fix max_new_token.

Weichao Luo · Weichao Luo · commit 046db94208b8 · 2025-05-20T15:01:54.000+08:00
diff --git a/lightllm/server/httpserver_for_pd_master/manager.py b/lightllm/server/httpserver_for_pd_master/manager.py
@@ -272,7 +272,7 @@ async def fetch_stream_nixl(
             "node_id": p_start_args["pd_node_id"],
             "ip": p_start_args["host"],
             "rpyc_port": p_start_args["pd_remote_prefill_port"],
-            "max_new_tokens": sampling_params.max_new_tokens - 1,
+            "max_new_tokens": sampling_params.max_new_tokens,
             "pd_master_node_id": self.args.pd_node_id,
         }
 

Original file line number	Diff line number	Diff line change
`@@ -272,7 +272,7 @@ async def fetch_stream_nixl(`
`272`	`272`	`"node_id": p_start_args["pd_node_id"],`
`273`	`273`	`"ip": p_start_args["host"],`
`274`	`274`	`"rpyc_port": p_start_args["pd_remote_prefill_port"],`
`275`		`- "max_new_tokens": sampling_params.max_new_tokens - 1,`
	`275`	`+ "max_new_tokens": sampling_params.max_new_tokens,`
`276`	`276`	`"pd_master_node_id": self.args.pd_node_id,`
`277`	`277`	`}`
`278`	`278`