chore: Remove PreprocessedRequest alias BackendInput (#1307)

It was confusing to have two names for one type.

This tidy up started in #1064 , is now complete.

launch/dynamo-run/src/input/common.rs

@@ -9,7 +9,7 @@ use dynamo_llm::{
     engines::StreamingEngineAdapter,
     model_card::ModelDeploymentCard,
     preprocessor::OpenAIPreprocessor,
-    protocols::common::llm_backend::{BackendInput, BackendOutput},
+    protocols::common::llm_backend::{BackendOutput, PreprocessedRequest},
     types::{
         openai::chat_completions::{
             NvCreateChatCompletionRequest, NvCreateChatCompletionStreamResponse,
@@ -113,7 +113,7 @@ where
     OpenAIPreprocessor: Operator<
         Context<Req>,
         Pin<Box<dyn AsyncEngineStream<Annotated<Resp>>>>,
-        Context<BackendInput>,
+        Context<PreprocessedRequest>,
         Pin<Box<dyn AsyncEngineStream<Annotated<BackendOutput>>>>,
     >,
 {

launch/dynamo-run/src/input/endpoint.rs

@@ -19,7 +19,7 @@ use dynamo_llm::{
     backend::Backend,
     engines::StreamingEngineAdapter,
     model_type::ModelType,
-    preprocessor::{BackendInput, BackendOutput},
+    preprocessor::{BackendOutput, PreprocessedRequest},
     types::{
         openai::chat_completions::{
             NvCreateChatCompletionRequest, NvCreateChatCompletionStreamResponse,
@@ -71,8 +71,10 @@ pub async fn run(
             mut model,
         } => {
             // Pre-processing is done ingress-side, so it should be already done.
-            let frontend =
-                SegmentSource::<SingleIn<BackendInput>, ManyOut<Annotated<BackendOutput>>>::new();
+            let frontend = SegmentSource::<
+                SingleIn<PreprocessedRequest>,
+                ManyOut<Annotated<BackendOutput>>,
+            >::new();
             let backend = Backend::from_mdc(model.card().clone())
                 .await?
                 .into_operator();

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

@@ -16,7 +16,7 @@
 use super::*;
 use crate::llm::model_card::ModelDeploymentCard;
 
-use llm_rs::protocols::common::llm_backend::{BackendInput, BackendOutput};
+use llm_rs::protocols::common::llm_backend::{BackendOutput, PreprocessedRequest};
 use llm_rs::types::Annotated;
 
 use dynamo_runtime::pipeline::{Operator, ServiceBackend, ServiceFrontend, Source};
@@ -44,8 +44,10 @@ impl Backend {
     }
 
     fn start<'p>(&self, py: Python<'p>, generator: PyObject) -> PyResult<Bound<'p, PyAny>> {
-        let frontend =
-            ServiceFrontend::<SingleIn<BackendInput>, ManyOut<Annotated<BackendOutput>>>::new();
+        let frontend = ServiceFrontend::<
+            SingleIn<PreprocessedRequest>,
+            ManyOut<Annotated<BackendOutput>>,
+        >::new();
 
         let backend = self.inner.into_operator();
         let engine = Arc::new(PythonAsyncEngine::new(

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

@@ -18,7 +18,7 @@ use crate::llm::model_card::ModelDeploymentCard;
 
 use llm_rs::{
     preprocessor::OpenAIPreprocessor,
-    protocols::common::llm_backend::{BackendInput, BackendOutput},
+    protocols::common::llm_backend::{BackendOutput, PreprocessedRequest},
     types::{
         openai::chat_completions::{
             NvCreateChatCompletionRequest, NvCreateChatCompletionStreamResponse,
@@ -60,7 +60,7 @@ impl OAIChatPreprocessor {
         >::new();
 
         let network =
-            SegmentSink::<SingleIn<BackendInput>, ManyOut<Annotated<BackendOutput>>>::new();
+            SegmentSink::<SingleIn<PreprocessedRequest>, ManyOut<Annotated<BackendOutput>>>::new();
 
         let preprocessor = self.inner.into_operator();
         let pipeline = frontend
@@ -77,7 +77,7 @@ impl OAIChatPreprocessor {
         let endpoint = Arc::new(self.next.inner.clone());
         pyo3_async_runtimes::tokio::future_into_py(py, async move {
             let client = endpoint.client().await.map_err(to_pyerr)?;
-            let router = PushRouter::<BackendInput, Annotated<BackendOutput>>::from_client(
+            let router = PushRouter::<PreprocessedRequest, Annotated<BackendOutput>>::from_client(
                 client,
                 Default::default(),
             )

lib/engines/llamacpp/src/lib.rs

@@ -23,7 +23,7 @@ use llama_cpp_2::{
     LogOptions,
 };
 
-use dynamo_llm::protocols::common::llm_backend::{BackendInput, LLMEngineOutput};
+use dynamo_llm::protocols::common::llm_backend::LLMEngineOutput;
 use dynamo_llm::protocols::common::preprocessor::PreprocessedRequest;
 use dynamo_llm::{backend::ExecutionContext, local_model::LocalModel};
 
@@ -119,12 +119,12 @@ fn load_model(backend: &LlamaBackend, model_path: &Path) -> Result<LlamaModel> {
 }
 
 #[async_trait]
-impl AsyncEngine<SingleIn<BackendInput>, ManyOut<Annotated<LLMEngineOutput>>, Error>
+impl AsyncEngine<SingleIn<PreprocessedRequest>, ManyOut<Annotated<LLMEngineOutput>>, Error>
     for LlamacppEngine
 {
     async fn generate(
         &self,
-        request: SingleIn<BackendInput>,
+        request: SingleIn<PreprocessedRequest>,
     ) -> Result<ManyOut<Annotated<LLMEngineOutput>>, Error> {
         let (request, context) = request.into_parts();
         let ctx = context.context();

lib/llm/src/backend.rs

@@ -44,7 +44,7 @@ use dynamo_runtime::{
 
 use crate::protocols::{
     common::{
-        llm_backend::{BackendInput, BackendOutput, FinishReason, LLMEngineOutput},
+        llm_backend::{BackendOutput, FinishReason, LLMEngineOutput, PreprocessedRequest},
         StopConditions,
     },
     TokenIdType,
@@ -56,7 +56,7 @@ use tokenizers::Tokenizer as HfTokenizer;
 pub type ExecutionOutputStream = Annotated<LLMEngineOutput>;
 
 /// Context for executing LLM inference, engine consumes backend input and produces execution output stream
-pub type ExecutionContext = ServerStreamingEngine<BackendInput, ExecutionOutputStream>;
+pub type ExecutionContext = ServerStreamingEngine<PreprocessedRequest, ExecutionOutputStream>;
 
 /// Backend handles resource management and orchestrates LLM execution
 #[allow(dead_code)]
@@ -121,16 +121,16 @@ impl Backend {
 #[async_trait]
 impl
     Operator<
-        SingleIn<BackendInput>,
+        SingleIn<PreprocessedRequest>,
         ManyOut<Annotated<BackendOutput>>,
-        SingleIn<BackendInput>,
+        SingleIn<PreprocessedRequest>,
         ManyOut<Annotated<LLMEngineOutput>>,
     > for Backend
 {
     async fn generate(
         &self,
-        request: SingleIn<BackendInput>,
-        next: ServerStreamingEngine<BackendInput, Annotated<LLMEngineOutput>>,
+        request: SingleIn<PreprocessedRequest>,
+        next: ServerStreamingEngine<PreprocessedRequest, Annotated<LLMEngineOutput>>,
     ) -> Result<ManyOut<Annotated<BackendOutput>>> {
         let stop_conditions = request.stop_conditions.clone();
         let next_stream = next.generate(request).await?;

lib/llm/src/discovery/watcher.rs

@@ -20,7 +20,7 @@ use crate::{
     backend::Backend,
     kv_router::KvPushRouter,
     model_type::ModelType,
-    preprocessor::{BackendInput, OpenAIPreprocessor},
+    preprocessor::{OpenAIPreprocessor, PreprocessedRequest},
     protocols::common::llm_backend::LLMEngineOutput,
     protocols::openai::chat_completions::{
         NvCreateChatCompletionRequest, NvCreateChatCompletionStreamResponse,
@@ -196,7 +196,8 @@ impl ModelWatcher {
                 >::new();
                 let preprocessor = OpenAIPreprocessor::new(card.clone()).await?.into_operator();
                 let backend = Backend::from_mdc(card.clone()).await?.into_operator();
-                let router = PushRouter::<BackendInput, Annotated<LLMEngineOutput>>::from_client(
+                let router =
+                    PushRouter::<PreprocessedRequest, Annotated<LLMEngineOutput>>::from_client(
                         client.clone(),
                         self.router_mode,
                     )
@@ -231,7 +232,8 @@ impl ModelWatcher {
                 >::new();
                 let preprocessor = OpenAIPreprocessor::new(card.clone()).await?.into_operator();
                 let backend = Backend::from_mdc(card.clone()).await?.into_operator();
-                let router = PushRouter::<BackendInput, Annotated<LLMEngineOutput>>::from_client(
+                let router =
+                    PushRouter::<PreprocessedRequest, Annotated<LLMEngineOutput>>::from_client(
                         client,
                         self.router_mode,
                     )

lib/llm/src/engines.rs

@@ -26,7 +26,7 @@ use dynamo_runtime::pipeline::{Error, ManyOut, SingleIn};
 use dynamo_runtime::protocols::annotated::Annotated;
 
 use crate::backend::ExecutionContext;
-use crate::preprocessor::BackendInput;
+use crate::preprocessor::PreprocessedRequest;
 use crate::protocols::common::llm_backend::LLMEngineOutput;
 use crate::protocols::openai::{
     chat_completions::{NvCreateChatCompletionRequest, NvCreateChatCompletionStreamResponse},
@@ -86,12 +86,12 @@ pub fn make_engine_core() -> ExecutionContext {
 }
 
 #[async_trait]
-impl AsyncEngine<SingleIn<BackendInput>, ManyOut<Annotated<LLMEngineOutput>>, Error>
+impl AsyncEngine<SingleIn<PreprocessedRequest>, ManyOut<Annotated<LLMEngineOutput>>, Error>
     for EchoEngineCore
 {
     async fn generate(
         &self,
-        incoming_request: SingleIn<BackendInput>,
+        incoming_request: SingleIn<PreprocessedRequest>,
     ) -> Result<ManyOut<Annotated<LLMEngineOutput>>, Error> {
         let (request, context) = incoming_request.into_parts();
         let ctx = context.context();

lib/llm/src/kv_router.rs

@@ -31,7 +31,7 @@ use crate::{
         scheduler::{KvScheduler, KvSchedulerError, SchedulingRequest},
         scoring::ProcessedEndpoints,
     },
-    preprocessor::BackendInput,
+    preprocessor::PreprocessedRequest,
     protocols::common::llm_backend::LLMEngineOutput,
     tokens::TokenBlockSequence,
 };
@@ -173,13 +173,13 @@ impl AsyncEngine<SingleIn<RouterRequest>, ManyOut<Annotated<RouterResponse>>, Er
 }
 
 pub struct KvPushRouter {
-    inner: PushRouter<BackendInput, Annotated<LLMEngineOutput>>,
+    inner: PushRouter<PreprocessedRequest, Annotated<LLMEngineOutput>>,
     chooser: Arc<KvRouter>,
 }
 
 impl KvPushRouter {
     pub fn new(
-        inner: PushRouter<BackendInput, Annotated<LLMEngineOutput>>,
+        inner: PushRouter<PreprocessedRequest, Annotated<LLMEngineOutput>>,
         chooser: Arc<KvRouter>,
     ) -> Self {
         KvPushRouter { inner, chooser }
@@ -187,12 +187,12 @@ impl KvPushRouter {
 }
 
 #[async_trait]
-impl AsyncEngine<SingleIn<BackendInput>, ManyOut<Annotated<LLMEngineOutput>>, Error>
+impl AsyncEngine<SingleIn<PreprocessedRequest>, ManyOut<Annotated<LLMEngineOutput>>, Error>
     for KvPushRouter
 {
     async fn generate(
         &self,
-        request: SingleIn<BackendInput>,
+        request: SingleIn<PreprocessedRequest>,
     ) -> Result<ManyOut<Annotated<LLMEngineOutput>>, Error> {
         match self.inner.client.instance_source.as_ref() {
             InstanceSource::Static => self.inner.r#static(request).await,

lib/llm/src/preprocessor.rs

@@ -55,7 +55,7 @@ use crate::tokenizers::{traits::Tokenizer, HuggingFaceTokenizer};
 
 use crate::preprocessor::prompt::PromptFormatter;
 
-pub use crate::protocols::common::llm_backend::{BackendInput, BackendOutput};
+pub use crate::protocols::common::llm_backend::{BackendOutput, PreprocessedRequest};
 
 pub const ANNOTATION_FORMATTED_PROMPT: &str = "formatted_prompt";
 pub const ANNOTATION_TOKEN_IDS: &str = "token_ids";
@@ -121,9 +121,9 @@ impl OpenAIPreprocessor {
     >(
         &self,
         request: &R,
-    ) -> Result<(BackendInput, HashMap<String, String>)> {
+    ) -> Result<(PreprocessedRequest, HashMap<String, String>)> {
         let mut annotations = HashMap::new();
-        let mut builder = BackendInput::builder();
+        let mut builder = PreprocessedRequest::builder();
 
         let use_raw_prompt = request
             .nvext()
@@ -266,7 +266,7 @@ impl
     Operator<
         SingleIn<NvCreateChatCompletionRequest>,
         ManyOut<Annotated<NvCreateChatCompletionStreamResponse>>,
-        SingleIn<BackendInput>,
+        SingleIn<PreprocessedRequest>,
         ManyOut<Annotated<BackendOutput>>,
     > for OpenAIPreprocessor
 {
@@ -274,7 +274,11 @@ impl
         &self,
         request: SingleIn<NvCreateChatCompletionRequest>,
         next: Arc<
-            dyn AsyncEngine<SingleIn<BackendInput>, ManyOut<Annotated<BackendOutput>>, Error>,
+            dyn AsyncEngine<
+                SingleIn<PreprocessedRequest>,
+                ManyOut<Annotated<BackendOutput>>,
+                Error,
+            >,
         >,
     ) -> Result<ManyOut<Annotated<NvCreateChatCompletionStreamResponse>>, Error> {
         // unpack the request
@@ -320,7 +324,7 @@ impl
     Operator<
         SingleIn<CompletionRequest>,
         ManyOut<Annotated<CompletionResponse>>,
-        SingleIn<BackendInput>,
+        SingleIn<PreprocessedRequest>,
         ManyOut<Annotated<BackendOutput>>,
     > for OpenAIPreprocessor
 {
@@ -328,7 +332,11 @@ impl
         &self,
         request: SingleIn<CompletionRequest>,
         next: Arc<
-            dyn AsyncEngine<SingleIn<BackendInput>, ManyOut<Annotated<BackendOutput>>, Error>,
+            dyn AsyncEngine<
+                SingleIn<PreprocessedRequest>,
+                ManyOut<Annotated<BackendOutput>>,
+                Error,
+            >,
         >,
     ) -> Result<ManyOut<Annotated<CompletionResponse>>, Error> {
         // unpack the request

lib/llm/src/protocols/common/llm_backend.rs

@@ -20,7 +20,6 @@ use crate::protocols::TokenIdType;
 pub type TokenType = Option<String>;
 pub type LogProbs = Vec<f64>;
 
-pub use super::preprocessor::PreprocessedRequest as BackendInput; // TODO stop renaming this
 pub use super::preprocessor::PreprocessedRequest;
 pub use super::FinishReason;