Apply for community grant: Academic project (gpu)

#2
by LXT - opened
Owner

Hi!I want to apply one GPU grant to supply our online demo. Thanks!

Paper Link: https://arxiv.org/abs/2401.10229

Code: https://github.com/lxtGH/OMG-Seg

Demo: https://huggingface.co./spaces/LXT/OMG_Seg

Hi @LXT , we've assigned ZeroGPU to this Space. Please check the usage section of this page so your Space can run on ZeroGPU. The @spaces decorator does nothing on non-ZeroGPU Spaces or local environments, so you can safely add it.

Owner

Configuration error.

No candidate PyTorch version found for ZeroGPU.

@hysts

How to handle this?

Hmm, not sure what is causing the error.
@cbensimon Could you take a look at this?

For now ZeroGPU is only compatible with PyTorch versions 2.0.0 and 2.0.1
(newer versions will come in a couple weeks)

@LXT do you think that your demo can work on PyTorch 2.0.0 or 2.0.1 ?

By the way I added you to https://huggingface.co./zero-gpu-explorers so you can create ZeroGPU Spaces freely as well as move this one back and forth between Zero Nvidia A10G and other hardware

Owner

Hi! I still meet the runtime error since we find our dependency (mmengine) can not run the Zero Nvidia A10G.

Hi @cbensimon ,

I have switched to PyTorch 2.0.1. However, I encountered some errors (here). It looks like the function I'm using cannot be compiled. Is there any way to bypass compilation?

Thanks for your help.

The error log is this:

===== Application Startup at 2024-01-23 03:53:10 =====

Traceback (most recent call last):
  File "/home/user/app/main.py", line 12, in <module>
    from mmdet.visualization import DetLocalVisualizer
  File "/home/user/.local/lib/python3.10/site-packages/mmdet/visualization/__init__.py", line 2, in <module>
    from .local_visualizer import DetLocalVisualizer, TrackLocalVisualizer
  File "/home/user/.local/lib/python3.10/site-packages/mmdet/visualization/local_visualizer.py", line 15, in <module>
    from mmengine.visualization import Visualizer
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/visualization/__init__.py", line 2, in <module>
    from .vis_backend import (AimVisBackend, BaseVisBackend, ClearMLVisBackend,
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/visualization/vis_backend.py", line 19, in <module>
    from mmengine.hooks.logger_hook import SUFFIX_TYPE
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/hooks/__init__.py", line 4, in <module>
    from .ema_hook import EMAHook
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/hooks/ema_hook.py", line 8, in <module>
    from mmengine.model import is_model_wrapper
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/model/__init__.py", line 6, in <module>
    from .base_model import BaseDataPreprocessor, BaseModel, ImgDataPreprocessor
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/model/base_model/__init__.py", line 2, in <module>
    from .base_model import BaseModel
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/model/base_model/base_model.py", line 9, in <module>
    from mmengine.optim import OptimWrapper
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/optim/__init__.py", line 2, in <module>
    from .optimizer import (OPTIM_WRAPPER_CONSTRUCTORS, OPTIMIZERS,
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/optim/optimizer/__init__.py", line 10, in <module>
    from .zero_optimizer import ZeroRedundancyOptimizer
  File "/home/user/.local/lib/python3.10/site-packages/mmengine/optim/optimizer/zero_optimizer.py", line 11, in <module>
    from torch.distributed.optim import \
  File "/home/user/.local/lib/python3.10/site-packages/torch/distributed/optim/__init__.py", line 11, in <module>
    from .functional_adadelta import _FunctionalAdadelta
  File "/home/user/.local/lib/python3.10/site-packages/torch/distributed/optim/functional_adadelta.py", line 20, in <module>
    class _FunctionalAdadelta:
  File "/home/user/.local/lib/python3.10/site-packages/torch/jit/_script.py", line 1321, in script
    _compile_and_register_class(obj, _rcb, qualified_name)
  File "/home/user/.local/lib/python3.10/site-packages/torch/jit/_recursive.py", line 51, in _compile_and_register_class
    script_class = torch._C._jit_script_class_compile(qualified_name, ast, defaults, rcb)
  File "/home/user/.local/lib/python3.10/site-packages/torch/jit/_recursive.py", line 867, in try_compile_fn
    return torch.jit.script(fn, _rcb=rcb)
  File "/home/user/.local/lib/python3.10/site-packages/torch/jit/_script.py", line 1338, in script
    ast = get_jit_def(obj, obj.__name__)
  File "/home/user/.local/lib/python3.10/site-packages/torch/jit/frontend.py", line 297, in get_jit_def
    return build_def(parsed_def.ctx, fn_def, type_line, def_name, self_name=self_name, pdt_arg_types=pdt_arg_types)
  File "/home/user/.local/lib/python3.10/site-packages/torch/jit/frontend.py", line 335, in build_def
    param_list = build_param_list(ctx, py_def.args, self_name, pdt_arg_types)
  File "/home/user/.local/lib/python3.10/site-packages/torch/jit/frontend.py", line 359, in build_param_list
    raise NotSupportedError(ctx_range, _vararg_kwarg_err)
torch.jit.frontend.NotSupportedError: Compiled functions can't take variable number of arguments or use keyword-only arguments with defaults:
  File "/home/user/.local/lib/python3.10/site-packages/spaces/zero/torch.py", line 74
    def _tensor_register(*args: Any, **kwargs: Any):
                                      ~~~~~~~ <--- HERE
        try:
            device = torch.device(kwargs.get('device', "cpu"))

@LXT @HarborYuan
Due to the time zone difference, it might take some time for @cbensimon to respond, so in the meantime, I'll just change the hardware to a10g-small, and we'll see if this Space can run on ZeroGPU later.

Hi @hysts ,

Thanks for your help. It works well now.

Hi @LXT , unfortunately ZeroGPU does not support torch.compile / torch.jit at the moment

@cbensimon Thanks for checking! I haven't looked into the details, but so torch.compile is used by the visualizer DetLocalVisualizer? Not 100% sure, but I kind of feel that a visualizer shouldn't block the use of ZeroGPU and that it would be nice to be able to support torch.compile/torch.jit. Anyway, I think we should keep using a10g-small for this Space, then.

Sign up or log in to comment