chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Multi-Stage Awake: Support Resume and Pause KV Cache and Weights separately (#7099)

Browse Source
This commit is contained in:
Stefan He
2025-06-19 00:56:37 -07:00
committed by GitHub
parent 9179ea1595
commit 3774f07825
14 changed files with 297 additions and 108 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/sglang/srt/managers/io_struct.py
View File

@@ -812,7 +812,9 @@ class GetWeightsByNameReqOutput:
@dataclass
class ReleaseMemoryOccupationReqInput:
pass
# Optional tags to identify the memory region, which is primarily used for RL
# Currently we only support `weights` and `kv_cache`
tags: Optional[List[str]] = None
@dataclass
@@ -822,7 +824,9 @@ class ReleaseMemoryOccupationReqOutput:
@dataclass
class ResumeMemoryOccupationReqInput:
pass
# Optional tags to identify the memory region, which is primarily used for RL
# Currently we only support `weights` and `kv_cache`
tags: Optional[List[str]] = None
@dataclass