Skip to content

feat(swe-bench): bind each unit to an endpoint by shard index #477

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

Closed
arekay-nv wants to merge 1 commit into from
Closed

feat(swe-bench): bind each unit to an endpoint by shard index #477

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: 9 additions & 3 deletions src/inference_endpoint/evaluation/swe_bench_fleet_scorer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -326,11 +326,17 @@ def _instance_ids(self) -> list[str]:
]

def _submit_unit(self, service_url: str, unit: Unit) -> str:
# The service accepts exactly one endpoint URL per run, so a unit is
# bound to exactly one endpoint. Sending every unit to endpoint 0 would
# funnel the whole fleet through a single engine while the rest idle,
# which is both a throughput ceiling and a measurement hazard: one
# engine's behaviour would decide the entire run's accuracy. Binding by
# shard index spreads units deterministically -- the same unit always
# gets the same endpoint, so a retry is comparable to its first attempt.
endpoint = self._endpoint_urls[unit.shard % len(self._endpoint_urls)]
payload = {
"model_name": self._model_name,
# The service accepts exactly one endpoint URL per run; the fleet's
# parallelism comes from running many units, not many endpoints.
"endpoint_urls": self._endpoint_urls[:1],
"endpoint_urls": [endpoint],
"endpoint_api_key": self._endpoint_api_key,
"generation_params": self._generation_params,
"subset": self.options["subset"],
Expand Down
Loading