feat: add auto-generated frontend OpenAPI spec and helper binary (#4802)

Signed-off-by: Satvik Matta <smatta@nvidia.com>

lib/llm/src/protocols/openai/embeddings/nvext.rs

@@ -3,6 +3,7 @@
 
 use derive_builder::Builder;
 use serde::{Deserialize, Serialize};
+use utoipa::ToSchema;
 use validator::{Validate, ValidationError};
 
 pub trait NvExtProvider {
@@ -10,7 +11,7 @@ pub trait NvExtProvider {
 }
 
 /// NVIDIA LLM extensions to the OpenAI API
-#[derive(Serialize, Deserialize, Builder, Validate, Debug, Clone)]
+#[derive(ToSchema, Serialize, Deserialize, Builder, Validate, Debug, Clone)]
 #[validate(schema(function = "validate_nv_ext"))]
 pub struct NvExt {
     /// Annotations

lib/llm/src/protocols/openai/nvext.rs

@@ -3,6 +3,7 @@
 
 use derive_builder::Builder;
 use serde::{Deserialize, Serialize};
+use utoipa::ToSchema;
 use validator::{Validate, ValidationError};
 
 pub use crate::protocols::common::timing::TimingInfo;
@@ -13,7 +14,7 @@ pub trait NvExtProvider {
 }
 
 /// Worker ID information for disaggregated serving
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq)]
+#[derive(ToSchema, Serialize, Deserialize, Debug, Clone, PartialEq)]
 pub struct WorkerIdInfo {
     /// The prefill worker ID that processed this request
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -25,7 +26,7 @@ pub struct WorkerIdInfo {
 }
 
 /// NVIDIA LLM response extensions
-#[derive(Serialize, Deserialize, Debug, Clone)]
+#[derive(ToSchema, Serialize, Deserialize, Debug, Clone)]
 pub struct NvExtResponse {
     /// Worker ID information (prefill and decode worker IDs)
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -43,7 +44,7 @@ pub struct NvExtResponse {
 }
 
 /// NVIDIA LLM extensions to the OpenAI API
-#[derive(Serialize, Deserialize, Builder, Validate, Debug, Clone)]
+#[derive(ToSchema, Serialize, Deserialize, Builder, Validate, Debug, Clone)]
 #[validate(schema(function = "validate_nv_ext"))]
 pub struct NvExt {
     /// If true, sampling will be forced to be greedy.

lib/llm/src/protocols/openai/responses.rs

@@ -11,6 +11,7 @@ use dynamo_async_openai::types::{
 };
 use dynamo_runtime::protocols::annotated::AnnotationsProvider;
 use serde::{Deserialize, Serialize};
+use utoipa::ToSchema;
 use uuid::Uuid;
 use validator::Validate;
 
@@ -18,8 +19,9 @@ use super::chat_completions::{NvCreateChatCompletionRequest, NvCreateChatComplet
 use super::nvext::{NvExt, NvExtProvider};
 use super::{OpenAISamplingOptionsProvider, OpenAIStopConditionsProvider};
 
-#[derive(Serialize, Deserialize, Validate, Debug, Clone)]
+#[derive(ToSchema, Serialize, Deserialize, Validate, Debug, Clone)]
 pub struct NvCreateResponse {
+    /// Flattened CreateResponse fields (model, input, temperature, etc.)
     #[serde(flatten)]
     pub inner: dynamo_async_openai::types::responses::CreateResponse,
 
@@ -27,8 +29,9 @@ pub struct NvCreateResponse {
     pub nvext: Option<NvExt>,
 }
 
-#[derive(Serialize, Deserialize, Validate, Debug, Clone)]
+#[derive(ToSchema, Serialize, Deserialize, Validate, Debug, Clone)]
 pub struct NvResponse {
+    /// Flattened Response fields.
     #[serde(flatten)]
     pub inner: dynamo_async_openai::types::responses::Response,