[MM][Core] Decouple ViT backend from LM backend (#27061)

Signed-off-by: Roger Wang <hey@rogerw.io>

tests/config/test_multimodal_config.py

@@ -0,0 +1,25 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import pytest
+
+from vllm.attention.backends.registry import _Backend
+from vllm.config.multimodal import MultiModalConfig
+
+
+def test_mm_encoder_attn_backend_str_conversion():
+    config = MultiModalConfig(mm_encoder_attn_backend="FLASH_ATTN")
+    assert config.mm_encoder_attn_backend == _Backend.FLASH_ATTN
+
+
+def test_mm_encoder_attn_backend_invalid():
+    with pytest.raises(ValueError):
+        MultiModalConfig(mm_encoder_attn_backend="not_a_backend")
+
+
+def test_mm_encoder_attn_backend_hash_updates():
+    base_hash = MultiModalConfig().compute_hash()
+    overridden_hash = MultiModalConfig(
+        mm_encoder_attn_backend=_Backend.FLASH_ATTN
+    ).compute_hash()
+    assert base_hash != overridden_hash