feat(inference): forward extra_body to remote VLM (DashScope enable_thinking)

src/trio_core/api/routers/inference.py

@@ -194,6 +194,14 @@ class DescribeRequest(BaseModel):
             "swap models per-request and will log a warning, then ignore."
         ),
     )
+    extra_body: dict | None = Field(
+        default=None,
+        description=(
+            "Backend-specific kwargs forwarded as the OpenAI SDK extra_body "
+            "(e.g. DashScope's enable_thinking). Honored by RemoteHTTPBackend; "
+            "ignored by local backends."
+        ),
+    )
 
 
 class DescribeResponse(BaseModel):
@@ -590,6 +598,7 @@ def _sync_describe():
                 max_tokens=req.max_tokens,
                 response_format=req.response_format,
                 model=req.model,
+                extra_body=req.extra_body,
             )
 
         try:

src/trio_core/backends/base.py

@@ -114,6 +114,7 @@ def generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> GenerationResult:
         """Run inference on video frames.
 
@@ -127,6 +128,9 @@ def generate(
                 Honored by ``RemoteHTTPBackend`` (passed through to the
                 upstream chat.completions call); local backends cannot swap
                 models per-request and log a one-shot warning, then ignore.
+            extra_body: Backend-specific kwargs forwarded as the OpenAI
+                SDK ``extra_body`` (e.g. DashScope's ``enable_thinking``).
+                Honored by ``RemoteHTTPBackend``; ignored by local backends.
 
         Returns:
             GenerationResult with text and metrics.
@@ -144,6 +148,7 @@ def stream_generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> Generator[StreamChunk, None, None]:
         """Stream inference token by token."""
         ...

src/trio_core/backends/compressed.py

@@ -59,8 +59,9 @@ def generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> GenerationResult:
-        del response_format  # remote-only spec; ignored by compressed local backend
+        del response_format, extra_body  # remote-only specs; ignored by compressed local backend
         self._warn_model_override_once(model)
         tic = time.perf_counter()
         y, prompt_cache, prompt_token_count = self._custom_prefill(

src/trio_core/backends/mlx.py

@@ -499,10 +499,11 @@ def generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> GenerationResult:
-        # response_format is a remote-only structured-output spec; local
-        # MLX inference doesn't honor it — ignored.
-        del response_format
+        # response_format / extra_body are remote-only specs; local MLX
+        # inference doesn't honor them — ignored.
+        del response_format, extra_body
         self._warn_model_override_once(model)
         formatted, kwargs = self._prepare(frames, prompt)
         input_ids = kwargs.pop("input_ids")

src/trio_core/backends/remote.py

@@ -97,6 +97,7 @@ def generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> GenerationResult:
         t0 = time.monotonic()
         uris = _frames_to_data_uris(frames)
@@ -118,6 +119,8 @@ def generate(
         extra_kwargs: dict[str, object] = {}
         if response_format is not None:
             extra_kwargs["response_format"] = response_format
+        if extra_body is not None:
+            extra_kwargs["extra_body"] = extra_body
 
         effective_model = model or self._remote_model
         response = self._client.chat.completions.create(
@@ -170,6 +173,7 @@ def stream_generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> Generator[StreamChunk, None, None]:
         """Stream from remote VLM API.
 

src/trio_core/backends/tome.py

@@ -228,9 +228,10 @@ def generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> GenerationResult:
         """Run inference with ToMe-compressed vision tokens."""
-        del response_format  # remote-only spec; ignored by ToMe local backend
+        del response_format, extra_body  # remote-only specs; ignored by ToMe local backend
         self._warn_model_override_once(model)
         formatted, kwargs = self._prepare(frames, prompt)
 

src/trio_core/backends/transformers.py

@@ -60,10 +60,11 @@ def generate(
         top_p: float = 1.0,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> GenerationResult:
         import torch
 
-        del response_format  # remote-only spec; ignored by local transformers backend
+        del response_format, extra_body  # remote-only specs; ignored by local transformers backend
         self._warn_model_override_once(model)
         inputs = self._prepare(frames, prompt)
 

src/trio_core/engine.py

@@ -243,6 +243,7 @@ def analyze_video(
         temperature: float | None = None,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> VideoResult:
         """Analyze a video with the loaded VLM.
 
@@ -315,6 +316,7 @@ def analyze_video(
                     top_p=self.config.top_p,
                     response_format=response_format,
                     model=model,
+                    extra_body=extra_body,
                 )
 
         # ── Phase 3: Postprocess ─────────────────────────────────────────
@@ -433,6 +435,7 @@ def analyze_frame(
         temperature: float | None = None,
         response_format: dict | None = None,
         model: str | None = None,
+        extra_body: dict | None = None,
     ) -> VideoResult:
         """Analyze a single frame (image). Convenience wrapper."""
         if frame.ndim == 3:
@@ -446,6 +449,7 @@ def analyze_frame(
             temperature=temperature,
             response_format=response_format,
             model=model,
+            extra_body=extra_body,
         )
 
     def health(self) -> dict[str, Any]: