refactor(llm): Rename EngineConfig::Static to InProcess (#4585)

Signed-off-by: Graham King <grahamk@nvidia.com>

Author: grahamking

launch/dynamo-run/src/lib.rs

@@ -148,12 +148,12 @@ async fn engine_for(
             // Auto-discover backends
             Ok(EngineConfig::Dynamic(Box::new(local_model)))
         }
-        Output::Echo => Ok(EngineConfig::StaticFull {
+        Output::Echo => Ok(EngineConfig::InProcessText {
             model: Box::new(local_model),
             engine: dynamo_llm::engines::make_echo_engine(),
         }),
         #[cfg(feature = "mistralrs")]
-        Output::MistralRs => Ok(EngineConfig::StaticFull {
+        Output::MistralRs => Ok(EngineConfig::InProcessText {
             engine: dynamo_engine_mistralrs::make_engine(&local_model).await?,
             model: Box::new(local_model),
         }),
@@ -164,7 +164,7 @@ async fn engine_for(
             let engine =
                 dynamo_llm::mocker::engine::make_mocker_engine(drt, endpoint, args).await?;
 
-            Ok(EngineConfig::StaticCore {
+            Ok(EngineConfig::InProcessTokens {
                 engine,
                 model: Box::new(local_model),
                 is_prefill: false,

lib/bindings/python/rust/llm/entrypoint.rs

@@ -253,7 +253,7 @@ async fn select_engine(
     let inner = match args.engine_type {
         EngineType::Echo => {
             // There is no validation for the echo engine
-            RsEngineConfig::StaticFull {
+            RsEngineConfig::InProcessText {
                 model: Box::new(local_model),
                 engine: dynamo_llm::engines::make_echo_engine(),
             }
@@ -284,7 +284,7 @@ async fn select_engine(
             )
             .await?;
 
-            RsEngineConfig::StaticCore {
+            RsEngineConfig::InProcessTokens {
                 engine,
                 model: Box::new(local_model),
                 is_prefill: args.is_prefill,

lib/llm/src/entrypoint.rs

@@ -51,14 +51,14 @@ pub enum EngineConfig {
     /// Remote networked engines that we discover via etcd
     Dynamic(Box<LocalModel>),
 
-    /// A Full service engine does it's own tokenization and prompt formatting.
-    StaticFull {
+    /// A Text engine receives text, does it's own tokenization and prompt formatting.
+    InProcessText {
         engine: Arc<dyn StreamingEngine>,
         model: Box<LocalModel>,
     },
 
-    /// A core engine expects to be wrapped with pre/post processors that handle tokenization.
-    StaticCore {
+    /// A Tokens engine receives tokens, expects to be wrapped with pre/post processors that handle tokenization.
+    InProcessTokens {
         engine: ExecutionContext,
         model: Box<LocalModel>,
         is_prefill: bool,
@@ -70,8 +70,8 @@ impl EngineConfig {
         use EngineConfig::*;
         match self {
             Dynamic(lm) => lm,
-            StaticFull { model, .. } => model,
-            StaticCore { model, .. } => model,
+            InProcessText { model, .. } => model,
+            InProcessTokens { model, .. } => model,
         }
     }
 }

lib/llm/src/entrypoint/input/common.rs

@@ -93,7 +93,7 @@ pub async fn prepare_engine(
                 request_template: local_model.request_template(),
             })
         }
-        EngineConfig::StaticFull { engine, model, .. } => {
+        EngineConfig::InProcessText { engine, model, .. } => {
             let service_name = model.service_name().to_string();
             tracing::debug!("Model: {service_name} with engine pre-processing");
             let engine = Arc::new(StreamingEngineAdapter::new(engine));
@@ -105,7 +105,7 @@ pub async fn prepare_engine(
                 card: Some(model.into_card()),
             })
         }
-        EngineConfig::StaticCore {
+        EngineConfig::InProcessTokens {
             engine: inner_engine,
             model,
             ..

lib/llm/src/entrypoint/input/endpoint.rs

@@ -38,7 +38,7 @@ pub async fn run(
     let endpoint = component.endpoint(&endpoint_id.name);
 
     let rt_fut: Pin<Box<dyn Future<Output = _> + Send + 'static>> = match engine_config {
-        EngineConfig::StaticFull { engine, mut model } => {
+        EngineConfig::InProcessText { engine, mut model } => {
             let engine = Arc::new(StreamingEngineAdapter::new(engine));
             let ingress_chat = Ingress::<
                 Context<NvCreateChatCompletionRequest>,
@@ -51,7 +51,7 @@ pub async fn run(
 
             Box::pin(fut_chat)
         }
-        EngineConfig::StaticCore {
+        EngineConfig::InProcessTokens {
             engine: inner_engine,
             mut model,
             is_prefill,
@@ -127,7 +127,7 @@ mod integration_tests {
             .await
             .map_err(|e| anyhow::anyhow!("Failed to create distributed runtime: {}", e))?;
 
-        let engine_config = EngineConfig::StaticFull {
+        let engine_config = EngineConfig::InProcessText {
             engine: crate::engines::make_echo_engine(),
             model: Box::new(
                 crate::local_model::LocalModelBuilder::default()

lib/llm/src/entrypoint/input/grpc.rs

@@ -45,7 +45,7 @@ pub async fn run(
             .await?;
             grpc_service
         }
-        EngineConfig::StaticFull { engine, model, .. } => {
+        EngineConfig::InProcessText { engine, model, .. } => {
             let grpc_service = grpc_service_builder.build()?;
             let engine = Arc::new(StreamingEngineAdapter::new(engine));
             let manager = grpc_service.model_manager();
@@ -54,7 +54,7 @@ pub async fn run(
             manager.add_chat_completions_model(model.service_name(), checksum, engine)?;
             grpc_service
         }
-        EngineConfig::StaticCore {
+        EngineConfig::InProcessTokens {
             engine: inner_engine,
             model,
             ..

lib/llm/src/entrypoint/input/http.rs

@@ -87,7 +87,7 @@ pub async fn run(
             .await?;
             http_service
         }
-        EngineConfig::StaticFull { engine, model, .. } => {
+        EngineConfig::InProcessText { engine, model, .. } => {
             let http_service = http_service_builder.build()?;
             let engine = Arc::new(StreamingEngineAdapter::new(engine));
             let manager = http_service.model_manager();
@@ -101,7 +101,7 @@ pub async fn run(
             }
             http_service
         }
-        EngineConfig::StaticCore {
+        EngineConfig::InProcessTokens {
             engine: inner_engine,
             model,
             ..

lib/llm/tests/http_metrics.rs

@@ -321,7 +321,7 @@ mod integration_tests {
             .unwrap();
 
         // Create EngineConfig with EchoEngine
-        let engine_config = EngineConfig::StaticFull {
+        let engine_config = EngineConfig::InProcessText {
             engine: make_echo_engine(),
             model: Box::new(local_model.clone()),
         };
@@ -355,9 +355,8 @@ mod integration_tests {
             model_watcher.watch(discovery_stream, None).await;
         });
 
-        // Set up the engine following the StaticFull pattern from http.rs
-        let EngineConfig::StaticFull { engine, model, .. } = engine_config else {
-            panic!("Expected StaticFull config");
+        let EngineConfig::InProcessText { engine, model, .. } = engine_config else {
+            panic!("Expected InProcessText config");
         };
 
         let card = local_model.card().clone();
@@ -373,7 +372,7 @@ mod integration_tests {
         let test_component = namespace.component("test-mdc-component").unwrap();
         let test_endpoint = test_component.endpoint("test-mdc-endpoint");
 
-        // This will store the MDC in etcd for discovery
+        // This will store the MDC in key-value store for discovery
         local_model
             .attach(
                 &test_endpoint,