saymrwulf/transformers
1
0
Fork 0
mirror of https://github.com/saymrwulf/transformers.git synced 2026-05-14 20:58:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[deepspeed] test on one node 2 gpus max (#11237)

Browse source 
* test on one node 2 gpus max

* fix the other place

* refactor

* fix

* cleanup

* more exact version
This commit is contained in:
Stas Bekman 2021-04-14 11:06:59 -07:00 committed by GitHub
parent 25e1af36e0
commit 83206ca6a8
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 11 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
setup.py
View file

@ -90,7 +90,7 @@ _deps = [
"cookiecutter==1.7.2",
"dataclasses",
"datasets",
"deepspeed>0.3.13",
"deepspeed>=0.3.14",
"docutils==0.16.0",
"fairscale>0.3",
"faiss-cpu",

15
tests/deepspeed/test_deepspeed.py
View file

@ -595,8 +595,7 @@ class TestDeepSpeedWithLauncher(TestCasePlus):
ds_args = f"--deepspeed {self.test_file_dir_str}/ds_config_{stage}.json".split()
script = [f"{self.examples_dir_str}/seq2seq/run_translation.py"]
num_gpus = get_gpu_count() if distributed else 1
launcher = f"deepspeed --num_gpus {num_gpus}".split()
launcher = self.get_launcher(distributed)
cmd = launcher + script + args + ds_args
# keep for quick debug
@ -629,11 +628,9 @@ class TestDeepSpeedWithLauncher(TestCasePlus):
--block_size 128
""".split()
distributed = True
ds_args = f"--deepspeed {self.test_file_dir_str}/ds_config_{stage}.json".split()
script = [f"{self.examples_dir_str}/language-modeling/run_clm.py"]
num_gpus = get_gpu_count() if distributed else 1
launcher = f"deepspeed --num_gpus {num_gpus}".split()
launcher = self.get_launcher(distributed=True)
cmd = launcher + script + args + ds_args
# keep for quick debug
@ -641,3 +638,11 @@ class TestDeepSpeedWithLauncher(TestCasePlus):
execute_subprocess_async(cmd, env=self.get_env())
return output_dir
def get_launcher(self, distributed=False):
# 1. explicitly set --num_nodes=1 just in case these tests end up run on a multi-node setup
# - it won't be able to handle that
# 2. for now testing with just 2 gpus max (since some quality tests may give different
# results with mode gpus because we use very little data)
num_gpus = min(2, get_gpu_count()) if distributed else 1
return f"deepspeed --num_nodes 1 --num_gpus {num_gpus}".split()