[BUG] RuntimeError: Error building extension 'utils' (`ninja` related?)

[josephrocca]

Describe the bug As shown in this notebook, I run these commands:

pip install deepspeed --upgrade
git clone https://github.com/microsoft/DeepSpeedExamples
cd DeepSpeedExamples/model_compression/gpt2
pip install -r requirements.txt
sudo apt-get install ninja-build # I don't think this line is actually needed, but I'm not sure
pip install ninja
bash ./bash_script/run_zero_quant.sh

This is exactly following the instructions in the readme of DeepSpeedExamples/tree/master/model_compression/gpt2 except that I had to install ninja because the machine didn't have it yet.

And after some progress, the run_zero_quant.sh script throws RuntimeError: Error building extension 'utils' (please see the notebook for full logs).

To Reproduce Steps to reproduce the behavior:

1. Run this notebook: https://gist.github.com/josephrocca/9ec65e8e5804286a475b5b6da85f7a28

Expected behavior There is a related issue here:

• https://github.com/microsoft/DeepSpeed/issues/694

The apparent solution there was to ensure that the deepspeed wheel was built with the same cuda version as the machine has installed. But the ds_report shows that the versions match. So I guess the "expected behavior" here is that it shouldn't throw the error that I'm seeing.

ds_report output As seen in the above-linked notebook:

--------------------------------------------------
DeepSpeed C++/CUDA extension op report
--------------------------------------------------
NOTE: Ops not installed will be just-in-time (JIT) compiled at
      runtime if needed. Op compatibility means that your system
      meet the required dependencies to JIT install the op.
--------------------------------------------------
JIT compiled ops requires ninja
ninja .................. [OKAY]
--------------------------------------------------
op name ................ installed .. compatible
--------------------------------------------------
cpu_adam ............... [NO] ....... [OKAY]
cpu_adagrad ............ [NO] ....... [OKAY]
fused_adam ............. [NO] ....... [OKAY]
fused_lamb ............. [NO] ....... [OKAY]
 [WARNING]  please install triton==1.0.0 if you want to use sparse attention
sparse_attn ............ [NO] ....... [NO]
transformer ............ [NO] ....... [OKAY]
stochastic_transformer . [NO] ....... [OKAY]
 [WARNING]  async_io requires the dev libaio .so object and headers but these were not found.
 [WARNING]  async_io: please install the libaio-dev package with apt
 [WARNING]  If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
async_io ............... [NO] ....... [NO]
utils .................. [NO] ....... [OKAY]
quantizer .............. [NO] ....... [OKAY]
transformer_inference .. [NO] ....... [OKAY]
--------------------------------------------------
DeepSpeed general environment info:
torch install path ............... ['/usr/lib/python3/dist-packages/torch']
torch version .................... 1.11.0
torch cuda version ............... 11.6
torch hip version ................ None
nvcc version ..................... 11.6
deepspeed install path ........... ['/home/ubuntu/.local/lib/python3.8/site-packages/deepspeed']
deepspeed info ................... 0.7.0, unknown, unknown
deepspeed wheel compiled w. ...... torch 1.11, cuda 11.6

System info (please complete the following information):

• OS: Ubuntu 20.04
• GPU count and types: 1x RTX 6000 (24GB)
• Python version: 3.8
• Any other relevant info about your setup: I used https://lambdalabs.com/ GPU cloud (using their Cloud IDE)

[mrwyattii]

Hi @josephrocca, thanks for using DeepSpeed. Could you try pre-compiling and let me know the outcome? To do so:

• Uninstall DeepSpeed pip uninstall -y deepspeed
• clone our repo git clone https://github.com/microsoft/DeepSpeed.git && cd DeepSpeed
• install with either DS_BUILD_OPS=1 pip install . or DS_BUILD_UTILS=1 pip install . (read more here: https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)

[josephrocca]

Hi @mrwyattii, I tried both the DS_BUILD_OPS option and the DS_BUILD_UTILS option on a fresh Lambda Cloud machine, and both gave errors. Please see here for the full error logs of both attempts: https://gist.github.com/josephrocca/8417c4665cbfef89ba85e439c17500da

[5hadytru]

Solution?

[tjruwase]

I see this error message in the gist log. Can you confirm that pybind11 is installed?

[loadams]

This looks to have been pybind11 related, if you are still having issues with this, please re-open.

[Robert11092002]

sudo apt install python3-pybind11

[CattleHome]

windows上，我在这里https://pypi.org/project/deepspeed/#files 下载了对应的包，解压之后直接放在虚拟环境里可以成功