Skip to content

fix(trainer): Keep the original runtime command in get_runtime_packages() API #64

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 19, 2025
+8 −4
Merged

fix(trainer): Keep the original runtime command in get_runtime_packages() API #64

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 8 additions & 4 deletions kubeflow/trainer/api/trainer_client.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.

import copy
import logging
import multiprocessing
import queue
Expand Down Expand Up @@ -177,11 +178,14 @@ def get_runtime_packages(self, runtime: types.Runtime):
if runtime.trainer.trainer_type == types.TrainerType.BUILTIN_TRAINER:
raise ValueError("Cannot get Runtime packages for BuiltinTrainer")

# Create a deepcopy of the runtime to avoid modifying the original command.
runtime_copy = copy.deepcopy(runtime)

# Run mpirun only within the single process.
if runtime.trainer.command[0] == "mpirun":
if runtime_copy.trainer.command[0] == "mpirun":
mpi_command = list(constants.MPI_COMMAND)
mpi_command[1:3] = ["-np", "1"]
runtime.trainer.set_command(tuple(mpi_command))
runtime_copy.trainer.set_command(tuple(mpi_command))

def print_packages():
import subprocess
Expand Down Expand Up @@ -211,12 +215,12 @@ def print_packages():
# Create the TrainJob and wait until it completes.
# If Runtime trainer has GPU resources use them, otherwise run TrainJob with 1 CPU.
job_name = self.train(
runtime=runtime,
runtime=runtime_copy,
trainer=types.CustomTrainer(
func=print_packages,
num_nodes=1,
resources_per_node=(
{"cpu": 1} if runtime.trainer.device != "gpu" else None
{"cpu": 1} if runtime_copy.trainer.device != "gpu" else None
),
),
)
Expand Down