add more changes

This commit is contained in:
Adil Hafeez 2025-02-07 19:01:42 -08:00
parent cd90faf50c
commit 66a971b086
No known key found for this signature in database
GPG key ID: 9B18EF7691369645
8 changed files with 35 additions and 9 deletions

View file

@ -25,6 +25,7 @@ pub struct Configuration {
#[derive(Debug, Clone, Serialize, Deserialize, Default)]
pub struct Overrides {
pub prompt_target_intent_matching_threshold: Option<f64>,
pub optimize_context_window: Option<bool>,
}
#[derive(Debug, Clone, Serialize, Deserialize, Default)]

View file

@ -137,9 +137,20 @@ impl HttpContext for StreamContext {
.map(|(_, pt)| pt.into())
.collect();
let mut metadata = deserialized_body.metadata.clone();
if let Some(overrides) = self.overrides.as_ref() {
if overrides.optimize_context_window.unwrap_or_default() {
if metadata.is_none() {
metadata = Some(HashMap::new());
}
metadata.as_mut().unwrap().insert("optimize_context_window".to_string(), "true".to_string());
}
}
let arch_fc_chat_completion_request = ChatCompletionsRequest {
messages: deserialized_body.messages.clone(),
metadata: deserialized_body.metadata.clone(),
metadata,
stream: deserialized_body.stream,
model: "--".to_string(),
stream_options: deserialized_body.stream_options.clone(),
@ -157,7 +168,7 @@ impl HttpContext for StreamContext {
};
debug!("sending request to model server");
trace!("request body: {}", json_data);
debug!("request body: {}", json_data);
let mut headers = vec![
(ARCH_UPSTREAM_HOST_HEADER, MODEL_SERVER_NAME),

View file

@ -46,7 +46,7 @@ pub struct StreamCallContext {
pub struct StreamContext {
system_prompt: Rc<Option<String>>,
pub prompt_targets: Rc<HashMap<String, PromptTarget>>,
_overrides: Rc<Option<Overrides>>,
pub overrides: Rc<Option<Overrides>>,
pub metrics: Rc<Metrics>,
pub callouts: RefCell<HashMap<u32, StreamCallContext>>,
pub context_id: u32,
@ -89,7 +89,7 @@ impl StreamContext {
streaming_response: false,
user_prompt: None,
is_chat_completions_request: false,
_overrides: overrides,
overrides: overrides,
request_id: None,
traceparent: None,
_tracing: tracing,