FileNotFoundError: Cannot find DGL C++ graphbolt library

[EthanGuo13]

在运行readme测试代码时，出现了如下报错：
(RFdiffusion) ethanguo@EthandeMacBook-Air RFdiffusion % ./scripts/run_inference.py 'contigmap.contigs=[150-150]' inference.output_prefix=test_outputs/test inference.num_designs=10
/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/rfdiffusion/util.py:253: UserWarning: Using torch.cross without specifying the dim arg is deprecated.
Please either pass the dim explicitly or simply use torch.linalg.cross.
The default value of dim will change to agree with that of linalg.cross in a future release. (Triggered internally at /private/var/folders/k1/30mswbxs7r1g6zwn8y4fyt500000gp/T/abs_e2rjusv8ek/croot/pytorch-select_1717607459930/work/aten/src/ATen/native/Cross.cpp:66.)
Z = torch.cross(Xn, Yn)
DGL backend not selected or invalid. Assuming PyTorch for now.
Setting the default backend to "pytorch". You can change it in the ~/.dgl/config.json file or export the DGLBACKEND environment variable. Valid options are: pytorch, mxnet, tensorflow (all lowercase)
Traceback (most recent call last):
File "/Users/ethanguo/RFdiffusion/./scripts/run_inference.py", line 25, in
from rfdiffusion.inference import utils as iu
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/rfdiffusion/inference/utils.py", line 6, in
from rfdiffusion.diffusion import get_beta_schedule
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/rfdiffusion/diffusion.py", line 12, in
from rfdiffusion.util_module import ComputeAllAtomCoords
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/rfdiffusion/util_module.py", line 7, in
import dgl
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/init.py", line 16, in
from . import (
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/dataloading/init.py", line 13, in
from .dataloader import *
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/dataloading/dataloader.py", line 27, in
from ..distributed import DistGraph
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/distributed/init.py", line 5, in
from .dist_graph import DistGraph, DistGraphServer, edge_split, node_split
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/distributed/dist_graph.py", line 11, in
from .. import backend as F, graphbolt as gb, heterograph_index
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/graphbolt/init.py", line 36, in
load_graphbolt()
File "/Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/graphbolt/init.py", line 26, in load_graphbolt
raise FileNotFoundError(
FileNotFoundError: Cannot find DGL C++ graphbolt library at /Users/ethanguo/anaconda3/envs/RFdiffusion/lib/python3.10/site-packages/dgl/graphbolt/libgraphbolt_pytorch_2.3.0.post100.dylib
请问前辈是因为M4芯片和PyTorch不兼容导致的吗 @YaoYinYing

[YaoYinYing]

Hello! @ethanguo-stack

The latest version dgl prebuild wheel that works on macos is 2.2.1. You need have to install the exact Pytorch version it supposes to support.

Please check the dgl page and release (v2.2.1) note for more helpful msgs.

Also, the performance on cpu only for Apple silicon is pretty bad. I personally recommend you to switch to the mps branch, so that the GPU acceleration will save you quite a lot of time.

Regards, Yinying.

[EthanGuo13]

好的，谢谢前辈，我想重新安装一边试一下（昨天安装步骤完全是按照您的文件来的，是pytorch2.3.0&dgl2.2.1，但是还是说这个library没有，所以说我怀疑是M4芯片是不是不兼容的问题），在您的RFD repo中克隆的仍是rosetta的库，是不是可以直接克隆您这个repo，还有使用mps分支是要单独设置吗，您的readme里是不是没有涉及，readme中pytorch安装注释的是CPUonly，是不是就不能用mps了，刚刚接触RFD以及Deep learning，没有什么计算机基础，问的问题可能比较无厘头望您海涵 @YaoYinYing

[YaoYinYing]

@ethanguo-stack Ah I see. The instruction on readme described a generalized installation so that users can follow it easily. I would personally prefer solely pip installs.

Heres the dependency configuration from one of my in-dev package, in which you may find some hints using pip installs.

rfdiffusion_mps=[
    "biotite>=1.1.0",
    "pyrsistent",
    "torch<=2.3.0",
    # dgl must be installed from https://data.dgl.ai/wheels/repo.html
    "torchvision", 
    "torchdata==0.7.1",
    "torchaudio",
    "nvtx-mock", 
    "nvtx",
    "se3-transformer @ git+https://github.com/YaoYinYing/SE3Transformer.git@rfdiffusion-mps-test",
    "dllogger @ git+https://github.com/NVIDIA/dllogger#egg=dllogger",
    "pydantic",
    "rfdiffusion @ git+https://github.com/YaoYinYing/RFdiffusion.git@mps-test",
    "wandb==0.16.0"
]

Hope this workaround to help you on your machine!

[EthanGuo13]

Thanks a lot! The problem has been solved, best wishes!

[EthanGuo13]

@YaoYinYing Sorry for my interruption, sir, I still have a question about how to use MPS while running RFD, the problem is origin from scripts/run_inference.py, when I use git clone https://github.com/YaoYinYing/RFdiffusion.git run_inference.pyis still the origin one from rosetta rather than your mps-test branch, and if I straightly change run_inference.py from rosetta to run_inference.py from mps-test branch in file RFDiffusion in user file, RFD can't run then. Sorry again for my stupid questions, best regards!

[YaoYinYing]

No worry @ethanguo-stack .

Did you run git branch mps-test on your local cloned repo?

I'm no sure how you setup and ran the script. But I could tell you that the default branch shares the original code base as the upstream , while the mps flavored branch has slight changes for utilizing the mps GPU.

So please do make sure that you have the mps branch installed via pip, as well as the mps related runtime dependencies. Also, just call the inference script out of the code dir. The mps GPU shall be recognized seamlessly.

Let me know if you have any questions about the installation.

[EthanGuo13]

@YaoYinYing Nope, I didn't do that, I'll try this tomorrow in my office, the network speed in my dorm in slow, it's hard to reinstall now, thank you again sir! thank you for your patience!