fix input_penalty token_id async update bug. (#1022)

hiworldwzj · web-flow · commit a7714121c264 · 2025-08-25T18:30:57.000+08:00
diff --git a/lightllm/common/req_manager.py b/lightllm/common/req_manager.py
@@ -159,6 +159,7 @@ def init_req_sampling_params(self, req):
                 token_id_counter(
                     prompt_ids=prompt_ids, out_token_id_counter=self.req_to_out_token_id_counter[req.req_idx]
                 )
+                torch.cuda.current_stream().synchronize()
 
         return
 

Original file line number	Diff line number	Diff line change
`@@ -159,6 +159,7 @@ def init_req_sampling_params(self, req):`
`159`	`159`	`token_id_counter(`
`160`	`160`	`prompt_ids=prompt_ids, out_token_id_counter=self.req_to_out_token_id_counter[req.req_idx]`
`161`	`161`	`)`
	`162`	`+ torch.cuda.current_stream().synchronize()`
`162`	`163`
`163`	`164`	`return`
`164`	`165`