mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-07-24 16:54:43 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 6

test: cover gemini-native max_tokens forwarding in _build_call_kwargs

Browse source 
Requested in review: builder-level assertions that the gemini-native
branch forwards max_tokens (provider names and the native
generativelanguage.googleapis.com base_url, max_tokens=600), plus a
control showing gemini models on OpenAI-compatible endpoints — including
Gemini's own /openai compatibility endpoint — keep the existing omission
behavior (#34530).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
aui 2026-07-16 08:14:13 +08:00 • committed by Teknium
parent 4ee74fa5df
commit ead9d7b256
1 changed files with 49 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

49
tests/agent/test_auxiliary_client.py
View file

@ -775,6 +775,55 @@ class TestBuildCallKwargsMaxTokens:
)
assert "max_tokens" not in kw3
@pytest.mark.parametrize(
"provider,model,base_url",
[
("gemini", "gemini-2.5-pro", None),
("google", "gemini-2.5-flash", None),
(
"custom",
"gemini-2.5-pro",
"https://generativelanguage.googleapis.com/v1beta",
),
],
)
def test_keeps_max_tokens_for_gemini_native(self, provider, model, base_url):
# Native generateContent maps max_tokens → maxOutputTokens; when it is
# omitted Gemini applies a fixed 65,535-token ceiling, which silently
# turned MoA's reference_max_tokens into a no-op for gemini advisors.
from agent.auxiliary_client import _build_call_kwargs
kwargs = _build_call_kwargs(
provider=provider,
model=model,
messages=[{"role": "user", "content": "hi"}],
max_tokens=600,
base_url=base_url,
)
assert kwargs["max_tokens"] == 600
assert "max_completion_tokens" not in kwargs
def test_omits_max_tokens_for_gemini_model_on_openai_compatible_endpoint(self):
# Control: the gemini branch keys on provider/base_url, never the model
# name. A gemini model served through an OpenAI-compatible endpoint
# keeps the default omission behavior (#34530), including Gemini's own
# /openai compatibility endpoint.
from agent.auxiliary_client import _build_call_kwargs
for provider, base_url in [
("openrouter", "https://openrouter.ai/api/v1"),
("custom", "https://generativelanguage.googleapis.com/v1beta/openai"),
]:
kwargs = _build_call_kwargs(
provider=provider,
model="google/gemini-2.5-pro",
messages=[{"role": "user", "content": "hi"}],
max_tokens=600,
base_url=base_url,
)
assert "max_tokens" not in kwargs
assert "max_completion_tokens" not in kwargs
class TestNousTagsScoping:
def test_tags_injected_when_provider_is_nous(self, monkeypatch):