load_from conflicts with resume

[serser]

Prerequisite

• [X] I have searched the existing and past issues but cannot get the expected help.
• [X] I have read the FAQ documentation but cannot get the expected help.
• [X] The bug has not been fixed in the latest version.

🐞 Describe the bug

I am training with COCO pretrained YOLOv5m model (specifying load_from=xxx in the config) on a custom dataset, after training for a while, the training stopped on exception. I now try with --resume to resume the training. It apparently loads load_from other than saved ckpt under work_dir. I now need to remove load_from from config to resume. So what is the suggested way to do this?

Traceback (most recent call last):
  File "tools/train.py", line 106, in <module>
    main()
  File "tools/train.py", line 102, in main
    runner.train()
  File "/anaconda3/envs/mm2_cu11/lib/python3.8/site-packages/mmengine/runner/runner.py", line 1653, in train
    self.load_or_resume()
  File "/anaconda3/envs/mm2_cu11/lib/python3.8/site-packages/mmengine/runner/runner.py", line 1599, in load_or_resume
    self.resume(resume_from)
  File "/anaconda3/envs/mm2_cu11/lib/python3.8/site-packages/mmengine/runner/runner.py", line 1954, in resume
    self.message_hub.load_state_dict(checkpoint['message_hub'])
KeyError: 'message_hub'
/anaconda3/envs/mm2_cu11/lib/python3.8/site-packages/mmengine/runner/runner.py:1948: UserWarning: The dataset metainfo from the resumed checkpoint is different from the current training dataset, please check the correctness of the checkpoint or the training dataset.

Environment

sys.platform: linux
Python: 3.8.13 (default, Oct 21 2022, 23:50:54) [GCC 11.2.0]
CUDA available: True
numpy_random_seed: 2147483648
GPU 0,1,2,3: Tesla V100-SXM2-32GB
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 11.0, V11.0.221
GCC: gcc (GCC) 5.4.0
PyTorch: 1.7.1+cu110
PyTorch compiling details: PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v1.6.0 (Git Hash 5ef631a030a6f73131c77892041042805a06064f)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 11.0
  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80
  - CuDNN 8.0.5
  - Magma 2.5.2
  - Build settings: BLAS=MKL, BUILD_TYPE=Release, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DUSE_VULKAN_WRAPPER -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, USE_CUDA=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, 

TorchVision: 0.8.2+cu110
OpenCV: 4.6.0
MMEngine: 0.3.0
MMCV: 2.0.0rc2
MMDetection: 3.0.0rc2
MMYOLO: 0.1.2+dc3377b

Additional information

No response

[RangeKing]

If load_from and resume=True are both set, only load_from will take effect according to MMEngine docs. It could be more convenient. I already sent feedback to mmengine developers. They will discuss it later.