From 7e5ff49d89e8982ae6d16e1333a49f9f5835cf45 Mon Sep 17 00:00:00 2001
From: Dinesh Reddy <dinesh@vlm.run>
Date: Mon, 23 Mar 2026 12:13:36 -0700
Subject: [PATCH] fix: exclude null fields from API payloads to match Node SDK
 behavior

GenerationConfig.model_dump() was sending all null fields (response_model,
prompt, json_schema, etc.) and AgentSkill null fields (skill_id, name,
source, bundle) in every request. This caused bloated payloads with ~20
unnecessary null fields vs the Node SDK which omits them.

Changes:
- GenerationConfig.model_dump(): default to exclude_none=True, always
  strip the internal response_model field
- document.generate(): only include domain/callback_url when non-None
- execute()/image.generate(): same callback_url cleanup

Made-with: Cursor
---
 vlmrun/client/predictions.py | 55 ++++++++++++++++++++----------------
 vlmrun/client/types.py       | 11 ++++----
 2 files changed, 37 insertions(+), 29 deletions(-)

diff --git a/vlmrun/client/predictions.py b/vlmrun/client/predictions.py
index 09eee42..06b577f 100644
--- a/vlmrun/client/predictions.py
+++ b/vlmrun/client/predictions.py
@@ -323,17 +323,19 @@ def generate(
             additional_kwargs["config"] = config.model_dump()
         if metadata:
             additional_kwargs["metadata"] = metadata.model_dump()
+        data: dict = {
+            "model": model,
+            "images": images_data,
+            "domain": domain,
+            "batch": batch,
+            **additional_kwargs,
+        }
+        if callback_url is not None:
+            data["callback_url"] = callback_url
         response, status_code, headers = self._requestor.request(
             method="POST",
             url="image/generate",
-            data={
-                "model": model,
-                "images": images_data,
-                "domain": domain,
-                "batch": batch,
-                "callback_url": callback_url,
-                **additional_kwargs,
-            },
+            data=data,
         )
         if not isinstance(response, dict):
             raise TypeError("Expected dict response")
@@ -453,17 +455,20 @@ def generate(
                 additional_kwargs["config"] = config.model_dump()
             if metadata:
                 additional_kwargs["metadata"] = metadata.model_dump()
+            data: dict = {
+                "model": model,
+                "url" if is_url else "file_id": file_or_url,
+                "batch": batch,
+                **additional_kwargs,
+            }
+            if domain is not None:
+                data["domain"] = domain
+            if callback_url is not None:
+                data["callback_url"] = callback_url
             response, status_code, headers = self._requestor.request(
                 method="POST",
                 url=f"{route}/generate",
-                data={
-                    "model": model,
-                    "url" if is_url else "file_id": file_or_url,
-                    "domain": domain,
-                    "batch": batch,
-                    "callback_url": callback_url,
-                    **additional_kwargs,
-                },
+                data=data,
             )
             if not isinstance(response, dict):
                 raise TypeError("Expected dict response")
@@ -519,17 +524,19 @@ def execute(
                 additional_kwargs["config"] = config.model_dump()
             if metadata:
                 additional_kwargs["metadata"] = metadata.model_dump()
+            data: dict = {
+                "name": name,
+                "version": version,
+                "url" if is_url else "file_id": file_or_url,
+                "batch": batch,
+                **additional_kwargs,
+            }
+            if callback_url is not None:
+                data["callback_url"] = callback_url
             response, status_code, headers = self._requestor.request(
                 method="POST",
                 url=f"{route}/execute",
-                data={
-                    "name": name,
-                    "version": version,
-                    "url" if is_url else "file_id": file_or_url,
-                    "batch": batch,
-                    "callback_url": callback_url,
-                    **additional_kwargs,
-                },
+                data=data,
             )
             if not isinstance(response, dict):
                 raise TypeError("Expected dict response")
diff --git a/vlmrun/client/types.py b/vlmrun/client/types.py
index 8aceef6..28c418d 100644
--- a/vlmrun/client/types.py
+++ b/vlmrun/client/types.py
@@ -505,20 +505,21 @@ class GenerationConfig(BaseModel):
 
     def model_dump(self, **kwargs) -> dict:
         """Dump the config as a dictionary, converting response_model to json_schema if present."""
-        data = super().model_dump(**kwargs)
-
         if self.response_model and self.json_schema:
             raise ValueError(
                 "`response_model` and `json_schema` cannot be used together"
             )
 
+        kwargs.setdefault("exclude_none", True)
+        data = super().model_dump(**kwargs)
+
         if self.response_model is not None:
             assert (
                 self.json_schema is None
             ), "`response_model` and `json_schema` cannot be used together"
-            json_schema = self.response_model.model_json_schema()
-            data["json_schema"] = json_schema
-            data.pop("response_model", None)
+            data["json_schema"] = self.response_model.model_json_schema()
+
+        data.pop("response_model", None)
 
         return data