[MM][Gemma4] Use video profiling hints in encoder budget (#41837)

Signed-off-by: lesj0610 <lesj0610@users.noreply.github.com>
Co-authored-by: lesj0610 <lesj0610@users.noreply.github.com>
This commit is contained in:
SeongJun Lee
2026-05-07 05:46:04 -07:00
committed by GitHub
co-authored by lesj0610
parent b3945cc316
commit f650ace6de
3 changed files with 50 additions and 4 deletions
+6 -3
View File
@@ -2,7 +2,7 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import warnings
from collections.abc import Sequence
from collections.abc import Mapping, Sequence
from dataclasses import dataclass
from typing import Any
@@ -277,7 +277,7 @@ def build_model_context(
dtype: ModelDType = "auto",
model_config_kwargs: dict[str, Any] | None = None,
mm_processor_kwargs: dict[str, Any] | None = None,
limit_mm_per_prompt: dict[str, int] | None = None,
limit_mm_per_prompt: Mapping[str, int | Mapping[str, int]] | None = None,
mm_processor_cache_gb: int = 0,
):
"""Creates an InputProcessingContext for a given model.
@@ -300,7 +300,10 @@ def build_model_context(
)
model_config_kwargs = model_config_kwargs or {}
limit_mm_per_prompt = limit_mm_per_prompt or {}
limit_mm_per_prompt = {
modality: dict(limit) if isinstance(limit, Mapping) else limit
for modality, limit in (limit_mm_per_prompt or {}).items()
}
model_config = ModelConfig(
model_id,
runner=runner,