Support both kimi (thinking) and nemotron (enable_thinking) in chat_template_kwargs

providers/nvidia_nim/request.py

@@ -63,7 +63,9 @@ def build_request_body(request_data: Any, nim: NimSettings) -> dict:
     if request_extra:
         extra_body.update(request_extra)
 
-    extra_body.setdefault("chat_template_kwargs", {"enable_thinking": True})
+    extra_body.setdefault(
+        "chat_template_kwargs", {"thinking": True, "enable_thinking": True}
+    )
     _set_extra(extra_body, "reasoning_budget", max_tokens)
 
     req_top_k = getattr(request_data, "top_k", None)

tests/providers/test_nvidia_nim.py

@@ -100,6 +100,7 @@ async def test_build_request_body(nim_provider):
 
     assert "extra_body" in body
     ctk = body["extra_body"]["chat_template_kwargs"]
+    assert ctk["thinking"] is True
     assert ctk["enable_thinking"] is True
     assert body["extra_body"]["reasoning_budget"] == body["max_tokens"]
 

tests/providers/test_nvidia_nim_request.py

@@ -101,7 +101,10 @@ class TestBuildRequestBody:
         nim = NimSettings()
         body = build_request_body(req, nim)
         extra = body["extra_body"]
-        assert extra["chat_template_kwargs"] == {"enable_thinking": True}
+        assert extra["chat_template_kwargs"] == {
+            "thinking": True,
+            "enable_thinking": True,
+        }
         assert extra["reasoning_budget"] == body["max_tokens"]
 
     def test_no_reasoning_params_in_extra_body(self):