mirror of
https://github.com/radixark/miles.git
synced 2026-10-02 07:14:53 +08:00
P6 round 1 / P6-26: Use the transfer selector for P2P checksum verification
Consolidate the raw fixes without changing their final tree. Raw commits: - bc2e74eefbcca7ab18b5950efcfac89df8211014
This commit is contained in:
@@ -161,7 +161,7 @@ class UpdateWeightP2P(WeightTransferProtocol):
|
||||
self.disconnect()
|
||||
self.rollout_engines = rollout_engines
|
||||
self.cell_updaters_of_cell_id = {
|
||||
cell_id: _P2PRolloutCellUpdater(args=self.args, cell_id=cell_id, api_client=api_client)
|
||||
cell_id: _P2PRolloutCellUpdater(args=self.args, cell_id=cell_id, api_client=api_client, selector=selector)
|
||||
for api_client, cell_id in zip(rollout_engines, engine_cell_ids, strict=True)
|
||||
}
|
||||
|
||||
|
||||
@@ -21,8 +21,10 @@ class _P2PRolloutCellUpdater(_RolloutCellUpdater):
|
||||
args: Namespace,
|
||||
cell_id: str,
|
||||
api_client: SGLangApiClient,
|
||||
selector: str,
|
||||
) -> None:
|
||||
super().__init__(args=args, cell_id=cell_id, api_client=api_client)
|
||||
self.selector = selector
|
||||
self._executor = ThreadPoolExecutor(max_workers=1)
|
||||
self.targets_by_rollout_engine_rank: dict[int, RemoteWeightInfo] = {}
|
||||
self._pending_writes: list[Future[None]] = []
|
||||
@@ -137,7 +139,9 @@ def _verify_transfer_checksums(
|
||||
names: list[str],
|
||||
sent_checksums: dict[str, str],
|
||||
) -> None:
|
||||
engine_body = cell_updater.submit_client_call("check_weights", action="raw_checksum", names=names).result()
|
||||
engine_body = cell_updater.submit_client_call(
|
||||
"check_weights", action="raw_checksum", names=names, selector=cell_updater.selector
|
||||
).result()
|
||||
if engine_body is None:
|
||||
return
|
||||
checksum_utils.verify_transfer_checksums(
|
||||
|
||||
Reference in New Issue
Block a user