Update examples/server/utils.hpp

ngxson · web-flow · commit 836015d451b5 · 2025-04-29T16:37:21.000+02:00
diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp
@@ -642,7 +642,8 @@ static json oaicompat_completion_params_parse(
         throw std::runtime_error("Cannot use custom grammar constraints with tools.");
     }
 
-    /* Prefill assistant message support */
+    // if the assistant message appears at the end of list, we do not add end-of-turn token
+    // for ex. this can be useful to modify the reasoning process in reasoning models
     bool prefill_assistant_message = !inputs.messages.empty() && inputs.messages.back().role == "assistant";
     common_chat_msg last_message;
     if (prefill_assistant_message) {

Original file line number	Diff line number	Diff line change
`@@ -642,7 +642,8 @@ static json oaicompat_completion_params_parse(`
`642`	`642`	`throw std::runtime_error("Cannot use custom grammar constraints with tools.");`
`643`	`643`	`}`
`644`	`644`
`645`		`- /* Prefill assistant message support */`
	`645`	`+ // if the assistant message appears at the end of list, we do not add end-of-turn token`
	`646`	`+ // for ex. this can be useful to modify the reasoning process in reasoning models`
`646`	`647`	`bool prefill_assistant_message = !inputs.messages.empty() && inputs.messages.back().role == "assistant";`
`647`	`648`	`common_chat_msg last_message;`
`648`	`649`	`if (prefill_assistant_message) {`