Tlntin / qwen-ascend-llm

Apache License 2.0
27 stars 2 forks source link

模型转换 #5

Open cuicui2023 opened 2 hours ago

cuicui2023 commented 2 hours ago

您好,能问一下您做onnx到om模型转换这步用了多久吗?我是Ascend310B1的板子,12G内存,24G交换区,onnx到om模型转换这步好事非常久(好几个小时),然后最后还没成功。 “ [INFO] soc_version is auto, will auto detect soc version [INFO] {'soc_full_name': 'Ascend310B1', 'soc_short_name': 'Ascend310B'} ============ run command ============== export MS_DEV_FORCE_ACL=1 && export MS_ENABLE_GE=1 && export TE_PARALLEL_COMPILER=1 && export MAX_COMPILE_CORE_NUMBER=1 && atc --framework=5 --model="./output/onnx2/qwen2_1.5b_chat.onnx" --output="./output/model/qwen2_1.5b_chat" --soc_version=Ascend310B1 --precision_mode_v2=mixed_float16 --modify_mixlist=/home/HwHiAiUser/model/qwen-ascend-llm/ops_info.json --input_format=ND --input_shape="input_ids:1,-1;attention_mask:1,-1;position_ids:1,-1;past_key_values:1,-1,112,128" --dynamic_dims "1,1025,1,1024;2,1026,2,1024;4,1028,4,1024;1,2049,1,2048;2,2050,2,2048;4,2052,4,2048"

ATC start working now, please wait for a moment. Traceback (most recent call last): File "/usr/local/Ascend/ascend-toolkit/latest/python/site-packages/te_fusion/fusion_manager.py", line 1912, in set_op_import_module importlib.import_module(module_name) File "/usr/local/miniconda3/lib/python3.9/importlib/init.py", line 127, in import_module return _bootstrap._gcd_import(name[level:], package, level) File "", line 1030, in _gcd_import File "", line 1007, in _find_and_load File "", line 986, in _find_and_load_unlocked File "", line 680, in _load_unlocked File "", line 786, in exec_module File "", line 922, in get_code File "", line 979, in get_data PermissionError: [Errno 13] Permission denied: '/usr/local/Ascend/ascend-toolkit/7.0.RC1/opp/vendors/customize/op_impl/ai_core/tbe/customize_impl/mrgba.py' ” 这是怎么回事啊?

Tlntin commented 2 hours ago

用了多久?回答:10-20分钟。 怎么回事呢?回答:看你的日志:PermissionError: [Errno 13] Permission denied: '/usr/local/Ascend/ascend-toolkit/7.0.RC1/opp/vendors/customize/op_impl/ai_core/tbe/customize_impl/mrgba.py',可能是cann版本较低(看readme推荐的版本),并且建议用root账号操作。