[BUG] M1 Mac has an issue with hostname -I not being a valid command

[AbhinavMir]

Not a particular issue I was facing, but a very offline friend who was using deepspeed initializes got an error where HostName -I isn't available on the M1 chip or OSX in general.

A workaround I can think of is using ifconfig | grep 'inet ' | awk '{print $2}' | paste -sd ' ' instead of hostname -I - and this code fails in ./deepspeed/launcher/runner.py: ssh_check_cmd += f" {first_host} hostname -I" so probably a check for OS type using uname and then running a specific bit of code. Currently I just added a new binary to replace hostname that runs the original hostname but also parses -I to run ifconfig but it probably isn't the best idea.

I am not an AI developer but just my 2 cents!

[loadams]

Hi @AbhinavMir - can you share the specific error? I know others have been able to run on M1 Macs in the past, so is it possible that the inetutils package wasn't installed which contains hostname?

[AbhinavMir]

Hi Logan, I'll talk to the friend really quick and get back to you, but iirc hostname was available, but not the -I flag.

[loadams]

Hi @AbhinavMir - closing this PR as stale for now, if you're able to get a little more info, please comment and we can investigate and re-open.

[RuslanProgrammer]

Hello. I got the same error on Apple M3, MacOS Solomona 14.5 (23F79)

[2024-07-01 11:07:31,491] [INFO] [logging.py:96:log_dist] [Rank -1] DeepSpeed info: version=0.14.4, git-hash=unknown, git-branch=unknown
[2024-07-01 11:07:31,491] [INFO] [comm.py:637:init_distributed] cdb=None
[2024-07-01 11:07:31,492] [INFO] [comm.py:652:init_distributed] Not using the DeepSpeed or dist launchers, attempting to detect MPI environment...
[...70358] shmem: mmap: an error occurred while determining whether or not /var/folders/75/... could be created.
hostname: illegal option -- I
usage: hostname [-f] [-s | -d] [name-of-host]
Traceback (most recent call last):
  File "/Users/.../class.py", line 139, in <module>
    model_engine, optimizer,__, __ = deepspeed.initialize(
  File "/Users/.../.venv/lib/python3.9/site-packages/deepspeed/__init__.py", line 143, in initialize
    dist.init_distributed(dist_backend=dist_backend,
  File "/Users/.../.venv/lib/python3.9/site-packages/deepspeed/comm/comm.py", line 658, in init_distributed
    mpi_discovery(distributed_port=distributed_port, verbose=verbose)
  File "/Users/.../.venv/lib/python3.9/site-packages/deepspeed/comm/comm.py", line 686, in mpi_discovery
    result = subprocess.check_output(hostname_cmd, shell=True)
  File "/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.9/lib/python3.9/subprocess.py", line 424, in check_output
    return run(*popenargs, stdout=PIPE, timeout=timeout, check=True,
  File "/Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.9/lib/python3.9/subprocess.py", line 528, in run
    raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command '['hostname -I']' returned non-zero exit status 1.

So when I run hostname -I in the terminal, I get the same error. Moreover inetutils 2.5 is already installed.

[loadams]

Hi @RuslanProgrammer - can you please share the output of running just hostname -I in the terminal?

[RuslanProgrammer]

$ hostname -I
hostname: illegal option -- I
usage: hostname [-f] [-s | -d] [name-of-host]

[loadams]

Thanks @RuslanProgrammer and @AbhinavMir - we've confirmed on an Intel Mac that the -I option is also not available there, so this appears to be an issue with OSX rather than the M series Macs.

We will need to find either another way to list the IPs on Macs, I'll have to look into this.