多卡并行训练
Closed this issue · 5 comments
官方手册有没有链接啊，这个失效了，怎么设置并行训练模型？
Exception info:
{
"exc_type": "RuntimeError",
"exc_value": "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 passing the keyword argument find_unused_parameters=True to torch.nn.parallel.DistributedDataParallel, and by \nmaking sure all forward function outputs participate in calculating loss. \nIf you already have done the above, 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).\nParameter indices which did not receive grad for rank 1: 390 391 412 413 414 415 416\n In addition, you can set the environment variable TORCH_DISTRIBUTED_DEBUG to either INFO or DETAIL to print out information about which particular parameters did not receive gradient on this rank as part of this error",
"exc_time": "2024-04-06-16:45:07",
"exc_global_rank": 1,
"exc_local_rank": 1
}
Start to stop these pids:[127129, 127430, 127650], please wait several seconds.
Traceback (most recent call last):
File "/data/tmp/EfficientSAM/mine/train.py", line 218, in
trainer.run(num_train_batch_per_epoch=-1, num_eval_batch_per_dl=-1, num_eval_sanity_batch=1)
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/fastNLP/core/controllers/trainer.py", line 711, in run
raise e
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/fastNLP/core/controllers/trainer.py", line 687, in run
self.train_batch_loop.run(self, self.dataloader)
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/fastNLP/core/controllers/loops/train_batch_loop.py", line 64, in run
raise e
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/fastNLP/core/controllers/loops/train_batch_loop.py", line 55, in run
self.batch_step_fn(trainer, batch)
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/fastNLP/core/controllers/loops/train_batch_loop.py", line 76, in batch_step_fn
outputs = trainer.train_step(batch)
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/fastNLP/core/controllers/trainer.py", line 1303, in train_step
outputs = self.driver.model_call(batch, self._train_step, self._train_step_signature_fn)
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/fastNLP/core/drivers/torch_driver/ddp.py", line 512, in model_call
return self.model(batch, fastnlp_fn=fn, fastnlp_signature_fn=signature_fn,
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/data/miniconda3/envs/efficientsam/lib/python3.9/site-packages/torch/nn/parallel/distributed.py", line 994, in forward
if torch.is_grad_enabled() and self.reducer._rebuild_buckets():
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 passing the keyword argument find_unused_parameters=True to torch.nn.parallel.DistributedDataParallel, and by
making sure all forward function outputs participate in calculating loss.
If you already have done the above, 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).
Parameter indices which did not receive grad for rank 1: 390 391 412 413 414 415 416
In addition, you can set the environment variable TORCH_DISTRIBUTED_DEBUG to either INFO or DETAIL to print out information about which particular parameters did not receive gradient on this rank as part of this error
已杀死
怎么解决？
应该是在Trainer初始化的时候传一个额外的参数
应该是在 Trainer 初始化的时候传一些额外的参数
您好 中文文档的链接失效 什么时候能够恢复一下
应该是在 Trainer 初始化的时候传一些额外的参数
您好 中文文档的链接失效 什么时候能够恢复一下
请问你解决了吗
应该是在 Trainer 初始化的时候传一些额外的参数
您好中文文档的链接什么时候能够恢复一下
请问你解决了吗
你好 没有