Skip to content

MegatronRun: fix .load test #866

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

Draft
podkidyshev wants to merge 3 commits into
base: main
Choose a base branch
from
+31 −32
Draft

MegatronRun: fix .load test #866

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f42136a
redefined is_run_successful function
podkidyshev Apr 10, 2026
dcf4199
adjusted was_run_successful
podkidyshev Apr 10, 2026
0ca6562
better is_run_successfull for megarun
podkidyshev Apr 13, 2026
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
6 changes: 4 additions & 2 deletions src/cloudai/workloads/megatron_run/megatron_run.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -145,11 +145,11 @@ def was_run_successful(self, tr: TestRun) -> JobStatusResult:
with slurm_job_path.open("r", encoding="utf-8") as file:
metadata = SlurmJobMetadata.model_validate(toml.load(file))

if not metadata.state.startswith("COMPLETED") or not metadata.exit_code.startswith("0:"):
if not metadata.exit_code.startswith("0:"):
return JobStatusResult(
is_successful=False,
error_message=(
f"Slurm job did not complete successfully for {tr.output_path}: "
f"Slurm job exited with a non-zero exit code for {tr.output_path}: "
f"state={metadata.state}, exit_code={metadata.exit_code}."
),
)
Expand All @@ -165,6 +165,8 @@ def was_run_successful(self, tr: TestRun) -> JobStatusResult:
for line in file:
if ITERATION_LOG_REGEX.search(line):
return JobStatusResult(is_successful=True)
if "validation loss at iteration" in line:
return JobStatusResult(is_successful=True)

return JobStatusResult(
is_successful=False,
Expand Down
57 changes: 27 additions & 30 deletions ...run/test_job_status_retrieval_strategy.py → ...rkloads/megatron_run/test_megatron_run.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@

from pathlib import Path

import pytest
import toml

from cloudai import TestRun
Expand All @@ -32,13 +33,13 @@ def setup_method(self) -> None:
cmd_args=MegatronRunCmdArgs(docker_image_url="http://url", run_script=Path(__file__)),
)

def _write_slurm_metadata(self, output_path: Path, *, state: str, exit_code: str = "0:0") -> None:
def _write_slurm_metadata(self, output_path: Path, *, exit_code: str = "0:0") -> None:
with (output_path / "slurm-job.toml").open("w", encoding="utf-8") as file:
toml.dump(
SlurmJobMetadata(
job_id=123,
name="megatron",
state=state,
state="WHATEVER",
exit_code=exit_code,
start_time="2026-03-22T11:44:22",
end_time="2026-03-22T11:54:22",
Expand All @@ -56,34 +57,30 @@ def test_missing_slurm_metadata_fails(self, base_tr: TestRun) -> None:
assert not result.is_successful
assert "slurm-job.toml file not found" in result.error_message

def test_failed_slurm_state_fails_even_if_stdout_has_metrics(self, base_tr: TestRun) -> None:
@pytest.mark.parametrize(
("exit_code", "log_text", "is_successful"),
(
("0:0", "bla", False),
(
"1:0",
"[2026-01-16 07:32:39] iteration 6/100 | elapsed time per iteration (ms): 15639.0 | "
"throughput per GPU (TFLOP/s/GPU): 494.6 |\n",
False,
),
(
"0:0",
"[2026-01-16 07:32:39] iteration 6/100 | elapsed time per iteration (ms): 15639.0 | "
"throughput per GPU (TFLOP/s/GPU): 494.6 |\n",
True,
),
("0:0", "validation loss at iteration 1.0", True),
("15:0", "validation loss at iteration 1.0", False),
),
)
def test_is_run_successful(self, base_tr: TestRun, exit_code: str, log_text: str, is_successful: bool) -> None:
base_tr.output_path.mkdir(parents=True, exist_ok=True)
self._write_slurm_metadata(base_tr.output_path, state="FAILED", exit_code="1:0")
(base_tr.output_path / "stdout.txt").write_text(
"[2026-01-16 07:32:39] iteration 6/100 | elapsed time per iteration (ms): 15639.0 | "
"throughput per GPU (TFLOP/s/GPU): 494.6 |\n"
)

result = self.megatron_tdef.was_run_successful(base_tr)
assert not result.is_successful
assert "state=FAILED" in result.error_message

def test_completed_slurm_job_with_iteration_metrics_succeeds(self, base_tr: TestRun) -> None:
base_tr.output_path.mkdir(parents=True, exist_ok=True)
self._write_slurm_metadata(base_tr.output_path, state="COMPLETED")
(base_tr.output_path / "stdout.txt").write_text(
"[2026-01-16 07:32:39] iteration 6/100 | elapsed time per iteration (ms): 15639.0 | "
"throughput per GPU (TFLOP/s/GPU): 494.6 |\n"
)

result = self.megatron_tdef.was_run_successful(base_tr)
assert result.is_successful

def test_completed_slurm_job_without_iteration_metrics_fails(self, base_tr: TestRun) -> None:
base_tr.output_path.mkdir(parents=True, exist_ok=True)
self._write_slurm_metadata(base_tr.output_path, state="COMPLETED")
(base_tr.output_path / "stdout.txt").write_text("training started\n")
self._write_slurm_metadata(base_tr.output_path, exit_code=exit_code)
(base_tr.output_path / "stdout.txt").write_text(log_text)

result = self.megatron_tdef.was_run_successful(base_tr)
assert not result.is_successful
assert "does not contain Megatron iteration metrics" in result.error_message
assert result.is_successful is is_successful
Loading