pytorch / PiPPy

Pipeline Parallelism for PyTorch
BSD 3-Clause "New" or "Revised" License
715 stars 86 forks source link

[Error] pipeline() got an unexpected keyword argument #1134

Open HieronZhang opened 2 months ago

HieronZhang commented 2 months ago

Hi, I used the pytorch version 2.5.0.dev20240613+cu124 with python version 3.10.14

When I ran the OPT example "torchrun --nproc-per-node 2 pippy_opt.py", I got this error:

[rank0]: Traceback (most recent call last):
[rank0]:   File "/home/zhang402/PiPPy-main/PiPPy/examples/huggingface/pippy_opt.py", line 106, in <module>
[rank0]:     run(args)
[rank0]:   File "/home/zhang402/PiPPy-main/PiPPy/examples/huggingface/pippy_opt.py", line 46, in run
[rank0]:     pipe = pipeline(
[rank0]: TypeError: pipeline() got an unexpected keyword argument 'num_chunks'
HieronZhang commented 2 months ago

Besides, the basic examples (basic/example.py and basic/example_train.py) are still using pippy, not torch.distributed.pipelining. Could you update them? Thanks! (The example_train.py is the only one training example...)