[Frontend] Replace legacy Gemma4 parsers with engine-based implementation (#45588)

Signed-off-by: Ben Browning <bbrownin@redhat.com>
Co-authored-by: Flora Feng <4florafeng@gmail.com>
This commit is contained in:
Ben Browning
2026-06-15 21:34:07 +00:00
committed by GitHub
co-authored by Flora Feng
parent 25ee659db0
commit 76a373eff4
20 changed files with 2809 additions and 1333 deletions
@@ -30,7 +30,9 @@ from openai.types.responses.tool_param import FunctionToolParam
from vllm.entrypoints.openai.responses.protocol import ResponsesRequest
from vllm.tool_parsers.abstract_tool_parser import ToolParser
from vllm.tool_parsers.gemma4_tool_parser import Gemma4ToolParser
from vllm.tool_parsers.gemma4_engine_tool_parser import (
Gemma4EngineToolParser as Gemma4ToolParser,
)
def _get_weather_tool() -> FunctionToolParam:
@@ -59,10 +61,16 @@ def _build_responses_request(*, tool_choice: str) -> ResponsesRequest:
class _StubTokenizer:
"""Minimal tokenizer stub to satisfy ``Gemma4ToolParser.__init__``."""
"""Minimal tokenizer stub to satisfy ``Gemma4EngineToolParser.__init__``."""
def get_vocab(self) -> dict[str, int]:
return {"<|tool_call>": 256_000, "<tool_call|>": 256_001, '<|"|>': 52}
return {
"<|tool_call>": 256_000,
"<tool_call|>": 256_001,
'<|"|>': 52,
"<|channel>": 256_002,
"<channel|>": 256_003,
}
def test_gemma4_adjust_request_sets_skip_special_tokens_on_responses() -> None:
@@ -74,15 +82,14 @@ def test_gemma4_adjust_request_sets_skip_special_tokens_on_responses() -> None:
path, causing raw ``call:fn{...}`` text to leak via
``response.output_text.delta``.
"""
parser = Gemma4ToolParser.__new__(Gemma4ToolParser)
parser.model_tokenizer = _StubTokenizer()
parser = Gemma4ToolParser(_StubTokenizer())
request = _build_responses_request(tool_choice="auto")
assert request.skip_special_tokens is True, (
"Precondition: ResponsesRequest.skip_special_tokens default is True"
)
Gemma4ToolParser.adjust_request(parser, request)
parser.adjust_request(request)
assert request.skip_special_tokens is False