我在用mmrazor通过yolov5-x蒸馏yolov5-s时候遇到了问题

[KungFuPandaPro]

Checklist

• I have searched related issues but cannot get the expected help.
• I have read related documents and don't know what to do.

Describe the question you meet

我使用mmyolo\configs\rtmdet\distillation\kd_s_rtmdet_m_neck_300e_coco.py这个蒸馏的配置文件进行修改提供给yolov5蒸馏使用，但是控制台报错，显示TypeError: optimizer can only optimize Tensors, but one of the params is NoneType

Post related information

这是控制台相关输出
Traceback (most recent call last):
File "D:\python\mmrazor\tools\train.py", line 121, in
main()
File "D:\python\mmrazor\tools\train.py", line 117, in main
runner.train()
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\mmengine\runner\runner.py", line 1734, in train
self.optim_wrapper = self.build_optim_wrapper(self.optim_wrapper)
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\mmengine\runner\runner.py", line 1115, in build_optim_wrapper
return build_optim_wrapper(self.model, optim_wrapper)
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\mmengine\optim\optimizer\builder.py", line 210, in build_optim_wrapper
optim_wrapper = optim_wrapper_constructor(model)
File "d:\python\mmyolo\mmyolo\engine\optimizers\yolov5_optim_constructor.py", line 129, in call
optimizer = OPTIMIZERS.build(optimizer_cfg)
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\mmengine\registry\registry.py", line 570, in build
return self.build_func(cfg, *args, **kwargs, registry=self)
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\mmengine\registry\build_functions.py", line 121, in build_from_cfg
obj = obj_cls(**args) # type: ignore
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\torch\optim\sgd.py", line 27, in init
super().init(params, defaults)
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\torch\optim\optimizer.py", line 284, in init
self.add_param_group(cast(dict, param_group))
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\torch_compile.py", line 24, in inner
return torch._dynamo.disable(fn, recursive)(*args, **kwargs)
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\torch_dynamo\eval_frame.py", line 451, in _fn
return fn(*args, **kwargs)
File "C:\Users\zhan.conda\envs\mmdetection\lib\site-packages\torch\optim\optimizer.py", line 891, in add_param_group
raise TypeError("optimizer can only optimize Tensors, "
TypeError: optimizer can only optimize Tensors, but one of the params is NoneType

这是我的蒸馏配置文件
base=['yolov5_m-v61_syncbn_fast_8xb16-300e_coco.py']
teacher_ckpt = 'D:\python\mmyolo/models/yolov5_x-v61_syncbn_fast_8xb16-300e_coco_20230305_152943-00776a4b.pth'
norm_cfg = dict(type='BN', affine=False, track_running_stats=False)
#mmyolo::yolov5/yolov5_x-v61_syncbn_fast_8xb16-300e_coco.py
model = dict(
delete=True,
scope='mmrazor',
type='FpnTeacherDistill',
architecture=dict(
cfg_path='mmyolo::yolov5/yolov5_s-v61_syncbn_fast_8xb16-300e_coco.py'),
teacher=dict(
cfg_path='mmyolo::yolov5/yolov5_x-v61_syncbn_fast_8xb16-300e_coco.py'),
teacher_ckpt=teacher_ckpt,
distiller=dict(
type='ConfigurableDistiller',
# recorders are used to record various intermediate results during
# the model forward.
student_recorders=dict(
fpn0=dict(type='ModuleOutputs', source='neck.out_layers.0'),
fpn1=dict(type='ModuleOutputs', source='neck.out_layers.1'),
fpn2=dict(type='ModuleOutputs', source='neck.out_layers.2'),
),
teacher_recorders=dict(
fpn0=dict(type='ModuleOutputs', source='neck.out_layers.0'),
fpn1=dict(type='ModuleOutputs', source='neck.out_layers.1'),
fpn2=dict(type='ModuleOutputs', source='neck.out_layers.2')),
# connectors are adaptive layers which usually map teacher's and
# students features to the same dimension.
connectors=dict(
fpn0_s=dict(
type='ConvModuleConnector',
in_channel=128,
out_channel=320,
bias=False,
norm_cfg=norm_cfg,
act_cfg=None),
fpn0_t=dict(
type='NormConnector', in_channels=320, norm_cfg=norm_cfg),
fpn1_s=dict(
type='ConvModuleConnector',
in_channel=256,
out_channel=640,
bias=False,
norm_cfg=norm_cfg,
act_cfg=None),
fpn1_t=dict(
type='NormConnector', in_channels=640, norm_cfg=norm_cfg),
fpn2_s=dict(
type='ConvModuleConnector',
in_channel=512,
out_channel=1280,
bias=False,
norm_cfg=norm_cfg,
act_cfg=None),
fpn2_t=dict(
type='NormConnector', in_channels=1280, norm_cfg=norm_cfg)),
distill_losses=dict(
loss_fpn0=dict(type='ChannelWiseDivergence', loss_weight=1),
loss_fpn1=dict(type='ChannelWiseDivergence', loss_weight=1),
loss_fpn2=dict(type='ChannelWiseDivergence', loss_weight=1)),
# loss_forward_mappings are mappings between distill loss forward
# arguments and records.
loss_forward_mappings=dict(
loss_fpn0=dict(
preds_S=dict(
from_student=True, recorder='fpn0', connector='fpn0_s'),
preds_T=dict(
from_student=False, recorder='fpn0', connector='fpn0_t')),
loss_fpn1=dict(
preds_S=dict(
from_student=True, recorder='fpn1', connector='fpn1_s'),
preds_T=dict(
from_student=False, recorder='fpn1', connector='fpn1_t')),
loss_fpn2=dict(
preds_S=dict(
from_student=True, recorder='fpn2', connector='fpn2_s'),
preds_T=dict(
from_student=False, recorder='fpn2',
connector='fpn2_t')))))

find_unused_parameters = True
[#](url)

[here]

3. Your train log file if you meet the problem during training.
   [here]
4. Other code you modified in the mmrazor folder.
   [here]

[ChinaStark]

snapshot is good