Skip to content

execute runs in parallel when possible #117

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
dphuang2 merged 1 commit into from
Aug 22, 2025
Merged

execute runs in parallel when possible #117

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
17 changes: 16 additions & 1 deletion eval_protocol/pytest/evaluation_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,7 @@
Status,
)
from eval_protocol.pytest.default_dataset_adapter import default_dataset_adapter
from eval_protocol.pytest.default_mcp_gym_rollout_processor import MCPGymRolloutProcessor
from eval_protocol.pytest.default_no_op_rollout_processor import NoOpRolloutProcessor
from eval_protocol.pytest.rollout_processor import RolloutProcessor
from eval_protocol.pytest.types import (
Expand Down Expand Up @@ -585,7 +586,9 @@ def _log_eval_error(status: Status, rows: Optional[List[EvaluationRow]] | None,
exception_handler_config=exception_handler_config,
)

for i in range(num_runs):
async def execute_run(i: int, config: RolloutProcessorConfig):
nonlocal all_results

# Regenerate outputs each run by deep-copying the pristine dataset
# so model responses are not reused across runs.
run_id = generate_id()
Expand Down Expand Up @@ -728,6 +731,18 @@ async def _collect_result(config, lst):
r.eval_metadata.status = Status.eval_finished()
active_logger.log(r)

tasks = []
for i in range(num_runs):
tasks.append(asyncio.create_task(execute_run(i, config)))

# if rollout_processor is McpGymRolloutProcessor, we execute runs sequentially since McpGym does not support concurrent runs
Copy link
Copy Markdown
Collaborator

@mayinghan mayinghan Aug 26, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why would this matter? i thought each task has its own mcp session id

# else, we execute runs in parallel
if isinstance(rollout_processor, MCPGymRolloutProcessor):
for task in tasks:
await task
else:
await asyncio.gather(*tasks)

# for groupwise mode, the result contains eval otuput from multiple completion_params, we need to differentiate them
# rollout_id is used to differentiate the result from different completion_params
if mode == "groupwise":
Expand Down
Loading