more log improvements

crates/common/src/tokenizer.rs

@@ -1,4 +1,4 @@
-use log::debug;
+use log::trace;
 
 #[derive(thiserror::Error, Debug, PartialEq, Eq)]
 #[allow(dead_code)]
@@ -9,7 +9,7 @@ pub enum Error {
 
 #[allow(dead_code)]
 pub fn token_count(model_name: &str, text: &str) -> Result<usize, Error> {
-    debug!("getting token count model={}", model_name);
+    trace!("getting token count model={}", model_name);
     // Consideration: is it more expensive to instantiate the BPE object every time, or to contend the singleton?
     let bpe = tiktoken_rs::get_bpe_from_model(model_name).map_err(|_| Error::UnknownModel {
         model_name: model_name.to_string(),

crates/llm_gateway/src/stream_context.rs

@@ -406,9 +406,10 @@ impl HttpContext for StreamContext {
         let body = if self.streaming_response {
             let chunk_start = 0;
             let chunk_size = body_size;
-            debug!(
+            trace!(
                 "streaming response reading, {}..{}",
-                chunk_start, chunk_size
+                chunk_start,
+                chunk_size
             );
             let streaming_chunk = match self.get_http_response_body(0, chunk_size) {
                 Some(chunk) => chunk,
@@ -531,9 +532,11 @@ impl HttpContext for StreamContext {
             }
         }
 
-        debug!(
+        trace!(
             "recv [S={}] total_tokens={} end_stream={}",
-            self.context_id, self.response_tokens, end_of_stream
+            self.context_id,
+            self.response_tokens,
+            end_of_stream
         );
 
         Action::Continue