[Bugfix] Check dimensions of multimodal embeddings in V1 (#15816)

DarkLight1337 · web-flow · commit 09e974d483ab · 2025-03-31T09:01:35.000-07:00
Signed-off-by: DarkLight1337 &lt;tlleungac@connect.ust.hk&gt;
diff --git a/examples/offline_inference/vision_language.py b/examples/offline_inference/vision_language.py
@@ -68,7 +68,7 @@ def run_blip2(questions: list[str], modality: str) -> ModelRequestData:
     # See https://huggingface.co/Salesforce/blip2-opt-2.7b/discussions/15#64ff02f3f8cf9e4f5b038262 #noqa
     prompts = [f"Question: {question} Answer:" for question in questions]
     engine_args = EngineArgs(
-        model="Salesforce/blip2-opt-2.7b",
+        model="Salesforce/blip2-opt-6.7b",
         disable_mm_preprocessor_cache=args.disable_mm_preprocessor_cache,
     )
 
@@ -128,7 +128,8 @@ def run_florence2(questions: list[str], modality: str) -> ModelRequestData:
     engine_args = EngineArgs(
         model="microsoft/Florence-2-large",
         tokenizer="facebook/bart-large",
-        max_num_seqs=8,
+        max_model_len=4096,
+        max_num_seqs=2,
         trust_remote_code=True,
         dtype="bfloat16",
         disable_mm_preprocessor_cache=args.disable_mm_preprocessor_cache,
@@ -511,7 +512,7 @@ def run_mllama(questions: list[str], modality: str) -> ModelRequestData:
     engine_args = EngineArgs(
         model=model_name,
         max_model_len=4096,
-        max_num_seqs=16,
+        max_num_seqs=2,
         disable_mm_preprocessor_cache=args.disable_mm_preprocessor_cache,
     )
 
@@ -700,7 +701,7 @@ def run_pixtral_hf(questions: list[str], modality: str) -> ModelRequestData:
     # NOTE: Need L40 (or equivalent) to avoid OOM
     engine_args = EngineArgs(
         model=model_name,
-        max_model_len=8192,
+        max_model_len=6144,
         max_num_seqs=2,
         disable_mm_preprocessor_cache=args.disable_mm_preprocessor_cache,
     )
diff --git a/tests/distributed/test_pipeline_parallel.py b/tests/distributed/test_pipeline_parallel.py
@@ -217,7 +217,7 @@ def iter_params(self, model_id: str):
 
 MULTIMODAL_MODELS = {
     # [Decoder-only]
-    "Salesforce/blip2-opt-2.7b": PPTestSettings.fast(),
+    "Salesforce/blip2-opt-6.7b": PPTestSettings.fast(),
     "facebook/chameleon-7b": PPTestSettings.fast(),
     "adept/fuyu-8b": PPTestSettings.fast(),
     "THUDM/glm-4v-9b": PPTestSettings.fast(),
diff --git a/tests/models/decoder_only/vision_language/test_models.py b/tests/models/decoder_only/vision_language/test_models.py
@@ -34,8 +34,6 @@
     # V1 Test: no way to fall back for head_dim = 80
     # https://github.com/vllm-project/vllm/issues/14524
     "qwen_vl",
-    "h2ovl",
-    "blip2",
     # V1 Test: not enough KV cache space in C1.
     "fuyu",
 ]
@@ -161,7 +159,8 @@
         marks=[large_gpu_mark(min_gb=64)],
     ),
     "blip2": VLMTestInfo(
-        models=["Salesforce/blip2-opt-2.7b"],
+        # TODO: Change back to 2.7b once head_dim = 80 is supported
+        models=["Salesforce/blip2-opt-6.7b"],
         test_type=VLMTestType.IMAGE,
         prompt_formatter=lambda img_prompt: f"Question: {img_prompt} Answer:",
         img_idx_to_prompt=lambda idx: "",
@@ -248,7 +247,8 @@
     "h2ovl": VLMTestInfo(
         models = [
             "h2oai/h2ovl-mississippi-800m",
-            "h2oai/h2ovl-mississippi-2b",
+            # TODO: Re-enable once head_dim = 80 is supported
+            # "h2oai/h2ovl-mississippi-2b",
         ],
         test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
         prompt_formatter=lambda img_prompt: f"<|prompt|>{img_prompt}<|end|><|answer|>", # noqa: E501
diff --git a/tests/models/registry.py b/tests/models/registry.py
@@ -259,7 +259,8 @@ def check_available_online(
 _MULTIMODAL_EXAMPLE_MODELS = {
     # [Decoder-only]
     "AriaForConditionalGeneration": _HfExamplesInfo("rhymes-ai/Aria"),
-    "Blip2ForConditionalGeneration": _HfExamplesInfo("Salesforce/blip2-opt-2.7b"),  # noqa: E501
+    "Blip2ForConditionalGeneration": _HfExamplesInfo("Salesforce/blip2-opt-2.7b",  # noqa: E501
+                                                     extras={"6b": "Salesforce/blip2-opt-6.7b"}),  # noqa: E501
     "ChameleonForConditionalGeneration": _HfExamplesInfo("facebook/chameleon-7b"),  # noqa: E501
     "DeepseekVLV2ForCausalLM": _HfExamplesInfo("deepseek-ai/deepseek-vl2-tiny",  # noqa: E501
                                                 extras={"fork": "Isotr0py/deepseek-vl2-tiny"},  # noqa: E501
diff --git a/vllm/model_executor/models/florence2.py b/vllm/model_executor/models/florence2.py
@@ -875,7 +875,8 @@ def _get_prompt_updates(
     Florence2MultiModalProcessor,
     info=Florence2ProcessingInfo,
     dummy_inputs=Florence2DummyInputsBuilder)
-class Florence2ForConditionalGeneration(nn.Module, SupportsMultiModal):
+class Florence2ForConditionalGeneration(nn.Module, SupportsMultiModal,
+                                        SupportsV0Only):
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
diff --git a/vllm/model_executor/models/fuyu.py b/vllm/model_executor/models/fuyu.py
@@ -39,7 +39,6 @@
                                         PromptUpdate, PromptUpdateDetails)
 from vllm.multimodal.profiling import BaseDummyInputsBuilder, ProcessorInputs
 from vllm.sequence import IntermediateTensors
-from vllm.utils import flatten_2d_lists
 
 from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
 from .utils import (AutoWeightsLoader, flatten_bn, maybe_prefix,
@@ -66,10 +65,13 @@ class FuyuImagePatchInputs(TypedDict):
     This is used to split the embeddings which has the first two dimensions
     flattened just like `flat_data`.
     """
+
     embed_is_patch: Union[torch.Tensor, list[torch.Tensor]]
     """
     A boolean mask indicating which image embeddings correspond
     to patch tokens.
+
+    Shape: `(batch_size * num_images, num_embeds)`
     """
 
 
@@ -322,16 +324,18 @@ def _validate_shape(d: torch.Tensor):
     def _parse_and_validate_image_input(
             self, **kwargs: object) -> Optional[FuyuImagePatchInputs]:
         image_patches = kwargs.pop("image_patches", None)
-        embed_is_patch = kwargs.pop("embed_is_patch", None)
         if image_patches is not None:
             if not isinstance(image_patches, (torch.Tensor, list)):
                 raise ValueError("Incorrect type of image patches. "
                                  f"Got type: {type(image_patches)}")
 
+            embed_is_patch = kwargs.pop("embed_is_patch")
             if not isinstance(embed_is_patch, (torch.Tensor, list)):
                 raise ValueError("Incorrect type of embed_is_patch. "
                                  f"Got type: {type(embed_is_patch)}")
+
             image_patches_flat = flatten_bn(image_patches)
+            embed_is_patch = flatten_bn(embed_is_patch)
 
             return FuyuImagePatchInputs(
                 type="image_patches",
@@ -351,20 +355,21 @@ def _process_image_input(
         assert self.vision_embed_tokens is not None
         vision_embeddings_flat, _ = self.vision_embed_tokens(
             image_patches_flat)
+
         return vision_embeddings_flat.split(patches_per_image, dim=0)
 
     def get_multimodal_embeddings(
             self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
         image_input = self._parse_and_validate_image_input(**kwargs)
         if image_input is None:
             return None
-        vision_embeddings = self._process_image_input(image_input)
-        #return vision_embeddings
-        return flatten_2d_lists(
-            scatter_patch_features(*args) for args in zip(
-                vision_embeddings,
-                image_input["embed_is_patch"],
-            ))
+
+        image_features = self._process_image_input(image_input)
+
+        return scatter_patch_features(
+            image_features,
+            image_input["embed_is_patch"],
+        )
 
     def get_input_embeddings(
         self,
diff --git a/vllm/model_executor/models/gemma3_mm.py b/vllm/model_executor/models/gemma3_mm.py
@@ -613,7 +613,7 @@ def _image_pixels_to_features(
     def _process_image_input(
         self,
         image_input: Gemma3ImageInputs,
-    ) -> tuple[torch.Tensor, ...]:
+    ) -> list[torch.Tensor]:
         assert self.vision_tower is not None
 
         pixel_values = image_input["pixel_values"]
@@ -625,7 +625,9 @@ def _process_image_input(
         )
         image_embeds = self.multi_modal_projector(image_features)
 
-        return image_embeds.split(num_patches.tolist())
+        return [
+            e.flatten(0, 1) for e in image_embeds.split(num_patches.tolist())
+        ]
 
     def get_multimodal_embeddings(
             self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
diff --git a/vllm/model_executor/models/idefics3.py b/vllm/model_executor/models/idefics3.py
@@ -733,15 +733,20 @@ def _process_image_pixels(
             pixel_attention_mask=pixel_attention_mask,
         )
 
-    def _process_image_input(self, image_input: ImageInputs) -> torch.Tensor:
+    def _process_image_input(
+        self,
+        image_input: ImageInputs,
+    ) -> Union[torch.Tensor, list[torch.Tensor]]:
         if image_input["type"] == "image_embeds":
             return image_input["data"]
 
         image_features = self._process_image_pixels(image_input)
         image_features = self.model.connector(image_features)
 
         num_patches = image_input["num_patches"]
-        return image_features.split(num_patches.tolist())
+        return [
+            e.flatten(0, 1) for e in image_features.split(num_patches.tolist())
+        ]
 
     def get_multimodal_embeddings(
             self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
diff --git a/vllm/model_executor/models/llava_next_video.py b/vllm/model_executor/models/llava_next_video.py
@@ -406,20 +406,21 @@ def _process_video_pixels(self, inputs: LlavaNextVideoPixelInputs):
                                                h, w)
             stacked_embeddings = self._video_pixels_to_features(
                 self.vision_tower, stacked_pixels)
-            return stacked_embeddings.view(b, num_frames,
-                                           *stacked_embeddings.shape[1:])
+            embeds = stacked_embeddings.view(b, num_frames,
+                                             *stacked_embeddings.shape[1:])
 
         elif is_list_of(video_pixels, torch.Tensor):
             frames_per_videos = [v.shape[0] for v in video_pixels]
             stacked_pixels = torch.cat(video_pixels, dim=0)
             stacked_embeddings = self._video_pixels_to_features(
                 self.vision_tower, stacked_pixels)
-            return torch.split(stacked_embeddings, frames_per_videos, dim=0)
-
+            embeds = torch.split(stacked_embeddings, frames_per_videos, dim=0)
         else:
             raise ValueError(
                 f"Unsupported type of video input {type(video_pixels)}")
 
+        return [e.flatten(0, 1) for e in embeds]
+
     def get_multimodal_embeddings(
             self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
         video_input = self._parse_and_validate_video_input(**kwargs)
diff --git a/vllm/model_executor/models/minicpmv.py b/vllm/model_executor/models/minicpmv.py
@@ -919,8 +919,11 @@ def _process_vision_input(
 
         image_features_flat = self.get_vision_hidden_states(image_input)
 
-        # Reconstruct the batch dimension
-        return image_features_flat.split(image_input["num_slices"].tolist())
+        num_slices = image_input["num_slices"]
+        return [
+            e.flatten(0, 1)
+            for e in image_features_flat.split(num_slices.tolist())
+        ]
 
     def _process_multimodal_inputs(self, modalities: dict):
         # The result multimodal_embeddings is tuple of tensors, with each
diff --git a/vllm/model_executor/models/vision.py b/vllm/model_executor/models/vision.py
@@ -204,7 +204,7 @@ def get_embed_one(patches_one: torch.Tensor, e_is_patch: torch.Tensor):
             (e_is_patch.shape[0], patches_one.shape[-1]),
             fill_value=torch.nan,
         )
-        embed_one[e_is_patch] = patches_one.flatten(0, -2)
+        embed_one[e_is_patch] = patches_one
         return embed_one
 
     return tuple(
diff --git a/vllm/v1/worker/gpu_model_runner.py b/vllm/v1/worker/gpu_model_runner.py
@@ -41,6 +41,8 @@
 from vllm.v1.worker.gpu_input_batch import CachedRequestState, InputBatch
 from vllm.v1.worker.lora_model_runner_mixin import LoRAModelRunnerMixin
 
+from .utils import sanity_check_mm_encoder_outputs
+
 if TYPE_CHECKING:
     import xgrammar as xgr
 
@@ -867,6 +869,11 @@ def _execute_encoder(self, scheduler_output: "SchedulerOutput"):
             curr_group_outputs = self.model.get_multimodal_embeddings(
                 **batched_mm_inputs)
 
+            sanity_check_mm_encoder_outputs(
+                curr_group_outputs,
+                expected_num_items=len(grouped_mm_inputs),
+            )
+
             for output in curr_group_outputs:
                 encoder_outputs.append(output)
 
@@ -1490,12 +1497,11 @@ def profile_run(self) -> None:
             # Run multimodal encoder.
             dummy_encoder_outputs = self.model.get_multimodal_embeddings(
                 **batched_dummy_mm_inputs)
-            assert len(dummy_encoder_outputs) == max_num_mm_items, (
-                "Expected dimension 0 of encoder outputs to match the number "
-                f"of multimodal data items: {max_num_mm_items}, got "
-                f"{len(dummy_encoder_outputs)=} instead. This is most likely "
-                "due to the 'get_multimodal_embeddings' method of the model "
-                "not implemented correctly.")
+
+            sanity_check_mm_encoder_outputs(
+                dummy_encoder_outputs,
+                expected_num_items=max_num_mm_items,
+            )
 
             # Cache the dummy encoder outputs.
             self.encoder_cache["tmp"] = dict(enumerate(dummy_encoder_outputs))
diff --git a/vllm/v1/worker/tpu_model_runner.py b/vllm/v1/worker/tpu_model_runner.py
@@ -37,6 +37,8 @@
 from vllm.v1.utils import bind_kv_cache
 from vllm.v1.worker.gpu_input_batch import CachedRequestState, InputBatch
 
+from .utils import sanity_check_mm_encoder_outputs
+
 if TYPE_CHECKING:
     from vllm.v1.core.sched.output import SchedulerOutput
 
@@ -512,6 +514,11 @@ def _execute_encoder(self, scheduler_output: "SchedulerOutput"):
             curr_group_outputs = self.model.get_multimodal_embeddings(
                 **batched_mm_inputs)
 
+            sanity_check_mm_encoder_outputs(
+                curr_group_outputs,
+                expected_num_items=len(grouped_mm_inputs),
+            )
+
             for output in curr_group_outputs:
                 encoder_outputs.append(output)
 
diff --git a/vllm/v1/worker/utils.py b/vllm/v1/worker/utils.py
@@ -0,0 +1,29 @@
+# SPDX-License-Identifier: Apache-2.0
+import torch
+
+
+def sanity_check_mm_encoder_outputs(
+    mm_embeddings: object,
+    expected_num_items: int,
+) -> None:
+    """
+    Perform sanity checks for the result of
+    :meth:`vllm.model_executor.models.SupportsMultiModal.get_multimodal_embeddings`.
+    """
+    assert isinstance(mm_embeddings, (list, tuple, torch.Tensor)), (
+        "Expected multimodal embeddings to be a list/tuple of 2D tensors, "
+        f"or a single 3D tensor, but got {type(mm_embeddings)} "
+        "instead. This is most likely due to incorrect implementation "
+        "of the model's `get_multimodal_embeddings` method.")
+
+    assert len(mm_embeddings) == expected_num_items, (
+        "Expected number of multimodal embeddings to match number of "
+        f"input items: {expected_num_items}, but got {len(mm_embeddings)=} "
+        "instead. This is most likely due to incorrect implementation "
+        "of the model's `get_multimodal_embeddings` method.")
+
+    assert all(e.ndim == 2 for e in mm_embeddings), (
+        "Expected multimodal embeddings to be a sequence of 2D tensors, "
+        f"but got tensors with shapes {[e.shape for e in mm_embeddings]} "
+        "instead. This is most likely due to incorrect implementation "
+        "of the model's `get_multimodal_embeddings` method.")
