Merge branch 'main' into patches

Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
2026-10-09 17:41:13 +00:00
1122 changed files with 60114 additions and 20956 deletions
+5 -18
View File
@@ -21,6 +21,7 @@ from typing import (
ClassVar,
Literal,
Protocol,
Self,
TypeAlias,
overload,
runtime_checkable,
@@ -31,7 +32,7 @@ import torch
import torch.nn as nn
from torch import Tensor
from transformers.models.whisper.tokenization_whisper import LANGUAGES
from typing_extensions import Self, TypeIs, deprecated
from typing_extensions import TypeIs
from vllm.logger import init_logger
from vllm.model_executor.layers.quantization import QuantizationConfig
@@ -423,17 +424,6 @@ class SupportsMultiModal(SupportsMultiModalEmbeddings, Protocol):
yield
@deprecated(
"get_num_mm_encoder_tokens is deprecated; use get_mm_lora_token_counts instead."
)
def get_num_mm_encoder_tokens(self, num_image_tokens: int) -> int: ...
@deprecated(
"get_num_mm_connector_tokens is deprecated; use "
"get_mm_lora_token_counts instead."
)
def get_num_mm_connector_tokens(self, num_vision_tokens: int) -> int: ...
def get_mm_lora_token_counts(
self,
*,
@@ -447,12 +437,9 @@ class SupportsMultiModal(SupportsMultiModalEmbeddings, Protocol):
connector forwards. Models with multiple modalities can override this
when each modality has different encoder padding or pooling behavior.
"""
del modality, mm_kwargs
num_encoder_tokens = self.get_num_mm_encoder_tokens(num_mm_embeds)
num_connector_tokens = self.get_num_mm_connector_tokens(num_encoder_tokens)
return (
num_encoder_tokens,
num_connector_tokens if isinstance(num_connector_tokens, int) else None,
raise NotImplementedError(
f"{type(self).__name__} must implement get_mm_lora_token_counts "
"to support tower/connector LoRA."
)
def _embed_text_input_ids(