fix(ollama): emit top-level reasoning_effort=none on /v1/chat/completions (#25758)

plugins/model-providers/custom/__init__.py

@@ -4,7 +4,9 @@
 Ollama instances and OpenAI-compatible reasoning endpoints (GLM-5.2 on
 Volcengine ARK, vLLM, llama.cpp). Key quirks:
   - ollama_num_ctx → extra_body.options.num_ctx (local context window)
-  - reasoning_config disabled → extra_body.think = False
+  - reasoning_config disabled → top-level reasoning_effort="none"
+    (Ollama /v1/chat/completions ignores think=False — ollama#14820)
+    + extra_body.think = False for /api/chat and proxies
   - reasoning_config enabled + effort → top-level reasoning_effort
     (the native OpenAI-compatible format GLM/ARK expect; unset omits it
     so the endpoint's server default applies)
@@ -53,6 +55,13 @@ def build_api_kwargs_extras(
             _effort = (reasoning_config.get("effort") or "").strip().lower()
             _enabled = reasoning_config.get("enabled", True)
             if _effort == "none" or _enabled is False:
+                # Ollama's /v1/chat/completions silently ignores
+                # extra_body.think (only /api/chat honours it — ollama#14820)
+                # but respects the top-level reasoning_effort field, so both
+                # are needed to actually stop a thinking-capable model from
+                # reasoning (#25758). Endpoints that recognize neither simply
+                # ignore them.
+                top_level["reasoning_effort"] = "none"
                 extra_body["think"] = False
             elif _effort:
                 top_level["reasoning_effort"] = _effort

scripts/release.py

@@ -46,6 +46,7 @@
 # Auto-extracted from noreply emails + manual overrides
 AUTHOR_MAP = {
     "75556242+webtecnica@users.noreply.github.com": "webtecnica",  # PR #63360 salvage (nous: restore inference-api base_url)
+    "skosarevivan@yandex.ru": "Epoxidex",  # PR #29820 salvage (ollama: top-level reasoning_effort=none; #25758)
     "changhyun.min@gmail.com": "minchang",  # PR #42231 salvage (providers: add Upstage Solar)
     "neo@neodeMac-mini.local": "neo-claw-bot",  # PR #58465 salvage (moa: drop empty user turns from advisory view)
     "2024104039@mails.szu.edu.cn": "pixel4039",  # PR #64420 salvage (streaming: retry zero-chunk streams)

tests/plugins/model_providers/test_custom_profile.py

@@ -49,20 +49,26 @@ def test_no_reasoning_config_emits_nothing(self, custom_profile):
         assert tl == {}
 
     def test_disabled_sends_think_false(self, custom_profile):
-        """enabled=False → extra_body.think = False (Ollama thinking-off flag)."""
+        """enabled=False → reasoning_effort='none' top-level + think=False.
+
+        Both fields are required: Ollama's /v1/chat/completions silently
+        ignores extra_body.think (only /api/chat honours it — ollama#14820)
+        but respects top-level reasoning_effort (#25758). think=False stays
+        for proxies and the native /api/chat path.
+        """
         eb, tl = custom_profile.build_api_kwargs_extras(
             reasoning_config={"enabled": False}, model="glm-5.2"
         )
         assert eb == {"think": False}
-        assert tl == {}
+        assert tl == {"reasoning_effort": "none"}
 
     def test_effort_none_sends_think_false(self, custom_profile):
-        """effort='none' is the disable alias → think=False, no effort."""
+        """effort='none' is the disable alias → same dual emission."""
         eb, tl = custom_profile.build_api_kwargs_extras(
             reasoning_config={"enabled": True, "effort": "none"}, model="glm-5.2"
         )
         assert eb == {"think": False}
-        assert tl == {}
+        assert tl == {"reasoning_effort": "none"}
 
     @pytest.mark.parametrize(
         "effort", ["minimal", "low", "medium", "high", "xhigh", "max"]