EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[Hoxfix] Fix incomplete token_to_kv_pool refactor (#4121)

Browse Source
This commit is contained in:
Wenxuan Tan
2025-03-05 21:32:42 -06:00
committed by GitHub
parent fc671f66c1
commit 718c391fd7
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/srt/managers/scheduler.py
View File

@@ -484,7 +484,7 @@ class Scheduler:
logger.error( logger.error(
f"{self.cur_batch.batch_size()=}, " f"{self.cur_batch.batch_size()=}, "
f"{self.cur_batch.reqs=}, " f"{self.cur_batch.reqs=}, "
f"{self.token_to_kv_pool.available_size()=}, " f"{self.token_to_kv_pool_allocator.available_size()=}, "
f"{self.tree_cache.evictable_size()=}, " f"{self.tree_cache.evictable_size()=}, "
) )
# Wait for some time so that the parent process can print the error. # Wait for some time so that the parent process can print the error.
@@ -932,7 +932,7 @@ class Scheduler:
): ):
# During idle time, also collect metrics every 30 seconds. # During idle time, also collect metrics every 30 seconds.
num_used = self.max_total_num_tokens - ( num_used = self.max_total_num_tokens - (
self.token_to_kv_pool.available_size() self.token_to_kv_pool_allocator.available_size()
+ self.tree_cache.evictable_size() + self.tree_cache.evictable_size()
) )
num_running_reqs = len(self.running_batch.reqs) if self.running_batch else 0 num_running_reqs = len(self.running_batch.reqs) if self.running_batch else 0