diff --git a/src/exo/worker/runner/batched_scoring_handler.py b/src/exo/worker/runner/batched_scoring_handler.py index 5d435909..c8942e4a 100644 --- a/src/exo/worker/runner/batched_scoring_handler.py +++ b/src/exo/worker/runner/batched_scoring_handler.py @@ -39,8 +39,8 @@ class BatchedScoringHandler: tokenizer: TokenizerWrapper model_id: ModelId device_rank: int - max_batch_size: int = 32 - batch_timeout_ms: int = 10 + max_batch_size: int = 128 + batch_timeout_ms: int = 20 pending: list[PendingScoringRequest] = field(default_factory=list) pending_start_time: float | None = None diff --git a/src/exo/worker/runner/runner.py b/src/exo/worker/runner/runner.py index a7925722..2ee2b08b 100644 --- a/src/exo/worker/runner/runner.py +++ b/src/exo/worker/runner/runner.py @@ -89,7 +89,7 @@ from exo.worker.runner.bootstrap import logger # Batching configuration BATCH_ENABLED = True -BATCH_MAX_SIZE = 64 +BATCH_MAX_SIZE = 128 def _should_use_serial_processing(