[Doc] Expand Multimodal API Reference

docs/source/api/multimodal/index.md

@@ -2,72 +2,27 @@
 
 # Multi-Modality
 
-```{eval-rst}
-.. currentmodule:: vllm.multimodal
-```
-
 vLLM provides experimental support for multi-modal models through the {mod}`vllm.multimodal` package.
 
 Multi-modal inputs can be passed alongside text and token prompts to [supported models](#supported-mm-models)
 via the `multi_modal_data` field in {class}`vllm.inputs.PromptType`.
 
 Looking to add your own multi-modal model? Please follow the instructions listed [here](#enabling-multimodal-inputs).
 
-
 ## Module Contents
 
-```{eval-rst}
-.. automodule:: vllm.multimodal
-```
-
-### Registry
-
 ```{eval-rst}
 .. autodata:: vllm.multimodal.MULTIMODAL_REGISTRY
 ```
 
-```{eval-rst}
-.. autoclass:: vllm.multimodal.MultiModalRegistry
-    :members:
-    :show-inheritance:
-```
-
-### Base Classes
-
-```{eval-rst}
-.. automodule:: vllm.multimodal.base
-    :members:
-    :show-inheritance:
-```
+## Submodules
 
-### Input Classes
+```{toctree}
+:maxdepth: 1
 
-```{eval-rst}
-.. automodule:: vllm.multimodal.inputs
-    :members:
-    :show-inheritance:
-```
-
-### Audio Classes
-
-```{eval-rst}
-.. automodule:: vllm.multimodal.audio
-    :members:
-    :show-inheritance:
-```
-
-### Image Classes
-
-```{eval-rst}
-.. automodule:: vllm.multimodal.image
-    :members:
-    :show-inheritance:
-```
-
-### Video Classes
-
-```{eval-rst}
-.. automodule:: vllm.multimodal.video
-    :members:
-    :show-inheritance:
+inputs
+parse
+processing
+profiling
+registry
 ```

docs/source/api/multimodal/inputs.md

@@ -0,0 +1,49 @@
+# Input Definitions
+
+## User-facing inputs
+
+```{eval-rst}
+.. autodata:: vllm.multimodal.MultiModalDataDict
+```
+
+## Internal data structures
+
+```{eval-rst}
+.. autoclass:: vllm.multimodal.inputs.PlaceholderRange
+    :members:
+    :show-inheritance:
+```
+
+```{eval-rst}
+.. autodata:: vllm.multimodal.inputs.NestedTensors
+```
+
+```{eval-rst}
+.. autoclass:: vllm.multimodal.inputs.MultiModalFieldElem
+    :members:
+    :show-inheritance:
+```
+
+```{eval-rst}
+.. autoclass:: vllm.multimodal.inputs.MultiModalFieldConfig
+    :members:
+    :show-inheritance:
+```
+
+```{eval-rst}
+.. autoclass:: vllm.multimodal.inputs.MultiModalKwargsItem
+    :members:
+    :show-inheritance:
+```
+
+```{eval-rst}
+.. autoclass:: vllm.multimodal.inputs.MultiModalKwargs
+    :members:
+    :show-inheritance:
+```
+
+```{eval-rst}
+.. autoclass:: vllm.multimodal.inputs.MultiModalInputsV2
+    :members:
+    :show-inheritance:
+```

docs/source/api/multimodal/parse.md

@@ -0,0 +1,9 @@
+# Data Parsing
+
+## Module Contents
+
+```{eval-rst}
+.. automodule:: vllm.multimodal.parse
+    :members:
+    :member-order: bysource
+```

docs/source/api/multimodal/processing.md

@@ -0,0 +1,9 @@
+# Data Processing
+
+## Module Contents
+
+```{eval-rst}
+.. automodule:: vllm.multimodal.processing
+    :members:
+    :member-order: bysource
+```

docs/source/api/multimodal/profiling.md

@@ -0,0 +1,9 @@
+# Memory Profiling
+
+## Module Contents
+
+```{eval-rst}
+.. automodule:: vllm.multimodal.profiling
+    :members:
+    :member-order: bysource
+```

docs/source/api/multimodal/registry.md

@@ -0,0 +1,9 @@
+# Registry
+
+## Module Contents
+
+```{eval-rst}
+.. automodule:: vllm.multimodal.registry
+    :members:
+    :member-order: bysource
+```

vllm/multimodal/parse.py

@@ -13,14 +13,16 @@
 
 from .audio import resample_audio
 from .inputs import (AudioItem, HfAudioItem, HfImageItem, HfVideoItem,
-                     ImageItem, ModalityData, MultiModalDataDict,
-                     NestedTensors, VideoItem)
+                     ImageItem, ModalityData, MultiModalDataDict, VideoItem)
 
 _T = TypeVar("_T")
 _I = TypeVar("_I")
 
 
 class ModalityDataItems(ABC, Generic[_T, _I]):
+    """
+    Represents data items for a modality in :class:`MultiModalDataItems`.
+    """
 
     def __init__(self, data: _T, modality: str) -> None:
         super().__init__()
@@ -69,6 +71,7 @@ def get_passthrough_data(self) -> Mapping[str, object]:
 
 
 class ProcessorBatchItems(ModalityDataItems[Sequence[_T], _T]):
+    """Base class for data items that are arranged in a list."""
 
     def get_count(self) -> int:
         return len(self.data)
@@ -83,7 +86,12 @@ def get_passthrough_data(self) -> Mapping[str, object]:
         return {}
 
 
-class EmbeddingItems(ModalityDataItems[NestedTensors, torch.Tensor]):
+class EmbeddingItems(ModalityDataItems[Union[torch.Tensor, list[torch.Tensor]],
+                                       torch.Tensor]):
+    """
+    Base class for data items that are expressed as a batched embedding tensor,
+    or a list of embedding tensors (one per item).
+    """
 
     def get_count(self) -> int:
         return len(self.data)
@@ -109,7 +117,7 @@ def __init__(self, data: Sequence[HfAudioItem]) -> None:
 
 class AudioEmbeddingItems(EmbeddingItems):
 
-    def __init__(self, data: NestedTensors) -> None:
+    def __init__(self, data: Union[torch.Tensor, list[torch.Tensor]]) -> None:
         super().__init__(data, "audio")
 
 
@@ -137,7 +145,7 @@ def get_image_size(self, item_idx: int) -> ImageSize:
 
 class ImageEmbeddingItems(EmbeddingItems):
 
-    def __init__(self, data: NestedTensors) -> None:
+    def __init__(self, data: Union[torch.Tensor, list[torch.Tensor]]) -> None:
         super().__init__(data, "image")
 
 
@@ -163,7 +171,7 @@ def get_frame_size(self, item_idx: int) -> ImageSize:
 
 class VideoEmbeddingItems(EmbeddingItems):
 
-    def __init__(self, data: NestedTensors) -> None:
+    def __init__(self, data: Union[torch.Tensor, list[torch.Tensor]]) -> None:
         super().__init__(data, "video")
 
 
@@ -172,8 +180,8 @@ def __init__(self, data: NestedTensors) -> None:
 
 class MultiModalDataItems(UserDict[str, ModalityDataItems[Any, Any]]):
     """
-    As :class:`MultiModalDataDict`, but normalized such that each entry
-    corresponds to a list.
+    As :data:`~vllm.multimodal.inputs.MultiModalDataDict`, but normalized
+    such that each entry corresponds to a list.
     """
 
     def get_count(self, modality: str, *, strict: bool = True) -> int:
@@ -226,7 +234,8 @@ def get_items(
 
 class MultiModalDataParser:
     """
-    Parses :class:`MultiModalDataDict` into :class:`MultiModalDataItems`.
+    Parses :data:`~vllm.multimodal.inputs.MultiModalDataDict` into
+    :class:`MultiModalDataItems`.
 
     Args:
         target_sr (float, optional): Enables automatic resampling of audio
@@ -238,7 +247,9 @@ def __init__(self, *, target_sr: Optional[float] = None) -> None:
 
         self.target_sr = target_sr
 
-    def _is_embeddings(self, data: object) -> TypeGuard[NestedTensors]:
+    def _is_embeddings(
+            self, data: object
+    ) -> TypeGuard[Union[torch.Tensor, list[torch.Tensor]]]:
         if isinstance(data, torch.Tensor):
             return data.ndim == 3
         if is_list_of(data, torch.Tensor):

vllm/multimodal/processing.py

@@ -33,20 +33,24 @@
 
 @dataclass
 class PromptReplacement:
+    """
+    Defines how to replace portions of an input prompt with placeholder tokens.
+    """
+
     modality: str
     """The modality for which the replacement is made."""
 
     target: _PromptSeq
-    """The text or token sequence to find and replace."""
+    """The token sequence (or text) to find and replace."""
 
     replacement: Union[Callable[[int], _PromptSeq],
                        _PromptSeq] = field(repr=False)
     """
-    Given the index of the processed item within :attr:`modality`, output the
-    replacement text or token sequence.
+    Given the index of the processed item within :attr:`modality`,
+    output the replacement token sequence (or text).
 
-    For convenience, you can pass in the replacement instead of a function
-    if it does not depend on the input.
+    For convenience, you can directly pass in the replacement token sequence
+    (or text) instead of a function if it does not depend on the input.
     """
 
     def bind(self, tokenizer: AnyTokenizer) -> "BoundPromptReplacement":
@@ -132,6 +136,11 @@ def token_ids(self) -> list[int]:
 
 @dataclass
 class BoundPromptReplacement:
+    """
+    A :class:`PromptReplacement` bound to a tokenizer to automatically
+    convert :attr:`target` and the result of :meth:`get_replacement` between
+    token sequence and text representations.
+    """
     tokenizer: AnyTokenizer = field(repr=False)
     modality: str
 
@@ -144,6 +153,7 @@ def __post_init__(self) -> None:
 
     @property
     def target(self) -> _BoundPromptSequence:
+        """The token sequence (or text) to find and replace."""
         target = self._target
 
         return _BoundPromptSequence(
@@ -153,6 +163,10 @@ def target(self) -> _BoundPromptSequence:
         )
 
     def get_replacement(self, item_idx: int) -> _BoundPromptSequence:
+        """
+        Given the index of the processed item within :attr:`modality`,
+        output the replacement token sequence (or text).
+        """
         replacement = self._replacement
         if callable(replacement):
             cache_key = item_idx
@@ -528,7 +542,7 @@ def put(
 
 
 class BaseProcessingInfo:
-    """Base class containing information to perform processing."""
+    """Base class to provide the information necessary for data processing."""
 
     def __init__(self, ctx: InputProcessingContext) -> None:
         super().__init__()

vllm/multimodal/profiling.py

@@ -19,7 +19,10 @@
 
 @dataclass
 class ProcessorInputs:
-    """Keyword arguments to :meth:`BaseMultiModalProcessor`."""
+    """
+    Represents the keyword arguments to
+    :meth:`vllm.multimodal.processing.BaseMultiModalProcessor.apply`.
+    """
     prompt_text: str
     mm_data: MultiModalDataDict
     hf_processor_mm_kwargs: Mapping[str, object] = field(default_factory=dict)
@@ -47,7 +50,7 @@ def get_dummy_processor_inputs(
     ) -> ProcessorInputs:
         """
         Build the input which, after processing, results in
-        `self.info.get_mm_max_tokens_per_item()` placeholder tokens.
+        :code:`self.info.get_mm_max_tokens_per_item()` placeholder tokens.
         """
         raise NotImplementedError