[Doc] Add model development API Reference (#11884)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-01-09 09:43:40 +00:00
committed by GitHub
parent 1d967acb45
commit 65097ca0af
9 changed files with 54 additions and 9 deletions
+7 -4
View File
@@ -38,13 +38,15 @@ class SupportsMultiModal(Protocol):
to be merged with text embeddings.
The output embeddings must be one of the following formats:
- A list or tuple of 2D tensors, where each tensor corresponds to
each input multimodal data item (e.g, image).
each input multimodal data item (e.g, image).
- A single 3D tensor, with the batch dimension grouping the 2D tensors.
NOTE: The returned multimodal embeddings must be in the same order as
the appearances of their corresponding multimodal data item in the
input prompt.
Note:
The returned multimodal embeddings must be in the same order as
the appearances of their corresponding multimodal data item in the
input prompt.
"""
...
@@ -59,6 +61,7 @@ class SupportsMultiModal(Protocol):
) -> torch.Tensor:
...
@overload
def get_input_embeddings(
self,
input_ids: torch.Tensor,
@@ -35,6 +35,7 @@ T_co = TypeVar("T_co", default=torch.Tensor, covariant=True)
@runtime_checkable
class VllmModel(Protocol[C_co, T_co]):
"""The interface required for all models in vLLM."""
def __init__(
self,
@@ -97,6 +98,7 @@ def is_vllm_model(
@runtime_checkable
class VllmModelForTextGeneration(VllmModel[C_co, T], Protocol[C_co, T]):
"""The interface required for all generative models in vLLM."""
def compute_logits(
self,
@@ -142,6 +144,7 @@ def is_text_generation_model(
@runtime_checkable
class VllmModelForPooling(VllmModel[C_co, T], Protocol[C_co, T]):
"""The interface required for all pooling models in vLLM."""
def pooler(
self,