[feature] [sgl-router] Add a dp-aware routing strategy (#6869)

This commit is contained in:
Rui Chen
2025-07-30 20:58:48 +08:00
committed by GitHub
parent 55ecdc0a8e
commit a730ce8162
19 changed files with 726 additions and 16 deletions

View File

@@ -31,6 +31,10 @@ class Router:
routing. Default: 60
max_payload_size: Maximum payload size in bytes. Default: 256MB
max_tree_size: Maximum size of the approximation tree for cache-aware routing. Default: 2^24
dp_aware: Enable data parallelism aware schedule. Default: False
api_key: The api key used for the authorization with the worker.
Useful when the dp aware scheduling strategy is enabled.
Default: None
log_dir: Directory to store log files. If None, logs are only output to console. Default: None
log_level: Logging level. Options: 'debug', 'info', 'warning', 'error', 'critical'.
service_discovery: Enable Kubernetes service discovery. When enabled, the router will
@@ -73,6 +77,8 @@ class Router:
eviction_interval_secs: int = 60,
max_tree_size: int = 2**24,
max_payload_size: int = 256 * 1024 * 1024, # 256MB
dp_aware: bool = False,
api_key: Optional[str] = None,
log_dir: Optional[str] = None,
log_level: Optional[str] = None,
service_discovery: bool = False,
@@ -110,6 +116,8 @@ class Router:
eviction_interval_secs=eviction_interval_secs,
max_tree_size=max_tree_size,
max_payload_size=max_payload_size,
dp_aware=dp_aware,
api_key=api_key,
log_dir=log_dir,
log_level=log_level,
service_discovery=service_discovery,