Enhance runtime memory check in CI (#15192)
This commit is contained in:
@@ -5,6 +5,7 @@ import signal
|
||||
import sys
|
||||
import threading
|
||||
import time
|
||||
import warnings
|
||||
from typing import TYPE_CHECKING
|
||||
|
||||
import psutil
|
||||
@@ -175,6 +176,13 @@ class SchedulerRuntimeCheckerMixin:
|
||||
if current_batch is None:
|
||||
return
|
||||
|
||||
spec_topk = self.server_args.speculative_eagle_topk or 1
|
||||
if spec_topk > 1:
|
||||
warnings.warn(
|
||||
"Runtime memory check (busy) is not supported when speculation topk > 1."
|
||||
)
|
||||
return
|
||||
|
||||
_, _, available_size, evictable_size = self._get_token_info()
|
||||
protected_size = self.tree_cache.protected_size()
|
||||
|
||||
|
||||
Reference in New Issue
Block a user