Support token low usage watermark in prefill delayer (#16814)
This commit is contained in:
@@ -224,6 +224,7 @@ class Envs:
|
||||
SGLANG_SCHEDULER_SKIP_ALL_GATHER = EnvBool(False)
|
||||
SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE = EnvBool(False)
|
||||
SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES = EnvInt(30)
|
||||
SGLANG_PREFILL_DELAYER_TOKEN_USAGE_LOW_WATERMARK = EnvFloat(None)
|
||||
SGLANG_DATA_PARALLEL_BUDGET_INTERVAL = EnvInt(1)
|
||||
|
||||
# Test: pd-disaggregation
|
||||
|
||||
Reference in New Issue
Block a user