Ignite: How to all_gather Tensor if not the same length

[Nic-Ma]

❓ Questions/Help/Support

Hi @vfdev-5 ,

I am developing distributed evaluation feature, and facing a problem that the preds and labels on different GPUs don't have the same length, then ignite.idist.all_gather() can't work. For example: GPU0 has 5 images to handle, GPU1 has 4 images, total=9 images.
Could you please help on how to idist.all_gather() the values?
I don't want to pad data for the input to make it evenly-divisible, because it will cause the metrics different on single GPU and multi-GPUs.

Thanks in advance.

[Nic-Ma]

BTW, I mean when using ignite.metrics.EpochMetric, the inputs preds and labels have different lengths on different GPUs.

Thanks.

[vfdev-5]

Hi @Nic-Ma ,

I think this can be done with the broadcast collective op:

• broadcast for each process its data to everyone
• concat results

Probably, this can be slow and thus could not be used everytime... I'll prototype a code snippet a bit later.

In this issue: https://github.com/pytorch/ignite/issues/1288, we though about providing a feature to customize reduction/gather ops for metrics, such that user could use its own adapted solution as well.

[Nic-Ma]

Hi @vfdev-5 ,

Or maybe we can pad NaN Tensor to make it evenly-divisible, and delete NaN number after all_gather()?

Thanks.

[vfdev-5]

Or maybe we can pad NaN Tensor to make it evenly-divisible, and delete NaN number after all_gather()?

Well, yes, padding can be an option too. But in the first message you said you did not want to pad...

[Nic-Ma]

Hi @vfdev-5 ,

Sorry I didn't make it clear, I mean we don't want to pad data for the DistributedSampler before model prediction.
So maybe it's possible to only pad NaN before all_gather() and delete NaN after.

Thanks.

[Nic-Ma]

Hi @vfdev-5 ,

I am facing an issue with idist.all_gather() for string:
Here I put all the filenames of 1 GPU to self.filenames list and join them into 1 string, then try to gather, but it always hangs at the all_gather line.

_filenames = "\t".join(self._filenames)
_filenames = idist.all_gather(_filenames)

My PyTorch version is 1.7.0 and GPU is V100, do you know any reason of this issue?

Thanks.

[vfdev-5]

Hi @Nic-Ma

here is a draft version for all_gather with padding

# Question : https://github.com/pytorch/ignite/issues/1569

import os
import time
from typing import Optional, List

import torch
import torch.nn.functional as F
import torch.distributed as dist


def compute_padding(shape, new_shape):
    padding = []
    for dim, new_dim in zip(shape, new_shape):
        padding.insert(0, new_dim - dim)
        padding.insert(0, 0)
    return padding


def all_gather(tensor: torch.Tensor, fixed_shape: Optional[List] = None) -> List[torch.Tensor]:
    input_shape = tensor.shape
    if fixed_shape is not None:
        padding = compute_padding(tensor.shape, fixed_shape)
        if sum(padding) > 0:
            tensor = F.pad(tensor, pad=padding, mode='constant', value=0)

    output = [torch.zeros_like(tensor) for _ in range(dist.get_world_size())]
    dist.all_gather(output, tensor)

    all_input_shapes = None
    if fixed_shape is not None:
        # gather all shapes
        tensor_shape = torch.tensor(input_shape, device=tensor.device)
        all_input_shapes = [torch.zeros_like(tensor_shape) for _ in range(dist.get_world_size())]
        dist.all_gather(all_input_shapes, tensor_shape)
        all_input_shapes = [t.tolist() for t in all_input_shapes]

    if all_input_shapes:
        for i, shape in enumerate(all_input_shapes):
            padding = compute_padding(output[i].shape, shape)
            if sum(padding) < 0:
                output[i] = F.pad(output[i], pad=padding)

    return output


if __name__ == "__main__":

    dist.init_process_group("nccl", init_method="env://")

    lrank = int(os.environ["LOCAL_RANK"])
    torch.cuda.set_device(lrank)
    rank = dist.get_rank()

    size = 5
    uneven_size = 2
    t = torch.arange(dist.get_world_size() * size - uneven_size, device="cuda")

    input_tensor = t[size * rank: size * (rank + 1)]
    time.sleep(0.5 * (dist.get_rank() + 1))
    print(rank, " - Input : ", input_tensor)

    result = all_gather(input_tensor, fixed_shape=(size, ))

    time.sleep(0.5 * (dist.get_rank() + 1))
    print(rank, " - Output: ", result)

    dist.destroy_process_group()

Run as

python -u -m torch.distributed.launch --nproc=2 --use_env question-uneven-input-all-gather.py 

0  - Input :  tensor([0, 1, 2, 3, 4], device='cuda:0')
1  - Input :  tensor([5, 6, 7], device='cuda:1')
0  - Output:  [tensor([0, 1, 2, 3, 4], device='cuda:0'), tensor([5, 6, 7], device='cuda:0')]
1  - Output:  [tensor([0, 1, 2, 3, 4], device='cuda:1'), tensor([5, 6, 7], device='cuda:1')]

As for idist.all_gather hang with string input. Let me try to reproduce it and will see.

BTW, what is the best way to reach out to you to communicate with messages to discuss about few things ?
I sent you few messages on MONAIBoot2020 slack...

[vfdev-5]

@Nic-Ma I couldn't reproduce your issue.

Here is my code:

import os
import time

import torch
import torch.distributed as dist
import ignite
import ignite.distributed as idist


if __name__ == "__main__":

    print(torch.__version__, ignite.__version__)

    dist.init_process_group("nccl", init_method="env://")

    lrank = int(os.environ["LOCAL_RANK"])
    torch.cuda.set_device(lrank)
    rank = dist.get_rank()

    all_filenames = [f"filename_{i}_{rank}" for i in range(10)]
    data = "\t".join(all_filenames)
    time.sleep(0.5 * (dist.get_rank() + 1))
    print(rank, " - Input : ", data)

    result = idist.all_gather(data)

    time.sleep(0.5 * (dist.get_rank() + 1))
    print(rank, " - Output: ", result)

    dist.destroy_process_group()

and output

# python -u -m torch.distributed.launch --nproc=2 --use_env issue-1569-repro-hang-idist-gather-string.py 
1.7.1 0.5.0.dev20210121
1.7.1 0.5.0.dev20210121
0  - Input :  filename_0_0      filename_1_0    filename_2_0    filename_3_0    filename_4_0    filename_5_0    filename_6_0    filename_7_0    filename_8_0    filename_9_0
1  - Input :  filename_0_1      filename_1_1    filename_2_1    filename_3_1    filename_4_1    filename_5_1    filename_6_1    filename_7_1    filename_8_1    filename_9_1
0  - Output:  ['filename_0_0\tfilename_1_0\tfilename_2_0\tfilename_3_0\tfilename_4_0\tfilename_5_0\tfilename_6_0\tfilename_7_0\tfilename_8_0\tfilename_9_0', 'filename_0_1\tfilename_1_1\tfilename_2_1\tfilename_3_1\tfilename_4_1\tfilename_5_1\tfilename_6_1\tfilename_7_1\tfilename_8_1\tfilename_9_1']
1  - Output:  ['filename_0_0\tfilename_1_0\tfilename_2_0\tfilename_3_0\tfilename_4_0\tfilename_5_0\tfilename_6_0\tfilename_7_0\tfilename_8_0\tfilename_9_0', 'filename_0_1\tfilename_1_1\tfilename_2_1\tfilename_3_1\tfilename_4_1\tfilename_5_1\tfilename_6_1\tfilename_7_1\tfilename_8_1\tfilename_9_1']

Please, let me know if my example correctly implements your issue. Thanks

[Nic-Ma]

Hi @vfdev-5 ,

Thanks very much for your detailed program!!
I also can't reproduce the issue with your program, let me try to make a program to reproduce it.
And BTW, I don't know why I can't login to the BootCamp workspace in slack anymore...So maybe let's communicate in email? my email [email protected] is always online.

Thanks.

[Nic-Ma]

I compared your program with mine and found the root cause, sorry for my mistake, the all_gather works for my string now.
Thanks very much for your help and example program!!!

[vfdev-5]

@Nic-Ma how about this code: https://github.com/pytorch/ignite/issues/1569#issuecomment-766739042 for all_gather tensors of different shapes ?

We may think to add more functions like that (e.g. all_reduce) and put them into contrib.distributed.utils module...

[Nic-Ma]

Hi @vfdev-5 ,

Your example code looks good, and I developed a evenly_divisible_all_gather() in MONAI now to handle this case:

def evenly_divisible_all_gather(data: torch.Tensor):
    """
    Utility function for distributed data parallel to pad tensor to make it evenly divisible for all_gather.
    Args:
        data: source tensor to pad and execute all_gather in distributed data parallel.

    """
    if idist.get_world_size() <= 1:
        return data
    # make sure the data is evenly-divisible on multi-GPUs
    length = data.shape[0]
    all_lens = idist.all_gather(length)
    max_len = max(all_lens).item()
    if length < max_len:
        size = [max_len - length] + list(data.shape[1:])
        data = torch.cat([data, data.new_full(size, float("NaN"))], dim=0)
    # all gather across all processes
    data = idist.all_gather(data)
    # delete the padding NaN items
    return torch.cat([data[i * max_len : i * max_len + l, ...] for i, l in enumerate(all_lens)], dim=0)

Thanks.