocess_num_proc 32 \ --lazy_tokenize True \ --deepspeed 'default-zero2'