chore: add mypy typing to frontend (#6861)

components/src/dynamo/frontend/main.py

@@ -22,7 +22,7 @@ import os
 import signal
 import sys
 from argparse import Namespace
-from typing import Optional
+from typing import TYPE_CHECKING, Any, Optional
 
 import uvloop
 
@@ -41,6 +41,9 @@ from dynamo.runtime.logging import configure_dynamo_logging
 
 from .frontend_args import FrontendArgGroup, FrontendConfig
 
+if TYPE_CHECKING:
+    from .vllm_processor import EngineFactory
+
 configure_dynamo_logging()
 logger = logging.getLogger(__name__)
 
@@ -50,7 +53,7 @@ def setup_engine_factory(
     router_config: RouterConfig,
     config: FrontendConfig,
     vllm_flags: Namespace,
-):
+) -> "EngineFactory":
     """
     When using vllm pre and post processor, create the EngineFactory that
     creates the engines that run requests.
@@ -196,7 +199,7 @@ async def async_main():
         active_prefill_tokens_threshold_frac=config.active_prefill_tokens_threshold_frac,
         decode_fallback=config.decode_fallback,
     )
-    kwargs = {
+    kwargs: dict[str, Any] = {
         "http_host": config.http_host,
         "http_port": config.http_port,
         "kv_cache_block_size": config.kv_cache_block_size,
@@ -245,7 +248,7 @@ async def async_main():
         pass
 
 
-async def graceful_shutdown(runtime):
+async def graceful_shutdown(runtime: DistributedRuntime) -> None:
     """Handle graceful shutdown of the distributed runtime.
 
     Args:
@@ -254,7 +257,7 @@ async def graceful_shutdown(runtime):
     runtime.shutdown()
 
 
-def main():
+def main() -> None:
     """Entry point for the Dynamo frontend CLI."""
     uvloop.run(async_main())
 

components/src/dynamo/frontend/prepost.py

@@ -401,10 +401,10 @@ class StreamingPostProcessor:
         # vLLM output_processor already applies stop-token/stop-string trimming
         # to text. Re-detokenizing from token_ids can reintroduce stop markers.
         delta_text = output.text or ""
-
+        delta: dict[str, Any] = {}
         if self._fast_plain_text:
             if delta_text:
-                delta: dict[str, Any] = {
+                delta = {
                     "role": "assistant",
                     "content": delta_text,
                 }
@@ -542,7 +542,7 @@ class StreamingPostProcessor:
                 # to drain the buffer.
                 choice = self._emit_tool_calls_choice(output)
         elif delta_message.content or delta_message.reasoning:
-            delta: dict[str, Any] = {"role": "assistant"}
+            delta = {"role": "assistant"}
             content = delta_message.content
             if self.in_progress_tool_calls and self._is_control_only_content(content):
                 content = None

components/src/dynamo/frontend/vllm_processor.py

@@ -37,7 +37,7 @@ from dynamo.llm import (
     RouterMode,
     fetch_model,
 )
-from dynamo.runtime import DistributedRuntime
+from dynamo.runtime import Client, DistributedRuntime
 
 from .prepost import (
     StreamingPostProcessor,
@@ -147,6 +147,7 @@ def _preprocess_worker(
     model_name: str,
 ) -> PreprocessWorkerResult:
     """Preprocess a request in a worker process and return a picklable result."""
+    assert _w_input_processor is not None
     pre = preprocess_chat_request_sync(
         request,
         tokenizer=_w_tokenizer,
@@ -271,7 +272,7 @@ class VllmProcessor:
         self,
         tokenizer: TokenizerLike,
         input_processor: InputProcessor,
-        router,  # Client or KvRouter
+        router: Any,  # Client or KvRouter
         output_processor: OutputProcessor,
         tool_parser_class: type[ToolParser] | None,
         reasoning_parser_class: type[ReasoningParser] | None,
@@ -644,7 +645,9 @@ class VllmProcessor:
 
         # --- Phase 1: Preprocess (semaphore held) ---
         try:
+            assert self._worker_semaphore is not None
             async with self._worker_semaphore:
+                assert self.preprocess_pool is not None
                 future = self.preprocess_pool.submit(
                     _preprocess_worker, request, request_id, request["model"]
                 )
@@ -793,7 +796,7 @@ class EngineFactory:
         generate_endpoint = self.runtime.endpoint(
             f"{namespace_name}.{component_name}.{endpoint_name}"
         )
-
+        router: Client | KvRouter
         if self.router_config.router_mode == RouterMode.KV:
             router = KvRouter(
                 endpoint=generate_endpoint,

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

@@ -412,6 +412,16 @@ class ModelDeploymentCard:
         """Deserialize a model deployment card from a JSON string."""
         ...
 
+    def model_type(self) -> ModelType:
+        """Return the model type of this deployment card."""
+        ...
+
+    def source_path(self) -> str:
+        """Return the source path of this deployment card."""
+        ...
+
+    def runtime_config(self) -> Any:
+        """Return the runtime configuration as a dict."""
         ...
 
 class ModelRuntimeConfig:
@@ -927,6 +937,9 @@ class ModelType:
     Images: ModelType
     Audios: ModelType
     Videos: ModelType
+
+    def supports_chat(self) -> bool:
+        """Return True if this model type supports chat."""
         ...
 
 class RouterMode:
@@ -939,6 +952,8 @@ class RouterMode:
 
 class RouterConfig:
     """How to route the request"""
+    router_mode: RouterMode
+    kv_router_config: KvRouterConfig
 
     def __init__(
         self,
@@ -982,6 +997,7 @@ class KvRouterConfig:
         router_prune_target_ratio: float = 0.8,
         router_queue_threshold: Optional[float] = None,
         router_event_threads: int = 4,
+        router_enable_cache_control: bool = False,
     ) -> None:
         """
         Create a KV router configuration.
@@ -1012,6 +1028,8 @@ class KvRouterConfig:
                 If None, queueing is disabled and all requests go directly to the scheduler.
             router_event_threads: Number of event processing threads (default: 4).
                 When > 1, uses a concurrent radix tree with a thread pool.
+            router_enable_cache_control: Enable cache control (PIN with TTL) via the worker's
+                cache_control service mesh endpoint (default: False).
         """
         ...