A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 03:30 UTC
Fix links in multi-modal model contributing page (#18615)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Jul 7, 2025 at 22:13 UTC
042d131f394d9069ea9b472a2f5adad64b6e2df1
2 files changed
+12
-1
docs/contributing/model/multimodal.md
+1
-1
@@ -819,7 +819,7 @@ Each [PromptUpdate][vllm.multimodal.processing.PromptUpdate] instance specifies
819
After you have defined [BaseProcessingInfo][vllm.multimodal.processing.BaseProcessingInfo] (Step 2),
820
[BaseDummyInputsBuilder][vllm.multimodal.profiling.BaseDummyInputsBuilder] (Step 3),
821
and [BaseMultiModalProcessor][vllm.multimodal.processing.BaseMultiModalProcessor] (Step 4),
822
-decorate the model class with {meth}`MULTIMODAL_REGISTRY.register_processor <vllm.multimodal.registry.MultiModalRegistry.register_processor>`
822
+decorate the model class with [MULTIMODAL_REGISTRY.register_processor][vllm.multimodal.processing.MultiModalRegistry.register_processor]
823
to register them to the multi-modal registry:
824
825
```diff
vllm/model_executor/models/interfaces.py
+11
@@ -89,11 +89,22 @@ class SupportsMultiModal(Protocol):
89
) -> Tensor:
90
...
91
92
+ # TODO: Remove this overload once v0 is deprecated
93
@overload
94
def get_input_embeddings(
95
self,
96
input_ids: Tensor,
97
multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
98
+ ) -> Tensor:
99
+ ...
100
+
101
+ def get_input_embeddings(
102
+ self,
103
+ input_ids: Tensor,
104
+ multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
105
+ # Only necessary so that the v0 overload is valid
106
+ # TODO: Remove attn_metadata once v0 is deprecated
107
+ attn_metadata: Optional["AttentionMetadata"] = None,
108
) -> Tensor:
109
"""
110
Returns the input embeddings merged from the text embeddings from