refactor: Split ModelType to ModelInput for request and response type;...

refactor: Split ModelType to ModelInput for request and response type; ModelType for the supported workloads (#2714)
Signed-off-by: Guan Luo <gluo@nvidia.com>
Signed-off-by: GuanLuo <41310872+GuanLuo@users.noreply.github.com>
Co-authored-by: Guan Luo <gluo@nvidia.com>
Co-authored-by: GuanLuo <41310872+GuanLuo@users.noreply.github.com>

Cargo.lock

@@ -726,6 +726,9 @@ name = "bitflags"
 version = "2.9.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "34efbcccd345379ca2868b2b2c9d3782e9cc58ba87bc7d79d5b53d9c9ae6f25d"
+dependencies = [
+ "serde",
+]
 
 [[package]]
 name = "bitstream-io"
@@ -1985,6 +1988,7 @@ dependencies = [
  "async_zmq",
  "axum 0.8.4",
  "axum-server",
+ "bitflags 2.9.3",
  "blake3",
  "bs62",
  "bytemuck",

components/backends/llama_cpp/src/dynamo/llama_cpp/main.py

@@ -11,7 +11,7 @@ from typing import Optional
 import uvloop
 from llama_cpp import Llama
 
-from dynamo.llm import ModelType, register_llm
+from dynamo.llm import ModelInput, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 from dynamo.runtime.logging import configure_dynamo_logging
 
@@ -41,10 +41,10 @@ async def worker(runtime: DistributedRuntime):
     component = runtime.namespace(config.namespace).component(config.component)
     await component.create_service()
 
-    model_type = ModelType.Chat  # llama.cpp does the pre-processing
     endpoint = component.endpoint(config.endpoint)
     await register_llm(
-        model_type,
+        ModelInput.Tokens,
+        ModelType.Chat,
         endpoint,
         config.model_path,
         config.model_name,

components/backends/sglang/src/dynamo/sglang/register.py

@@ -8,7 +8,7 @@ import sglang as sgl
 from sglang.srt.server_args import ServerArgs
 
 from dynamo._core import Endpoint
-from dynamo.llm import ModelRuntimeConfig, ModelType, register_llm
+from dynamo.llm import ModelInput, ModelRuntimeConfig, ModelType, register_llm
 from dynamo.sglang.args import DynamoArgs
 
 
@@ -26,7 +26,8 @@ async def register_llm_with_runtime_config(
     runtime_config = await _get_runtime_config(engine, dynamo_args)
     try:
         await register_llm(
-            ModelType.Backend,
+            ModelInput.Tokens,
+            ModelType.Chat | ModelType.Completions,
             endpoint,
             server_args.model_path,
             server_args.served_model_name,

components/backends/trtllm/src/dynamo/trtllm/main.py

@@ -22,7 +22,7 @@ from torch.cuda import device_count
 from transformers import AutoConfig
 
 import dynamo.nixl_connect as nixl_connect
-from dynamo.llm import ModelRuntimeConfig, ModelType, register_llm
+from dynamo.llm import ModelInput, ModelRuntimeConfig, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 from dynamo.runtime.logging import configure_dynamo_logging
 from dynamo.trtllm.engine import TensorRTLLMEngine, get_llm_engine
@@ -223,7 +223,8 @@ async def init(runtime: DistributedRuntime, config: Config):
     default_sampling_params = SamplingParams()
     default_sampling_params._setup(tokenizer)
     default_sampling_params.stop = None
-    modelType = ModelType.Backend
+    model_input = ModelInput.Tokens
+    model_type = ModelType.Chat | ModelType.Completions
     multimodal_processor = None
 
     if os.getenv("DYNAMO_ENABLE_TEST_LOGITS_PROCESSOR") == "1":
@@ -234,7 +235,7 @@ async def init(runtime: DistributedRuntime, config: Config):
 
     if modality == "multimodal":
         engine_args["skip_tokenizer_init"] = False
-        modelType = ModelType.Chat
+        model_input = ModelInput.Text
         model_config = AutoConfig.from_pretrained(
             config.model_path, trust_remote_code=True
         )
@@ -292,7 +293,8 @@ async def init(runtime: DistributedRuntime, config: Config):
         if is_first_worker(config):
             # Register the model with runtime config
             await register_llm(
-                modelType,
+                model_input,
+                model_type,
                 endpoint,
                 config.model_path,
                 config.served_model_name,

components/backends/vllm/src/dynamo/vllm/main.py

@@ -12,6 +12,7 @@ from vllm.usage.usage_lib import UsageContext
 from vllm.v1.engine.async_llm import AsyncLLM
 
 from dynamo.llm import (
+    ModelInput,
     ModelRuntimeConfig,
     ModelType,
     ZmqKvEventPublisher,
@@ -251,7 +252,8 @@ async def init(runtime: DistributedRuntime, config: Config):
         runtime_config.reasoning_parser = config.reasoning_parser
 
         await register_llm(
-            ModelType.Backend,
+            ModelInput.Tokens,
+            ModelType.Chat | ModelType.Completions,
             generate_endpoint,
             config.model,
             config.served_model_name,

docs/guides/backend.md

@@ -16,7 +16,7 @@ The Python file must do three things:
 3. Attach a request handler
 
 ```
-from dynamo.llm import ModelType, register_llm
+from dynamo.llm import ModelInput, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 
    # 1. Decorate a function to get the runtime
@@ -29,10 +29,11 @@ from dynamo.runtime import DistributedRuntime, dynamo_worker
     component = runtime.namespace("namespace").component("component")
     await component.create_service()
     model_path = "Qwen/Qwen3-0.6B" # or "/data/models/Qwen3-0.6B"
-    model_type = ModelType.Backend
+    model_input = ModelInput.Tokens # or ModelInput.Text if engine handles pre-processing
+    model_type = ModelType.Chat # or ModelType.Chat | ModelType.Completions if model can be deployed on chat and completions endpoints
     endpoint = component.endpoint("endpoint")
     # Optional last param to register_llm is model_name. If not present derives it from model_path
-    await register_llm(model_type, endpoint, model_path)
+    await register_llm(model_input, model_type, endpoint, model_path)
 
     # Initialize your engine here
     # engine = ...
@@ -62,10 +63,13 @@ The `model_path` can be:
 - The path to a checkout of a HuggingFace repo - any folder containing safetensor files as well as `config.json`, `tokenizer.json` and `tokenizer_config.json`.
 - The path to a GGUF file, if your engine supports that.
 
+The `model_input` can be:
+- ModelInput.Tokens. Your engine expects pre-processed input (token IDs). Dynamo handles tokenization and pre-processing.
+- ModelInput.Text. Your engine expects raw text input and handles its own tokenization and pre-processing.
+
 The `model_type` can be:
-- ModelType.Backend. Dynamo handles pre-processing. Your `generate` method receives a `request` dict containing a `token_ids` array of int. It must return a dict also containing a `token_ids` array and an optional `finish_reason` string.
-- ModelType.Chat. Your `generate` method receives a `request` and must return a response dict of type [OpenAI Chat Completion](https://platform.openai.com/docs/api-reference/chat). Your engine handles pre-processing.
-- ModelType.Completion. Your `generate` method receives a `request` and must return a response dict of the older [Completions](https://platform.openai.com/docs/api-reference/completions). Your engine handles pre-processing.
+- ModelType.Chat. Your `generate` method receives a `request` and must return a response dict of type [OpenAI Chat Completion](https://platform.openai.com/docs/api-reference/chat).
+- ModelType.Completions. Your `generate` method receives a `request` and must return a response dict of the older [Completions](https://platform.openai.com/docs/api-reference/completions).
 
 `register_llm` can also take the following kwargs:
 - `model_name`: The name to call the model. Your incoming HTTP requests model name must match this. Defaults to the hugging face repo name, the folder name, or the GGUF file name.

docs/guides/dynamo_run.md

@@ -389,7 +389,7 @@ The Python file must do three things:
 3. Attach a request handler
 
 ```
-from dynamo.llm import ModelType, register_llm
+from dynamo.llm import ModelInput, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 
    # 1. Decorate a function to get the runtime
@@ -402,10 +402,11 @@ from dynamo.runtime import DistributedRuntime, dynamo_worker
     component = runtime.namespace("namespace").component("component")
     await component.create_service()
     model_path = "Qwen/Qwen3-0.6B" # or "/data/models/Qwen3-0.6B"
-    model_type = ModelType.Backend
+    model_input = ModelInput.Tokens # or ModelInput.Text if engine handles pre-processing
+    model_type = ModelType.Chat # or ModelType.Chat | ModelType.Completions if model can be deployed on chat and completions endpoints
     endpoint = component.endpoint("endpoint")
     # Optional last param to register_llm is model_name. If not present derives it from model_path
-    await register_llm(model_type, endpoint, model_path)
+    await register_llm(model_input, model_type, endpoint, model_path)
 
     # Initialize your engine here
     # engine = ...
@@ -435,10 +436,13 @@ The `model_path` can be:
 - The path to a checkout of a HuggingFace repo - any folder containing safetensor files as well as `config.json`, `tokenizer.json` and `tokenizer_config.json`.
 - The path to a GGUF file, if your engine supports that.
 
+The `model_input` can be:
+- ModelInput.Tokens. Your engine expects pre-processed input (token IDs). Dynamo handles tokenization and pre-processing.
+- ModelInput.Text. Your engine expects raw text input and handles its own tokenization and pre-processing.
+
 The `model_type` can be:
-- ModelType.Backend. Dynamo handles pre-processing. Your `generate` method receives a `request` dict containing a `token_ids` array of int. It must return a dict also containing a `token_ids` array and an optional `finish_reason` string.
-- ModelType.Chat. Your `generate` method receives a `request` and must return a response dict of type [OpenAI Chat Completion](https://platform.openai.com/docs/api-reference/chat). Your engine handles pre-processing.
-- ModelType.Completion. Your `generate` method receives a `request` and must return a response dict of the older [Completions](https://platform.openai.com/docs/api-reference/completions). Your engine handles pre-processing.
+- ModelType.Chat. Your `generate` method receives a `request` and must return a response dict of type [OpenAI Chat Completion](https://platform.openai.com/docs/api-reference/chat).
+- ModelType.Completions. Your `generate` method receives a `request` and must return a response dict of the older [Completions](https://platform.openai.com/docs/api-reference/completions).
 
 `register_llm` can also take the following kwargs:
 - `model_name`: The name to call the model. Your incoming HTTP requests model name must match this. Defaults to the hugging face repo name, the folder name, or the GGUF file name.

examples/multimodal/components/processor.py

@@ -32,7 +32,7 @@ from vllm.outputs import RequestOutput
 from vllm.transformers_utils.tokenizer import AnyTokenizer
 from vllm.utils import FlexibleArgumentParser
 
-from dynamo.llm import ModelType, register_llm
+from dynamo.llm import ModelInput, ModelType, register_llm
 from dynamo.runtime import Client, DistributedRuntime, dynamo_worker
 from dynamo.runtime.logging import configure_dynamo_logging
 
@@ -321,7 +321,8 @@ async def init(runtime: DistributedRuntime, args: argparse.Namespace, config: Co
 
     # Register the endpoint as entrypoint to a model
     await register_llm(
-        ModelType.Chat,  # Custom processor is used and this type bypasses SDK processor
+        ModelInput.Text,  # Custom processor is used and this type bypasses SDK processor
+        ModelType.Chat,
         generate_endpoint,
         config.model,
         config.served_model_name,

lib/bindings/python/Cargo.lock

@@ -590,6 +590,9 @@ name = "bitflags"
 version = "2.9.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "34efbcccd345379ca2868b2b2c9d3782e9cc58ba87bc7d79d5b53d9c9ae6f25d"
+dependencies = [
+ "serde",
+]
 
 [[package]]
 name = "bitstream-io"
@@ -1383,6 +1386,7 @@ dependencies = [
  "async_zmq",
  "axum",
  "axum-server",
+ "bitflags 2.9.3",
  "blake3",
  "bs62",
  "bytemuck",

lib/bindings/python/examples/hello_world/server_sglang.py

@@ -8,7 +8,7 @@
 # request via NATS to this python script, which runs sglang.
 #
 # The key differences between this and `server_sglang_tok.py` are:
-# - The `register_llm` function registers us a `Backend` model
+# - The `register_llm` function registers us a `Chat` and `Completions` model that accepts `Tokens` input
 # - The `generate` function receives a pre-tokenized request and must return token_ids in the response.
 #
 # Setup a virtualenv with dynamo.llm, dynamo.runtime and sglang[all] installed
@@ -27,7 +27,7 @@ import sglang
 import uvloop
 from sglang.srt.server_args import ServerArgs
 
-from dynamo.llm import ModelType, register_llm
+from dynamo.llm import ModelInput, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 
 DEFAULT_ENDPOINT = "dyn://dynamo.backend.generate"
@@ -91,7 +91,12 @@ async def init(runtime: DistributedRuntime, config: Config):
     await component.create_service()
 
     endpoint = component.endpoint(config.endpoint)
-    await register_llm(ModelType.Backend, endpoint, config.model)
+    await register_llm(
+        ModelInput.Tokens,
+        ModelType.Chat | ModelType.Completions,
+        endpoint,
+        config.model,
+    )
 
     engine_args = ServerArgs(
         model_path=config.model,

lib/bindings/python/examples/hello_world/server_sglang_tok.py

@@ -9,7 +9,7 @@
 # do the pre/post-processing.
 #
 # The key differences between this and `server_sglang.py` are:
-# - The `register_llm` function registers us a `Chat` model
+# - The `register_llm` function registers us a `Chat` and `Completions` model that accepts `Text` input
 # - The `generate` function receives a chat completion request and must return matching response
 #
 # Setup a virtualenv with dynamo.llm, dynamo.runtime and sglang[all] installed
@@ -31,7 +31,7 @@ from sglang.srt.openai_api.adapter import v1_chat_generate_request
 from sglang.srt.openai_api.protocol import ChatCompletionRequest
 from sglang.srt.server_args import ServerArgs
 
-from dynamo.llm import ModelType, register_llm
+from dynamo.llm import ModelInput, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 
 DEFAULT_ENDPOINT = "dyn://dynamo.backend.generate"
@@ -104,7 +104,9 @@ async def init(runtime: DistributedRuntime, config: Config):
     await component.create_service()
 
     endpoint = component.endpoint(config.endpoint)
-    await register_llm(ModelType.Chat, endpoint, config.model)
+    await register_llm(
+        ModelInput.Text, ModelType.Chat | ModelType.Completions, endpoint, config.model
+    )
 
     server_args = ServerArgs(model_path=config.model)
     tokenizer_manager, _scheduler_info = _launch_subprocesses(server_args=server_args)

lib/bindings/python/examples/hello_world/server_vllm.py

@@ -39,7 +39,7 @@ from vllm.entrypoints.openai.api_server import (
 )
 from vllm.inputs import TokensPrompt
 
-from dynamo.llm import ModelType, register_llm
+from dynamo.llm import ModelInput, ModelType, register_llm
 from dynamo.runtime import DistributedRuntime, dynamo_worker
 
 DEFAULT_ENDPOINT = "dyn://dynamo.backend.generate"
@@ -114,7 +114,12 @@ async def init(runtime: DistributedRuntime, config: Config):
     await component.create_service()
 
     endpoint = component.endpoint(config.endpoint)
-    await register_llm(ModelType.Backend, endpoint, config.model)
+    await register_llm(
+        ModelInput.Tokens,
+        ModelType.Chat | ModelType.Completions,
+        endpoint,
+        config.model,
+    )
 
     engine_args = AsyncEngineArgs(
         model=config.model,

lib/bindings/python/rust/lib.rs

@@ -109,6 +109,7 @@ fn _core(m: &Bound<'_, PyModule>) -> PyResult<()> {
     m.add_class::<context::Context>()?;
     m.add_class::<EtcdKvCache>()?;
     m.add_class::<ModelType>()?;
+    m.add_class::<ModelInput>()?;
     m.add_class::<llm::kv::ForwardPassMetrics>()?;
     m.add_class::<llm::kv::WorkerStats>()?;
     m.add_class::<llm::kv::KvStats>()?;
@@ -141,10 +142,11 @@ fn log_message(level: &str, message: &str, module: &str, file: &str, line: u32)
 }
 
 #[pyfunction]
-#[pyo3(signature = (model_type, endpoint, model_path, model_name=None, context_length=None, kv_cache_block_size=None, router_mode=None, migration_limit=0, runtime_config=None, user_data=None, custom_template_path=None))]
+#[pyo3(signature = (model_input, model_type, endpoint, model_path, model_name=None, context_length=None, kv_cache_block_size=None, router_mode=None, migration_limit=0, runtime_config=None, user_data=None, custom_template_path=None))]
 #[allow(clippy::too_many_arguments)]
 fn register_llm<'p>(
     py: Python<'p>,
+    model_input: ModelInput,
     model_type: ModelType,
     endpoint: Endpoint,
     model_path: &str,
@@ -157,13 +159,13 @@ fn register_llm<'p>(
     user_data: Option<&Bound<'p, PyDict>>,
     custom_template_path: Option<&str>,
 ) -> PyResult<Bound<'p, PyAny>> {
-    let model_type_obj = match model_type {
-        ModelType::Chat => llm_rs::model_type::ModelType::Chat,
-        ModelType::Completion => llm_rs::model_type::ModelType::Completion,
-        ModelType::Backend => llm_rs::model_type::ModelType::Backend,
-        ModelType::Embedding => llm_rs::model_type::ModelType::Embedding,
+    let model_input = match model_input {
+        ModelInput::Text => llm_rs::model_type::ModelInput::Text,
+        ModelInput::Tokens => llm_rs::model_type::ModelInput::Tokens,
     };
 
+    let model_type_obj = model_type.inner;
+
     let inner_path = model_path.to_string();
     let model_name = model_name.map(|n| n.to_string());
     let router_mode = router_mode.unwrap_or(RouterMode::RoundRobin);
@@ -205,7 +207,7 @@ fn register_llm<'p>(
         let mut local_model = builder.build().await.map_err(to_pyerr)?;
         // Advertise ourself on etcd so ingress can find us
         local_model
-            .attach(&endpoint.inner, model_type_obj)
+            .attach(&endpoint.inner, model_type_obj, model_input)
             .await
             .map_err(to_pyerr)?;
 
@@ -272,14 +274,44 @@ struct Client {
     router: rs::pipeline::PushRouter<serde_json::Value, RsAnnotated<serde_json::Value>>,
 }
 
+#[pyclass]
+#[derive(Clone, PartialEq)]
+struct ModelType {
+    inner: llm_rs::model_type::ModelType,
+}
+
+#[pymethods]
+#[allow(non_upper_case_globals)]
+impl ModelType {
+    #[classattr]
+    const Chat: Self = ModelType {
+        inner: llm_rs::model_type::ModelType::Chat,
+    };
+    #[classattr]
+    const Completions: Self = ModelType {
+        inner: llm_rs::model_type::ModelType::Completions,
+    };
+    #[classattr]
+    const Embedding: Self = ModelType {
+        inner: llm_rs::model_type::ModelType::Embedding,
+    };
+
+    fn __or__(&self, other: &Self) -> Self {
+        ModelType {
+            inner: self.inner | other.inner,
+        }
+    }
+
+    fn __str__(&self) -> String {
+        self.inner.to_string()
+    }
+}
+
 #[pyclass(eq, eq_int)]
 #[derive(Clone, PartialEq)]
-#[repr(i32)]
-enum ModelType {
-    Chat = 1,
-    Completion = 2,
-    Backend = 3,
-    Embedding = 4,
+enum ModelInput {
+    Text = 1,
+    Tokens = 2,
 }
 
 #[pymethods]

lib/bindings/python/src/dynamo/_core.pyi

@@ -843,8 +843,12 @@ class HttpAsyncEngine:
 
     ...
 
+class ModelInput:
+    """What type of request this model needs: Text or Tokens"""
+    ...
+
 class ModelType:
-    """What type of request this model needs: Chat, Component or Backend (pre-processed)"""
+    """What type of request this model needs: Chat, Completions or Embedding"""
     ...
 
 class RouterMode:
@@ -859,7 +863,19 @@ class KvRouterConfig:
     """Values for KV router"""
     ...
 
-async def register_llm(model_type: ModelType, endpoint: Endpoint, model_path: str, model_name: Optional[str] = None, context_length: Optional[int] = None, kv_cache_block_size: Optional[int] = None, router_mode: Optional[RouterMode] = None, migration_limit: int = 0, runtime_config: Optional[ModelRuntimeConfig] = None, user_data: Optional[dict] = None, custom_template_path: Optional[str] = None) -> None:
+async def register_llm(
+    model_input: ModelInput,
+    model_type: ModelType,
+    endpoint: Endpoint,
+    model_path: str,
+    model_name: Optional[str] = None,
+    context_length: Optional[int] = None,
+    kv_cache_block_size: Optional[int] = None,
+    migration_limit: int = 0,
+    router_mode: Optional[RouterMode] = None,
+    user_data: Optional[Dict[str, Any]] = None,
+    custom_template_path: Optional[str] = None,
+) -> None:
     """Attach the model at path to the given endpoint, and advertise it as model_type"""
     ...
 

lib/bindings/python/src/dynamo/llm/__init__.py

@@ -28,6 +28,7 @@ from dynamo._core import KvMetricsAggregator as KvMetricsAggregator
 from dynamo._core import KvRecorder as KvRecorder
 from dynamo._core import KvRouterConfig as KvRouterConfig
 from dynamo._core import KvStats as KvStats
+from dynamo._core import ModelInput as ModelInput
 from dynamo._core import ModelRuntimeConfig as ModelRuntimeConfig
 from dynamo._core import ModelType as ModelType
 from dynamo._core import OverlapScores as OverlapScores

lib/llm/Cargo.toml

@@ -89,6 +89,7 @@ xxhash-rust = { workspace = true }
 # model_express_common = { version = "0.1.0", optional = true }
 
 akin = "0.4.0"
+bitflags = { version = "2.4", features = ["serde"] }
 blake3 = "1"
 bytemuck = "1.22"
 candle-core = { version = "0.9.1" }

lib/llm/src/discovery/model_entry.rs

@@ -14,7 +14,7 @@ use serde::{Deserialize, Serialize};
 use crate::{
     local_model::runtime_config::ModelRuntimeConfig,
     model_card::{self, ModelDeploymentCard},
-    model_type::ModelType,
+    model_type::{ModelInput, ModelType},
 };
 
 /// [ModelEntry] contains the information to discover models from the etcd cluster.
@@ -34,6 +34,11 @@ pub struct ModelEntry {
     /// Runtime configuration specific to this model instance
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub runtime_config: Option<ModelRuntimeConfig>,
+
+    /// Specifies the model input type.
+    /// `Tokens` for engines that expect pre-processed input.
+    /// `Text` for engines that take care of pre-processing themselves.
+    pub model_input: ModelInput,
 }
 
 impl ModelEntry {
@@ -43,7 +48,7 @@ impl ModelEntry {
     }
 
     pub fn requires_preprocessing(&self) -> bool {
-        matches!(self.model_type, ModelType::Backend)
+        matches!(self.model_input, ModelInput::Tokens)
     }
 
     /// Fetch the ModelDeploymentCard from etcd.

lib/llm/src/discovery/watcher.rs

@@ -21,8 +21,8 @@ use crate::{
     backend::Backend,
     entrypoint,
     kv_router::KvRouterConfig,
-    model_type::ModelType,
-    preprocessor::{OpenAIPreprocessor, PreprocessedEmbeddingRequest},
+    model_type::{ModelInput, ModelType},
+    preprocessor::{OpenAIPreprocessor, PreprocessedEmbeddingRequest, prompt::PromptFormatter},
     protocols::{
         common::llm_backend::EmbeddingsEngineOutput,
         openai::{
@@ -54,8 +54,11 @@ pub struct ModelWatcher {
     busy_threshold: Option<f64>,
 }
 
-const ALL_MODEL_TYPES: &[ModelType] =
-    &[ModelType::Chat, ModelType::Completion, ModelType::Embedding];
+const ALL_MODEL_TYPES: &[ModelType] = &[
+    ModelType::Chat,
+    ModelType::Completions,
+    ModelType::Embedding,
+];
 
 impl ModelWatcher {
     pub fn new(
@@ -236,7 +239,7 @@ impl ModelWatcher {
         } else {
             for model_type in ALL_MODEL_TYPES {
                 if ((chat_model_removed && *model_type == ModelType::Chat)
-                    || (completions_model_removed && *model_type == ModelType::Completion)
+                    || (completions_model_removed && *model_type == ModelType::Completions)
                     || (embeddings_model_removed && *model_type == ModelType::Embedding))
                     && let Some(tx) = &self.model_update_tx
                 {
@@ -273,10 +276,13 @@ impl ModelWatcher {
             }
         };
 
-        match model_entry.model_type {
-            ModelType::Backend => {
-                // A Backend model expects pre-processed requests meaning it's up to us whether we
-                // handle Chat or Completions requests, so handle both.
+        if model_entry.model_input == ModelInput::Tokens
+            && (model_entry.model_type.supports_chat()
+                || model_entry.model_type.supports_completions())
+        {
+            // Case 1: Tokens + (Chat OR Completions OR Both)
+            // A model that expects pre-processed requests meaning it's up to us whether we
+            // handle Chat or Completions requests, so handle whatever the model supports.
 
             let Some(mut card) = card else {
                 anyhow::bail!("Missing model deployment card");
@@ -302,6 +308,8 @@ impl ModelWatcher {
                 None
             };
 
+            // Add chat engine only if the model supports chat
+            if model_entry.model_type.supports_chat() {
                 let chat_engine = entrypoint::build_routed_pipeline::<
                     NvCreateChatCompletionRequest,
                     NvCreateChatCompletionStreamResponse,
@@ -315,8 +323,15 @@ impl ModelWatcher {
                 .await?;
                 self.manager
                     .add_chat_completions_model(&model_entry.name, chat_engine)?;
+            }
 
-                let completions_engine = entrypoint::build_routed_pipeline::<
+            // Add completions engine only if the model supports completions
+            if model_entry.model_type.supports_completions() {
+                let formatter = PromptFormatter::no_op();
+                let PromptFormatter::OAI(formatter) = formatter;
+                let preprocessor =
+                    OpenAIPreprocessor::new_with_formatter(card.clone(), formatter).await?;
+                let completions_engine = entrypoint::build_routed_pipeline_with_preprocessor::<
                     NvCreateCompletionRequest,
                     NvCreateCompletionResponse,
                 >(
@@ -325,36 +340,44 @@ impl ModelWatcher {
                     self.router_mode,
                     self.busy_threshold,
                     kv_chooser,
+                    preprocessor,
                 )
                 .await?;
                 self.manager
                     .add_completions_model(&model_entry.name, completions_engine)?;
             }
-            ModelType::Chat => {
+        } else if model_entry.model_input == ModelInput::Text
+            && model_entry.model_type.supports_chat()
+        {
+            // Case 3: Text + Chat
             let push_router = PushRouter::<
                 NvCreateChatCompletionRequest,
                 Annotated<NvCreateChatCompletionStreamResponse>,
             >::from_client_with_threshold(
-                    client, Default::default(), self.busy_threshold
+                client, self.router_mode, self.busy_threshold
             )
             .await?;
             let engine = Arc::new(push_router);
             self.manager
                 .add_chat_completions_model(&model_entry.name, engine)?;
-            }
-            ModelType::Completion => {
+        } else if model_entry.model_input == ModelInput::Text
+            && model_entry.model_type.supports_completions()
+        {
+            // Case 2: Text + Completions
             let push_router = PushRouter::<
                 NvCreateCompletionRequest,
                 Annotated<NvCreateCompletionResponse>,
             >::from_client_with_threshold(
-                    client, Default::default(), self.busy_threshold
+                client, self.router_mode, self.busy_threshold
             )
             .await?;
             let engine = Arc::new(push_router);
             self.manager
                 .add_completions_model(&model_entry.name, engine)?;
-            }
-            ModelType::Embedding => {
+        } else if model_entry.model_input == ModelInput::Tokens
+            && model_entry.model_type.supports_embedding()
+        {
+            // Case 4: Tokens + Embeddings
             let Some(mut card) = card else {
                 anyhow::bail!("Missing model deployment card for embedding model");
             };
@@ -393,7 +416,14 @@ impl ModelWatcher {
 
             self.manager
                 .add_embeddings_model(&model_entry.name, embedding_engine)?;
-            }
+        } else {
+            // Reject unsupported combinations
+            anyhow::bail!(
+                "Unsupported model configuration: {} with {} input. Supported combinations: \
+                Tokens+(Chat|Completions), Text+Chat, Text+Completions, Tokens+Embeddings",
+                model_entry.model_type,
+                model_entry.model_input.as_str()
+            );
         }
 
         Ok(())

lib/llm/src/entrypoint.rs

@@ -6,7 +6,7 @@
 //! - Connect it to an Input
 
 pub mod input;
-pub use input::build_routed_pipeline;
+pub use input::{build_routed_pipeline, build_routed_pipeline_with_preprocessor};
 
 use std::sync::Arc;
 

lib/llm/src/entrypoint/input.rs

@@ -16,7 +16,7 @@ use std::{
 
 pub mod batch;
 mod common;
-pub use common::build_routed_pipeline;
+pub use common::{build_routed_pipeline, build_routed_pipeline_with_preprocessor};
 pub mod endpoint;
 pub mod grpc;
 pub mod http;