This commit is contained in:
Ryuichi Leo Takashige
2026-01-23 15:00:11 +00:00
parent 457debc338
commit 8798ab52ee
2 changed files with 3 additions and 3 deletions
@@ -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
+1 -1
View File
@@ -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(