Migrate Paligemma inputs to TensorSchema (#23470)

Signed-off-by: Benji Beck <benjibeck@meta.com>

vllm/model_executor/models/paligemma.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 from collections.abc import Iterable, Mapping, Sequence
-from typing import Literal, Optional, TypedDict, Union
+from typing import Annotated, Literal, Optional, Union
 
 import torch
 from torch import nn
@@ -21,6 +21,7 @@ from vllm.multimodal.processing import (BaseMultiModalProcessor,
                                         PromptUpdateDetails)
 from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
+from vllm.utils.tensor_schema import TensorSchema, TensorShape
 
 from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
 from .siglip import SiglipVisionModel
@@ -32,19 +33,27 @@ from .vision import get_vision_encoder_info
 logger = init_logger(__name__)
 
 
-class PaliGemmaImagePixelInputs(TypedDict):
-    type: Literal["pixel_values"]
-    data: torch.Tensor
-    """Shape: `(batch_size * num_images, num_channels, height, width)`"""
-
+class PaliGemmaImagePixelInputs(TensorSchema):
+    """
+    Dimensions:
+        - bn: Batch size * number of images
+        - c: Number of channels (3)
+        - h: Height
+        - w: Width
+    """
+    type: Literal["pixel_values"] = "pixel_values"
+    data: Annotated[torch.Tensor, TensorShape("bn", 3, "h", "w")]
 
-class PaliGemmaImageEmbeddingInputs(TypedDict):
-    type: Literal["image_embeds"]
-    data: torch.Tensor
-    """Shape: `(batch_size * num_images, image_feature_size, hidden_size)`
 
-    `hidden_size` must match the hidden size of language model backbone.
+class PaliGemmaImageEmbeddingInputs(TensorSchema):
+    """
+    Dimensions:
+        - bn: Batch size * number of images
+        - ifs: Image feature size
+        - hs: Hidden size (must match language model backbone)
     """
+    type: Literal["image_embeds"] = "image_embeds"
+    data: Annotated[torch.Tensor, TensorShape("bn", "ifs", "hs")]
 
 
 PaliGemmaImageInputs = Union[PaliGemmaImagePixelInputs,
@@ -279,19 +288,6 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsMultiModal,
         self.make_empty_intermediate_tensors = (
             self.language_model.make_empty_intermediate_tensors)
 
-    def _validate_pixel_values(self, data: torch.Tensor) -> torch.Tensor:
-        h = w = self.config.vision_config.image_size
-        expected_dims = (3, h, w)
-        actual_dims = tuple(data.shape[1:])
-
-        if actual_dims != expected_dims:
-            expected_expr = ("batch_size", *map(str, expected_dims))
-            raise ValueError(
-                f"The expected shape of pixel values is {expected_expr}. "
-                f"You supplied {tuple(data.shape)}.")
-
-        return data
-
     def _parse_and_validate_image_input(
             self, **kwargs: object) -> Optional[PaliGemmaImageInputs]:
         pixel_values = kwargs.pop("pixel_values", None)
@@ -301,22 +297,17 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsMultiModal,
             return None
 
         if pixel_values is not None:
-            if not isinstance(pixel_values, (torch.Tensor, list)):
-                raise ValueError("Incorrect type of pixel values. "
-                                 f"Got type: {type(pixel_values)}")
-
             pixel_values = flatten_bn(pixel_values, concat=True)
 
-            return PaliGemmaImagePixelInputs(
-                type="pixel_values",
-                data=self._validate_pixel_values(pixel_values),
-            )
+            h = w = self.config.vision_config.image_size
+            return PaliGemmaImagePixelInputs(type="pixel_values",
+                                             data=pixel_values,
+                                             resolve_bindings={
+                                                 "h": h,
+                                                 "w": w
+                                             })
 
         if image_embeds is not None:
-            if not isinstance(image_embeds, torch.Tensor):
-                raise ValueError("Incorrect type of image embeddings. "
-                                 f"Got type: {type(image_embeds)}")
-
             image_embeds = flatten_bn(image_embeds, concat=True)
 
             return PaliGemmaImageEmbeddingInputs(