diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml index 13b9e96ca..3e7ac15a0 100644 --- a/.github/workflows/build.yaml +++ b/.github/workflows/build.yaml @@ -5,6 +5,7 @@ on: push: branches: - 'main' + - 'optimizations' tags: - 'v*' @@ -69,10 +70,7 @@ jobs: images: | ghcr.io/predibase/lorax tags: | - type=semver,pattern={{version}} - type=semver,pattern={{major}}.{{minor}} - type=sha,prefix=,suffix=,format=short - type=raw,value=main,enable=${{ github.ref == 'refs/heads/main' }} + type=raw,value=optimizations-5,enable=${{ github.ref == 'refs/heads/optimizations' }} - name: Create a hash from tags env: diff --git a/router/src/batch.rs b/router/src/batch.rs index 2363dbdf6..39f8d32c7 100644 --- a/router/src/batch.rs +++ b/router/src/batch.rs @@ -222,7 +222,6 @@ impl BatchEntriesState { // TODO(travis): clone is not ideal, find a way to do this cleanly in place for r in self.batch_requests.clone().into_iter().rev() { let id = r.id; - tracing::info!("!!! drain::remove entry id={id:?}"); let entry = self.batch_entries.remove(&id).unwrap(); let adapter_index = r.adapter_index; let adapter = self.index_to_adapter.get_mut(&adapter_index).unwrap(); diff --git a/router/src/infer.rs b/router/src/infer.rs index b3b608dc3..0a4c4e5cc 100644 --- a/router/src/infer.rs +++ b/router/src/infer.rs @@ -1323,7 +1323,6 @@ fn filter_send_generations(generations: Vec, entries: &mut IntMap, entries: &mut IntMap "dropped"); }).unwrap_or(true); if stopped { - tracing::info!("!!! filter_send_generations::remove entry id={id:?}"); entries.remove(&id).expect("ID not found in entries. This is a bug."); } }); @@ -1351,7 +1349,6 @@ fn send_responses( ) -> Result>>> { // Return directly if the channel is disconnected if entry.response_tx.is_closed() { - tracing::info!("!!! send_responses::disconnected"); return Ok(true); } @@ -1410,9 +1407,6 @@ fn send_responses( match (&generation.generated_text, iterator.peek()) { (Some(generated_text), None) => { - tracing::info!( - "!!! send_responses::generation_ended id={id:?} generated_text={generated_text:?}" - ); // Generation has ended stopped = true; // Send message