[Env] centralize pd vars in environ.py (#16264)
This commit is contained in:
@@ -221,6 +221,15 @@ class Envs:
|
||||
SGLANG_SCHEDULER_RECV_SKIPPER_WEIGHT_TARGET_VERIFY = EnvInt(1)
|
||||
SGLANG_SCHEDULER_RECV_SKIPPER_WEIGHT_NONE = EnvInt(1)
|
||||
|
||||
# PD Disaggregation (runtime)
|
||||
# NOTE: For SGLANG_DISAGGREGATION_THREAD_POOL_SIZE, the effective default is
|
||||
# computed dynamically at runtime based on cpu_count; see disaggregation backends.
|
||||
SGLANG_DISAGGREGATION_THREAD_POOL_SIZE = EnvInt(None)
|
||||
SGLANG_DISAGGREGATION_QUEUE_SIZE = EnvInt(4)
|
||||
SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT = EnvInt(300)
|
||||
SGLANG_DISAGGREGATION_HEARTBEAT_INTERVAL = EnvFloat(5.0)
|
||||
SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE = EnvInt(2)
|
||||
SGLANG_DISAGGREGATION_WAITING_TIMEOUT = EnvInt(300)
|
||||
|
||||
# Scheduler: others:
|
||||
SGLANG_EMPTY_CACHE_INTERVAL = EnvFloat(-1) # in seconds. Set if you observe high memory accumulation over a long serving period.
|
||||
|
||||
Reference in New Issue
Block a user