Support page size > 1 (#4356)
This commit is contained in:
@@ -7,13 +7,13 @@ from typing import List, Optional
|
||||
import torch
|
||||
|
||||
from sglang.srt.managers.cache_controller import HiCacheController
|
||||
from sglang.srt.managers.schedule_batch import Req
|
||||
from sglang.srt.mem_cache.memory_pool import (
|
||||
MHATokenToKVPoolHost,
|
||||
ReqToTokenPool,
|
||||
TokenToKVPoolAllocator,
|
||||
)
|
||||
from sglang.srt.mem_cache.radix_cache import RadixCache, TreeNode, _key_match
|
||||
from sglang.srt.mem_cache.radix_cache import RadixCache, TreeNode
|
||||
from sglang.srt.mem_cache.radix_cache import _key_match_page_size1 as _key_match
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
@@ -122,7 +122,7 @@ class HiRadixCache(RadixCache):
|
||||
def evictable_size(self):
|
||||
return self.evictable_size_
|
||||
|
||||
def evict(self, num_tokens: int, evict_callback=None):
|
||||
def evict(self, num_tokens: int):
|
||||
leaves = self._collect_leaves_device()
|
||||
heapq.heapify(leaves)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user