diff --git a/router/src/lib.rs b/router/src/lib.rs index 2395e3e2..17ab00e7 100644 --- a/router/src/lib.rs +++ b/router/src/lib.rs @@ -709,6 +709,10 @@ pub(crate) struct ChatRequest { #[schema(nullable = true, example = "null")] #[serde(deserialize_with = "deserialize_tool_choice::deserialize")] pub tool_choice: Option, + + #[serde(default)] + #[schema(default = "false", example = true)] + pub do_sample: bool, } fn default_tool_prompt() -> Option { diff --git a/router/src/server.rs b/router/src/server.rs index 302a4753..e728455d 100644 --- a/router/src/server.rs +++ b/router/src/server.rs @@ -1060,7 +1060,7 @@ async fn chat_completions( top_k: None, top_p: req.top_p, typical_p: None, - do_sample: true, + do_sample: req.do_sample, max_new_tokens, return_full_text: None, stop,