Add kvcache-centric profiling and admission controls
This commit is contained in:
@@ -64,6 +64,7 @@ def launch_pd_stack(
|
||||
prefill_policy: str,
|
||||
decode_policy: str,
|
||||
timeout_s: float = 1200.0,
|
||||
router_request_timeout_s: float | None = None,
|
||||
include_router: bool = True,
|
||||
) -> ManagedPdStack:
|
||||
run_dir.mkdir(parents=True, exist_ok=True)
|
||||
@@ -75,6 +76,7 @@ def launch_pd_stack(
|
||||
prefill_policy=prefill_policy,
|
||||
decode_policy=decode_policy,
|
||||
include_router=include_router,
|
||||
router_request_timeout_s=router_request_timeout_s,
|
||||
)
|
||||
|
||||
prefill_processes = [
|
||||
@@ -186,7 +188,7 @@ def _build_process_env(topology: SingleNodeTopology) -> dict[str, str]:
|
||||
env["NO_PROXY"] = "*"
|
||||
env["no_proxy"] = "*"
|
||||
env.setdefault("SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT", "600")
|
||||
env.setdefault("SGLANG_DISAGGREGATION_WAITING_TIMEOUT", "60")
|
||||
env.setdefault("SGLANG_DISAGGREGATION_WAITING_TIMEOUT", "600")
|
||||
env.setdefault("FLASHINFER_DISABLE_VERSION_CHECK", "1")
|
||||
if topology.force_rdma:
|
||||
env["MOONCAKE_PROTOCOL"] = "rdma"
|
||||
|
||||
Reference in New Issue
Block a user