[VLM] Implement merged multimodal processor for Mllama (#11427)

This commit is contained in:
Isotr0py
2025-02-12 20:26:21 -08:00
committed by GitHub
parent d88c8666a1
commit bc55d13070
8 changed files with 444 additions and 221 deletions
+16
View File
@@ -739,3 +739,19 @@ class MultiModalInputs(TypedDict):
For each modality, information about the placeholder tokens in
:code:`prompt_token_ids`.
"""
class MultiModalEncDecInputs(MultiModalInputs):
"""
Represents the outputs of :class:`vllm.multimodal.EncDecMultiModalProcessor`
ready to be passed to vLLM internals.
"""
encoder_prompt: str
"""The processed encoder prompt text."""
encoder_prompt_token_ids: list[int]
"""The processed token IDs of the encoder prompt."""
encoder_token_type_ids: NotRequired[list[int]]
"""The token type IDs of the encoder prompt."""