Skip to content

Personal AI, On Personal Devices

Python

fix(engine): retry without temperature in streaming and OpenRouter cloud paths (#1047)

Arush Wadhawanbd766e4bfc7f
diff --git a/src/openjarvis/engine/cloud.py b/src/openjarvis/engine/cloud.py
index 5dae86c9..17bff9d9 100644
--- a/src/openjarvis/engine/cloud.py
+++ b/src/openjarvis/engine/cloud.py
@@ -249,6 +249,20 @@ def _is_unsupported_temperature_error(exc: Exception) -> bool:
)
+def _chat_completion_with_temperature_retry(
+ client: Any,
+ create_kwargs: Dict[str, Any],
+) -> Any:
+ """Invoke client.chat.completions.create, retrying without temperature on 400."""
+ try:
+ return client.chat.completions.create(**create_kwargs)
+ except Exception as exc:
+ if "temperature" in create_kwargs and _is_unsupported_temperature_error(exc):
+ create_kwargs.pop("temperature", None)
+ return client.chat.completions.create(**create_kwargs)
+ raise
+
+
def estimate_cost(model: str, prompt_tokens: int, completion_tokens: int) -> float:
"""Estimate USD cost based on the hardcoded pricing table."""
# Try exact match first, then prefix match
@@ -690,19 +704,9 @@ class CloudEngine(InferenceEngine):
create_kwargs["response_format"] = response_format
t0 = time.monotonic()
- try:
- resp = self._openai_client.chat.completions.create(**create_kwargs)
- except Exception as exc:
- # Some models reject a non-default temperature with a 400
- # unsupported_value (see #426). Retry once without it rather
- # than failing the user's first prompt.
- if "temperature" in create_kwargs and _is_unsupported_temperature_error(
- exc
- ):
- create_kwargs.pop("temperature", None)
- resp = self._openai_client.chat.completions.create(**create_kwargs)
- else:
- raise
+ resp = _chat_completion_with_temperature_retry(
+ self._openai_client, create_kwargs
+ )
elapsed = time.monotonic() - t0
choice = _first_choice_or_raise(resp, provider="OpenAI", model=model)
usage = resp.usage
@@ -1045,7 +1049,9 @@ class CloudEngine(InferenceEngine):
if tool_choice is not None:
create_kwargs["tool_choice"] = tool_choice
t0 = time.monotonic()
- resp = self._openrouter_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._openrouter_client, create_kwargs
+ )
elapsed = time.monotonic() - t0
choice = _first_choice_or_raise(resp, provider="OpenRouter", model=actual_model)
usage = resp.usage
@@ -1109,7 +1115,9 @@ class CloudEngine(InferenceEngine):
if tool_choice is not None:
create_kwargs["tool_choice"] = tool_choice
t0 = time.monotonic()
- resp = self._atlascloud_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._atlascloud_client, create_kwargs
+ )
elapsed = time.monotonic() - t0
choice = resp.choices[0]
usage = resp.usage
@@ -1167,7 +1175,9 @@ class CloudEngine(InferenceEngine):
"temperature": temperature,
}
t0 = time.monotonic()
- resp = self._minimax_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._minimax_client, create_kwargs
+ )
elapsed = time.monotonic() - t0
choice = _first_choice_or_raise(resp, provider="MiniMax", model=model)
usage = resp.usage
@@ -1217,7 +1227,9 @@ class CloudEngine(InferenceEngine):
"temperature": temperature,
}
t0 = time.monotonic()
- resp = self._deepseek_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._deepseek_client, create_kwargs
+ )
elapsed = time.monotonic() - t0
choice = _first_choice_or_raise(resp, provider="DeepSeek", model=model)
usage = resp.usage
@@ -1393,7 +1405,9 @@ class CloudEngine(InferenceEngine):
}
if not _is_openai_reasoning_model(model):
create_kwargs["temperature"] = temperature
- resp = self._openai_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._openai_client, create_kwargs
+ )
for chunk in resp:
delta = chunk.choices[0].delta if chunk.choices else None
if delta and delta.content:
@@ -1647,7 +1661,9 @@ class CloudEngine(InferenceEngine):
tool_choice = kwargs.pop("tool_choice", None)
if tool_choice is not None:
create_kwargs["tool_choice"] = tool_choice
- resp = self._openrouter_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._openrouter_client, create_kwargs
+ )
for chunk in resp:
delta = chunk.choices[0].delta if chunk.choices else None
if delta and delta.content:
@@ -1679,7 +1695,9 @@ class CloudEngine(InferenceEngine):
tool_choice = kwargs.pop("tool_choice", None)
if tool_choice is not None:
create_kwargs["tool_choice"] = tool_choice
- resp = self._atlascloud_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._atlascloud_client, create_kwargs
+ )
for chunk in resp:
delta = chunk.choices[0].delta if chunk.choices else None
if delta and delta.content:
@@ -1705,7 +1723,9 @@ class CloudEngine(InferenceEngine):
"temperature": temperature,
"stream": True,
}
- resp = self._minimax_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._minimax_client, create_kwargs
+ )
for chunk in resp:
delta = chunk.choices[0].delta if chunk.choices else None
if delta and delta.content:
@@ -1729,7 +1749,9 @@ class CloudEngine(InferenceEngine):
"temperature": temperature,
"stream": True,
}
- resp = self._deepseek_client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(
+ self._deepseek_client, create_kwargs
+ )
for chunk in resp:
delta = chunk.choices[0].delta if chunk.choices else None
if delta and delta.content:
@@ -1825,7 +1847,7 @@ class CloudEngine(InferenceEngine):
}
if not _is_openai_reasoning_model(model):
create_kwargs["temperature"] = temperature
- resp = client.chat.completions.create(**create_kwargs)
+ resp = _chat_completion_with_temperature_retry(client, create_kwargs)
for chunk in resp:
choice = chunk.choices[0] if chunk.choices else None
if not choice:
diff --git a/tests/engine/test_cloud.py b/tests/engine/test_cloud.py
index 5e7a82e6..2d3a624f 100644
--- a/tests/engine/test_cloud.py
+++ b/tests/engine/test_cloud.py
@@ -202,6 +202,18 @@ class TestOpenAIUnsupportedTemperatureRetry:
model="gpt-5",
)
+ def _fake_stream_resp(self):
+ return [
+ SimpleNamespace(
+ choices=[
+ SimpleNamespace(
+ delta=SimpleNamespace(content="stream-ok", tool_calls=None),
+ finish_reason="stop",
+ )
+ ]
+ )
+ ]
+
def test_retries_without_temperature_on_unsupported_value(
self, monkeypatch: pytest.MonkeyPatch
) -> None:
@@ -338,6 +350,160 @@ class TestOpenAIUnsupportedTemperatureRetry:
# No temperature-retry for an unrelated 400 — exactly one attempt.
assert len(calls) == 1
+ @pytest.mark.asyncio
+ async def test_stream_openai_retries_without_temperature(
+ self, monkeypatch: pytest.MonkeyPatch
+ ) -> None:
+ monkeypatch.setenv("OPENAI_API_KEY", "sk-test")
+ monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
+
+ calls: list[dict] = []
+ err = Exception(
+ "Error code: 400 - Unsupported value: 'temperature' does not support 0.7"
+ )
+
+ def create(**kwargs):
+ calls.append(kwargs)
+ if "temperature" in kwargs:
+ raise err
+ return self._fake_stream_resp()
+
+ fake_client = mock.MagicMock()
+ fake_client.chat.completions.create.side_effect = create
+
+ EngineRegistry.register_value("cloud", CloudEngine)
+ engine = CloudEngine()
+ engine._openai_client = fake_client
+
+ tokens = [
+ t
+ async for t in engine.stream(
+ [Message(role=Role.USER, content="Hi")],
+ model="gpt-5",
+ temperature=0.7,
+ )
+ ]
+ assert tokens == ["stream-ok"]
+ assert len(calls) == 2
+ assert "temperature" in calls[0]
+ assert "temperature" not in calls[1]
+
+ @pytest.mark.asyncio
+ async def test_stream_full_openai_retries_without_temperature(
+ self, monkeypatch: pytest.MonkeyPatch
+ ) -> None:
+ monkeypatch.setenv("OPENAI_API_KEY", "sk-test")
+ monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
+
+ calls: list[dict] = []
+ err = Exception(
+ "Error code: 400 - Unsupported value: 'temperature' does not support 0.7"
+ )
+
+ def create(**kwargs):
+ calls.append(kwargs)
+ if "temperature" in kwargs:
+ raise err
+ return self._fake_stream_resp()
+
+ fake_client = mock.MagicMock()
+ fake_client.chat.completions.create.side_effect = create
+
+ EngineRegistry.register_value("cloud", CloudEngine)
+ engine = CloudEngine()
+ engine._openai_client = fake_client
+
+ chunks = [
+ c
+ async for c in engine.stream_full(
+ [Message(role=Role.USER, content="Hi")],
+ model="gpt-5",
+ temperature=0.7,
+ )
+ ]
+ assert len(chunks) == 1
+ assert chunks[0].content == "stream-ok"
+ assert chunks[0].finish_reason == "stop"
+ assert len(calls) == 2
+ assert "temperature" in calls[0]
+ assert "temperature" not in calls[1]
+
+ def test_generate_openrouter_retries_without_temperature(
+ self, monkeypatch: pytest.MonkeyPatch
+ ) -> None:
+ monkeypatch.setenv("OPENROUTER_API_KEY", "sk-or-test")
+ monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+ monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
+
+ calls: list[dict] = []
+ err = Exception(
+ "Error code: 400 - Unsupported parameter: "
+ "'temperature' is not supported with this model."
+ )
+
+ def create(**kwargs):
+ calls.append(kwargs)
+ if "temperature" in kwargs:
+ raise err
+ return self._fake_resp()
+
+ fake_client = mock.MagicMock()
+ fake_client.chat.completions.create.side_effect = create
+
+ EngineRegistry.register_value("cloud", CloudEngine)
+ engine = CloudEngine()
+ engine._openrouter_client = fake_client
+
+ result = engine.generate(
+ [Message(role=Role.USER, content="Hi")],
+ model="openrouter/openai/o3-mini",
+ temperature=0.7,
+ )
+ assert result["content"] == "ok"
+ assert len(calls) == 2
+ assert "temperature" in calls[0]
+ assert "temperature" not in calls[1]
+
+ @pytest.mark.asyncio
+ async def test_stream_openrouter_retries_without_temperature(
+ self, monkeypatch: pytest.MonkeyPatch
+ ) -> None:
+ monkeypatch.setenv("OPENROUTER_API_KEY", "sk-or-test")
+ monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+ monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
+
+ calls: list[dict] = []
+ err = Exception(
+ "Error code: 400 - Unsupported parameter: "
+ "'temperature' is not supported with this model."
+ )
+
+ def create(**kwargs):
+ calls.append(kwargs)
+ if "temperature" in kwargs:
+ raise err
+ return self._fake_stream_resp()
+
+ fake_client = mock.MagicMock()
+ fake_client.chat.completions.create.side_effect = create
+
+ EngineRegistry.register_value("cloud", CloudEngine)
+ engine = CloudEngine()
+ engine._openrouter_client = fake_client
+
+ tokens = [
+ t
+ async for t in engine.stream(
+ [Message(role=Role.USER, content="Hi")],
+ model="openrouter/openai/o3-mini",
+ temperature=0.7,
+ )
+ ]
+ assert tokens == ["stream-ok"]
+ assert len(calls) == 2
+ assert "temperature" in calls[0]
+ assert "temperature" not in calls[1]
+
# ---------------------------------------------------------------------------
# Codex provider support (OpenAI Responses API)
diff --git a/tests/engine/test_cloud_atlascloud.py b/tests/engine/test_cloud_atlascloud.py
index f3f6a25f..a9fbd40d 100644
--- a/tests/engine/test_cloud_atlascloud.py
+++ b/tests/engine/test_cloud_atlascloud.py
@@ -209,6 +209,37 @@ class TestAtlasCloudGenerate:
assert result["usage"]["completion_tokens"] == 5
assert result["finish_reason"] == "stop"
+ def test_generate_retries_without_unsupported_temperature(
+ self, monkeypatch: pytest.MonkeyPatch
+ ) -> None:
+ engine = _make_cloud_engine(monkeypatch)
+ client = mock.MagicMock()
+ calls: list[dict] = []
+
+ def create(**kwargs):
+ calls.append(kwargs)
+ if "temperature" in kwargs:
+ raise Exception(
+ "Error code: 400 - Unsupported value: 'temperature' "
+ "does not support 0.7"
+ )
+ return _fake_response(model="openai/gpt-5.4-mini")
+
+ client.chat.completions.create.side_effect = create
+ engine._atlascloud_client = client
+
+ result = engine.generate(
+ [Message(role=Role.USER, content="Hi")],
+ model="atlascloud/openai/gpt-5.4-mini",
+ temperature=0.7,
+ )
+
+ assert result["content"] == "Hello from Atlas Cloud!"
+ assert len(calls) == 2
+ assert calls[0]["model"] == calls[1]["model"] == "openai/gpt-5.4-mini"
+ assert "temperature" in calls[0]
+ assert "temperature" not in calls[1]
+
@pytest.mark.parametrize("model", _ATLASCLOUD_POPULAR)
def test_every_listed_model_routes_to_the_atlas_client(
self, monkeypatch: pytest.MonkeyPatch, model: str
@@ -326,6 +357,46 @@ class TestAtlasCloudStream:
assert sent["model"] == "openai/gpt-4.1-mini"
assert sent["stream"] is True
+ @pytest.mark.asyncio
+ async def test_stream_retries_without_unsupported_temperature(
+ self, monkeypatch: pytest.MonkeyPatch
+ ) -> None:
+ engine = _make_cloud_engine(monkeypatch)
+ client = mock.MagicMock()
+ calls: list[dict] = []
+
+ def create(**kwargs):
+ calls.append(kwargs)
+ if "temperature" in kwargs:
+ raise Exception(
+ "Error code: 400 - Unsupported value: 'temperature' "
+ "does not support 0.7"
+ )
+ return [
+ SimpleNamespace(
+ choices=[SimpleNamespace(delta=SimpleNamespace(content="ok"))]
+ )
+ ]
+
+ client.chat.completions.create.side_effect = create
+ engine._atlascloud_client = client
+
+ tokens = [
+ token
+ async for token in engine.stream(
+ [Message(role=Role.USER, content="Hi")],
+ model="atlascloud/openai/gpt-5.4-mini",
+ temperature=0.7,
+ )
+ ]
+
+ assert tokens == ["ok"]
+ assert len(calls) == 2
+ assert calls[0]["model"] == calls[1]["model"] == "openai/gpt-5.4-mini"
+ assert calls[0]["stream"] is calls[1]["stream"] is True
+ assert "temperature" in calls[0]
+ assert "temperature" not in calls[1]
+
@pytest.mark.asyncio
async def test_stream_without_a_client_raises(
self, monkeypatch: pytest.MonkeyPatch