feat(models): implement Google Gemma 4 architecture support (MoE, Multimodal, Reasoning, Tool-Use) (#38826)

Signed-off-by: Luciano Martins <lucianommartins@users.noreply.github.com>
Signed-off-by: Luciano Martins <lucianomartins@google.com>
Co-authored-by: Luciano Martins <lucianommartins@users.noreply.github.com>
Co-authored-by: Isotr0py <2037008807@qq.com>
(cherry picked from commit 08ed2b9688)

tests/models/multimodal/generation/test_common.py

@@ -394,6 +394,22 @@ VLM_TEST_SETTINGS = {
         vllm_runner_kwargs={"mm_processor_kwargs": {"do_pan_and_scan": True}},
         patch_hf_runner=model_utils.gemma3_patch_hf_runner,
     ),
+    "gemma4": VLMTestInfo(
+        models=["google/gemma-4-E2B-it"],
+        test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
+        prompt_formatter=lambda img_prompt: f"<bos><start_of_turn>user\n{img_prompt}<end_of_turn>\n<start_of_turn>model\n",  # noqa: E501
+        single_image_prompts=IMAGE_ASSETS.prompts(
+            {
+                "stop_sign": "What's the content in the center of the image?",
+                "cherry_blossom": "What is the season?",
+            }
+        ),
+        multi_image_prompt="Describe the two images in detail.",
+        max_model_len=4096,
+        max_num_seqs=2,
+        auto_cls=AutoModelForImageTextToText,
+        vllm_runner_kwargs={"limit_mm_per_prompt": {"image": 4}},
+    ),
     "granite_vision": VLMTestInfo(
         models=["ibm-granite/granite-vision-3.3-2b"],
         test_type=(VLMTestType.IMAGE),

tests/models/multimodal/processing/test_gemma4.py

@@ -0,0 +1,44 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import pytest
+
+from vllm.multimodal import MULTIMODAL_REGISTRY
+
+from ....conftest import ImageTestAssets
+from ...utils import build_model_context
+
+# TODO: to be updated to "google/gemma-4-e2b-it" once the models are available
+GEMMA4_MODEL_ID = "google/gemma-4-E2B-it"
+
+
+@pytest.mark.parametrize("model_id", [GEMMA4_MODEL_ID])
+def test_limit_mm_per_prompt(
+    image_assets: ImageTestAssets,
+    model_id: str,
+):
+    """Test that limit_mm_per_prompt accurately restricts multiple images."""
+    # We only allow 1 image
+    ctx = build_model_context(
+        model_id,
+        mm_processor_kwargs={},
+        limit_mm_per_prompt={"image": 1},
+    )
+    processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
+
+    # Provide 2 images in the prompt
+    prompt = "<image><image>"
+    # image_assets usually has multiple images
+    images = [asset.pil_image for asset in image_assets][:2]
+    if len(images) < 2:
+        images = [images[0], images[0]]
+
+    mm_data = {"image": images}
+
+    # Expect ValueError when exceeding limit
+    with pytest.raises(ValueError, match="At most 1 image"):
+        processor(
+            prompt,
+            mm_items=processor.info.parse_mm_data(mm_data),
+            hf_processor_mm_kwargs={},
+        )