multi-gpu training error: some model parameters don't calculate grad when training?
YongtaoGe opened this issue · comments
backbone.bottom_up.stem.conv1.weight
backbone.bottom_up.res2.0.shortcut.weight
backbone.bottom_up.res2.0.conv1.weight
backbone.bottom_up.res2.0.conv2.weight
backbone.bottom_up.res2.0.conv3.weight
backbone.bottom_up.res2.1.conv1.weight
backbone.bottom_up.res2.1.conv2.weight
backbone.bottom_up.res2.1.conv3.weight
backbone.bottom_up.res2.2.conv1.weight
backbone.bottom_up.res2.2.conv2.weight
backbone.bottom_up.res2.2.conv3.weight
head.head_series.5.self_attn.in_proj_weight
head.head_series.5.self_attn.in_proj_bias
head.head_series.5.self_attn.out_proj.weight
head.head_series.5.self_attn.out_proj.bias
head.head_series.5.inst_interact.dynamic_layer.weight
head.head_series.5.inst_interact.dynamic_layer.bias
head.head_series.5.inst_interact.norm1.weight
head.head_series.5.inst_interact.norm1.bias
head.head_series.5.inst_interact.norm2.weight
head.head_series.5.inst_interact.norm2.bias
head.head_series.5.inst_interact.out_layer.weight
head.head_series.5.inst_interact.out_layer.bias
head.head_series.5.inst_interact.norm3.weight
head.head_series.5.inst_interact.norm3.bias
head.head_series.5.linear1.weight
head.head_series.5.linear1.bias
head.head_series.5.linear2.weight
head.head_series.5.linear2.bias
head.head_series.5.norm1.weight
head.head_series.5.norm1.bias
head.head_series.5.norm2.weight
head.head_series.5.norm2.bias
head.head_series.5.norm3.weight
head.head_series.5.norm3.bias
head.head_series.5.cls_module.0.weight
head.head_series.5.cls_module.1.weight
head.head_series.5.cls_module.1.bias
head.head_series.5.reg_module.0.weight
head.head_series.5.reg_module.1.weight
head.head_series.5.reg_module.1.bias
head.head_series.5.reg_module.3.weight
head.head_series.5.reg_module.4.weight
head.head_series.5.reg_module.4.bias
head.head_series.5.reg_module.6.weight
head.head_series.5.reg_module.7.weight
head.head_series.5.reg_module.7.bias
head.head_series.5.class_logits.weight
head.head_series.5.class_logits.bias
head.head_series.5.bboxes_delta.weight
head.head_series.5.bboxes_delta.bias
head.iter_series.0.reg_module.0.weight
head.iter_series.0.reg_module.1.weight
head.iter_series.0.reg_module.1.bias
head.iter_series.0.reg_module.3.weight
head.iter_series.0.reg_module.4.weight
head.iter_series.0.reg_module.4.bias
head.iter_series.0.reg_module.6.weight
head.iter_series.0.reg_module.7.weight
head.iter_series.0.reg_module.7.bias
head.iter_series.0.bboxes_delta.weight
head.iter_series.0.bboxes_delta.bias
你好 怎么解决这个问题的 我用两张卡训练出现这个问题：
RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one. This error indicates that your module has parameters that were not used in producing loss. You can enable unused parameter detection by (1) passing the keyword argument find_unused_parameters=True to torch.nn.parallel.DistributedDataParallel; (2) making sure all forward function outputs participate in calculating loss. If you already have done the above two steps, then the distributed data parallel module wasn't able to locate the output tensors in the return value of your module's forward function. Please include the loss function and the structure of the return value of forward of your module when reporting this issue (e.g. list, dict, iterable).
