[Doc] Convert Sphinx directives ( {class}, {meth}, {attr}, ...) to MkDocs format for better documentation linking (#18663)

Signed-off-by: Zerohertz <ohg3417@gmail.com>

vllm/model_executor/models/interfaces.py

@@ -226,9 +226,11 @@ class SupportsPP(Protocol):
         intermediate_tensors: Optional["IntermediateTensors"],
     ) -> Union[Tensor, "IntermediateTensors"]:
         """
-        Accept {class}`IntermediateTensors` when PP rank > 0.
+        Accept [`IntermediateTensors`][vllm.sequence.IntermediateTensors] when
+        PP rank > 0.
 
-        Return {class}`IntermediateTensors` only for the last PP rank.
+        Return [`IntermediateTensors`][vllm.sequence.IntermediateTensors] only
+        for the last PP rank.
         """
         ...
 

vllm/model_executor/models/molmo.py

@@ -965,7 +965,7 @@ def select_tiling(
 
 class MolmoProcessorWrapper:
     """
-    Wraps {class}`MolmoProcessor` so that it can be called directly.
+    Wraps `MolmoProcessor` so that it can be called directly.
 
     The original definition can be found here:
     https://huggingface.co/allenai/Molmo-7B-D-0924/blob/main/preprocessing_molmo.py

vllm/model_executor/models/pixtral.py

@@ -67,14 +67,14 @@ class PixtralImagePixelInputs(TypedDict):
     """
     Shape: `(batch_size * num_images, num_channels, image_width, image_height)`
 
-    The result of stacking {attr}`ImageEncoding.tokens` from each prompt.
+    The result of stacking `ImageEncoding.tokens` from each prompt.
     """
 
 
 class PixtralProcessorAdapter:
     """
     Provide a HF-compatible interface for
-    {class}`mistral_common.tokens.tokenizers.multimodal.ImageEncoder`.
+    `mistral_common.tokens.tokenizers.multimodal.ImageEncoder`.
     """
 
     def __init__(self, tokenizer: MistralTokenizer) -> None:

vllm/model_executor/models/qwen_vl.py

@@ -382,7 +382,8 @@ def _get_tokenizer_without_image_pad(
         tokenizer: PreTrainedTokenizer) -> PreTrainedTokenizer:
     """
     The logic of adding image pad tokens should only be applied in
-    {class}`QwenVLProcessor`, so they are patched out here.
+    [`QwenVLProcessor`][vllm.model_executor.models.qwen_vl.QwenVLProcessor],
+    so they are patched out here.
 
     The definition of the wrapped tokenizer can be found here:
     https://huggingface.co/Qwen/Qwen-VL/blob/main/tokenization_qwen.py

vllm/model_executor/models/registry.py

@@ -383,7 +383,7 @@ class _ModelRegistry:
 
         `model_cls` can be either:
 
-        - A {class}`torch.nn.Module` class directly referencing the model.
+        - A [`torch.nn.Module`][] class directly referencing the model.
         - A string in the format `<module>:<class>` which can be used to
           lazily import the model. This is useful to avoid initializing CUDA
           when importing the model and thus the related error

vllm/model_executor/models/utils.py

@@ -66,7 +66,7 @@ class WeightsMapper:
 
 class AutoWeightsLoader:
     """
-    Helper class to load weights into a {class}`torch.nn.Module`. It is able
+    Helper class to load weights into a [`torch.nn.Module`][]. It is able
     to automatically detect child modules and parameters while iterating over
     the weights only once.