[Bugfix][Gemma4] Align parser enable_thinking default with template (#52430)

Signed-off-by: lxy-alexander <lxy_alexander@outlook.com>
Co-authored-by: Chauncey <chaunceyjiang@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Alexander Lee
2026-08-18 12:41:26 +00:00
committed by GitHub
co-authored by Chauncey mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent be3f614ff1
commit d29dc3ab87
3 changed files with 31 additions and 9 deletions
@@ -170,7 +170,7 @@ def mock_tokenizer():
@pytest.fixture
def parser(mock_tokenizer):
return Gemma4Parser(mock_tokenizer)
return Gemma4Parser(mock_tokenizer, chat_template_kwargs={"enable_thinking": True})
@pytest.fixture
@@ -289,7 +289,10 @@ class TestGemma4PromptOpenReasoning:
@pytest.fixture
def open_reasoning_parser(self, open_reasoning_tokenizer):
return Gemma4Parser(open_reasoning_tokenizer)
return Gemma4Parser(
open_reasoning_tokenizer,
chat_template_kwargs={"enable_thinking": True},
)
@staticmethod
def _prompt_ids_open_channel() -> list[int]:
@@ -394,7 +397,9 @@ class TestGemma4PreInitReasoningRobustness:
@pytest.fixture
def pre_init_parser(self, pre_init_tokenizer):
return Gemma4Parser(pre_init_tokenizer)
return Gemma4Parser(
pre_init_tokenizer, chat_template_kwargs={"enable_thinking": True}
)
def test_model_emitted_channel_open_after_new_turn(
self, pre_init_parser, pre_init_tokenizer, request_obj
@@ -489,7 +494,9 @@ class TestGemma4ChannelLessOutputConsistency:
@pytest.fixture
def plain_parser(self, plain_tokenizer):
return Gemma4Parser(plain_tokenizer)
return Gemma4Parser(
plain_tokenizer, chat_template_kwargs={"enable_thinking": True}
)
def test_streaming_channel_less_output_is_content(
self, plain_parser, plain_tokenizer, request_obj
@@ -521,7 +528,9 @@ class TestGemma4ChannelLessOutputConsistency:
assert reasoning is None
assert content == _PLAIN_ANSWER_TEXT
stream_parser = Gemma4Parser(plain_tokenizer)
stream_parser = Gemma4Parser(
plain_tokenizer, chat_template_kwargs={"enable_thinking": True}
)
results = _stream_tokens_batched(
stream_parser,
plain_tokenizer,