Distributed/ND-Parallel (#2977)
This commit is contained in:
@@ -2,7 +2,7 @@
|
||||
set -e
|
||||
|
||||
# Only run two tests at a time to avoid OOM on GPU (with coverage collection)
|
||||
pytest -v -n2 \
|
||||
pytest -v --durations=10 -n2 \
|
||||
--ignore=/workspace/axolotl/tests/e2e/multigpu/solo/ \
|
||||
--ignore=/workspace/axolotl/tests/e2e/multigpu/patched/ \
|
||||
/workspace/axolotl/tests/e2e/multigpu/ \
|
||||
|
||||
@@ -65,6 +65,9 @@ GPU_CONFIG = f"L40S:{N_GPUS}"
|
||||
def run_cmd(cmd: str, run_folder: str):
|
||||
import subprocess # nosec
|
||||
|
||||
sp_env = os.environ.copy()
|
||||
sp_env["AXOLOTL_DATASET_PROCESSES"] = "8"
|
||||
|
||||
# Propagate errors from subprocess.
|
||||
if exit_code := subprocess.call(cmd.split(), cwd=run_folder): # nosec
|
||||
if exit_code := subprocess.call(cmd.split(), cwd=run_folder, env=sp_env): # nosec
|
||||
exit(exit_code) # pylint: disable=consider-using-sys-exit
|
||||
|
||||
Reference in New Issue
Block a user