[UX] Add --language-model-only for hybrid models (#34120)

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

vllm/config/model.py

@@ -297,6 +297,7 @@ class ModelConfig:
     multimodal_config: MultiModalConfig | None = None
     """Configuration for multimodal model. If `None`, this will be inferred
     from the architecture of `self.model`."""
+    language_model_only: InitVar[bool] = False
     limit_mm_per_prompt: InitVar[dict[str, int | dict[str, int]] | None] = None
     enable_mm_embeds: InitVar[bool | None] = None
     media_io_kwargs: InitVar[dict[str, dict[str, Any]] | None] = None
@@ -411,6 +412,7 @@ class ModelConfig:
     def __post_init__(
         self,
         # Multimodal config init vars
+        language_model_only: bool,
         limit_mm_per_prompt: dict[str, int | dict[str, int]] | None,
         enable_mm_embeds: bool | None,
         media_io_kwargs: dict[str, dict[str, Any]] | None,
@@ -576,6 +578,7 @@ class ModelConfig:
                 mm_encoder_tp_mode = "weights"
 
             mm_config_kwargs = dict(
+                language_model_only=language_model_only,
                 limit_per_prompt=limit_mm_per_prompt,
                 enable_mm_embeds=enable_mm_embeds,
                 media_io_kwargs=media_io_kwargs,

vllm/config/multimodal.py

@@ -54,8 +54,12 @@ DummyOptions: TypeAlias = (
 class MultiModalConfig:
     """Controls the behavior of multimodal models."""
 
+    language_model_only: bool = False
+    """If True, disables all multimodal inputs by setting all modality limits
+    to 0. Equivalent to setting --limit-mm-per-prompt to 0 for every
+    modality."""
     limit_per_prompt: dict[str, DummyOptions] = Field(default_factory=dict)
-    """The maximum number of input items and options allowed per 
+    """The maximum number of input items and options allowed per
         prompt for each modality.
     Defaults to 999 for each modality.
 
@@ -63,11 +67,11 @@ class MultiModalConfig:
         {"image": 16, "video": 2}
 
     Configurable format (with options):
-        {"video": {"count": 1, "num_frames": 32, "width": 512, "height": 512}, 
+        {"video": {"count": 1, "num_frames": 32, "width": 512, "height": 512},
         "image": {"count": 5, "width": 512, "height": 512}}
 
     Mixed format (combining both):
-        {"image": 16, "video": {"count": 1, "num_frames": 32, "width": 512, 
+        {"image": 16, "video": {"count": 1, "num_frames": 32, "width": 512,
         "height": 512}}
     """
     enable_mm_embeds: bool = False
@@ -215,6 +219,7 @@ class MultiModalConfig:
         the final hidden states.
         """
         factors: list[Any] = [
+            self.language_model_only,
             self.mm_encoder_attn_backend.name
             if self.mm_encoder_attn_backend is not None
             else None,
@@ -228,6 +233,9 @@ class MultiModalConfig:
         Get the maximum number of input items allowed per prompt
         for the given modality (backward compatible).
         """
+        if self.language_model_only:
+            return 0
+
         limit_data = self.limit_per_prompt.get(modality)
 
         if limit_data is None: