chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

Fix: Complete int32 to int64 conversion (#4465)

Browse Source
This commit is contained in:
Zhiqiang Xie
2025-03-16 18:14:27 -07:00
committed by GitHub
parent 5cbd709ea1
commit f5bbf6037d
3 changed files with 5 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
python/sglang/srt/mem_cache/memory_pool.py
View File

@@ -622,11 +622,10 @@ class HostKVCache(abc.ABC):
self.mem_state = torch.zeros(
(self.size,), dtype=torch.uint8, device=self.device
)
self.free_slots = torch.arange(self.size, dtype=torch.int32)
self.can_use_mem_size = self.size
# A lock for synchronized operations on memory allocation and state transitions.
self.lock = threading.RLock()
self.clear()
@abc.abstractmethod
def get_size_per_token(self):
@@ -656,7 +655,7 @@ class HostKVCache(abc.ABC):
def clear(self):
self.mem_state.fill_(0)
self.can_use_mem_size = self.size
self.free_slots = torch.arange(self.size, dtype=torch.int32)
self.free_slots = torch.arange(self.size, dtype=torch.int64)
@synchronized
def get_state(self, indices: torch.Tensor) -> MemoryStateInt: