change resuming policy TODO to right pos

WukLab · Aug 15, 2023 · 5eb4f55 · 5eb4f55
1 parent 7041160
commit 5eb4f55
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 5 deletions.
diff --git a/vllm/core/scheduler.py b/vllm/core/scheduler.py
@@ -102,7 +102,10 @@ def get_paused_seq_group(self, request_id: str) \
         else:
             raise ValueError(f"No paused request with request id {request_id}")
 
-    # TODO: add logic for other two modes
+    # TODO: (Need discussion) different policy of resuming APIs:
+    #  1. (implememted) API with preserve mode: FCFS in the running group
+    #  2. (?) API with recompute mode: First in the waiting group
+    #  3. (?) API with swap mode: FCFS in the swapped group
     def resume_seq_group(self, request_id: str) -> None:
         seq_group, mode = self.get_paused_seq_group(request_id)
         del self.paused[request_id]

diff --git a/vllm/engine/llm_engine.py b/vllm/engine/llm_engine.py
@@ -276,10 +276,6 @@ def resume_request(self,
         seq_group.clear_and_set(seqs)
         self.scheduler.resume_seq_group(request_id)
 
-    # TODO: (Need discussion) different policy of resuming APIs, currently is 
-    #  1. API with preserve mode: FCFS in the running group
-    #  2. (?) API with recompute mode: First in the waiting group
-    #  3. (?) API with swap mode: FCFS in the swapped group
     # NOTE: seq_id_2_api_rel is to support multiple API calls in a SeqGroup
     # need mapping seq_id -> API tokens to append result to the right seq
     def new_resume_request(self, request_id: str,