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, 08:30 UTC
Replace `{func}` with mkdocs style links (#18610)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
May 23, 2025 at 14:51 UTC
2edb533af26d2cdf7e4b7bdd3da0df11c009f654
5 files changed
+6
-6
vllm/model_executor/models/llava_next.py
+1
-1
@@ -540,7 +540,7 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal,
540
Unlike in LLaVA-1.5, the number of image tokens inputted to the language
541
model depends on the original size of the input image. Including the
542
original image token in the input, the required number of image tokens
543
- is given by {func}`get_llava_next_image_feature_size`.
543
+ is given by [get_llava_next_image_feature_size][].
544
545
This way, the `positions` and `attn_metadata` are consistent
546
with the `input_ids`.
vllm/multimodal/processing.py
+1
-1
@@ -387,7 +387,7 @@ _M = TypeVar("_M", bound=Union[_HasModalityAttr, _HasModalityProp])
387
388
389
def full_groupby_modality(values: Iterable[_M]) -> ItemsView[str, list[_M]]:
390
- """Convenience function to apply {func}`full_groupby` based on modality."""
390
+ """Convenience function to apply [full_groupby][] based on modality."""
391
return full_groupby(values, key=lambda x: x.modality)
392
393
vllm/platforms/interface.py
+2
-2
@@ -157,7 +157,7 @@ class Platform:
157
return self._enum == PlatformEnum.OOT
158
159
def is_cuda_alike(self) -> bool:
160
- """Stateless version of {func}`torch.cuda.is_available`."""
160
+ """Stateless version of [torch.cuda.is_available][]."""
161
return self._enum in (PlatformEnum.CUDA, PlatformEnum.ROCM)
162
163
def is_sleep_mode_available(self) -> bool:
@@ -194,7 +194,7 @@ class Platform:
194
cls,
195
device_id: int = 0,
196
) -> Optional[DeviceCapability]:
197
- """Stateless version of {func}`torch.cuda.get_device_capability`."""
197
+ """Stateless version of [torch.cuda.get_device_capability][]."""
198
return None
199
200
@classmethod
vllm/sequence.py
+1
-1
@@ -27,7 +27,7 @@ VLLM_INVALID_TOKEN_ID = -1
27
28
29
def array_full(token_id: int, count: int):
30
- """{class}`array` equivalent of {func}`numpy.full`."""
30
+ """{class}`array` equivalent of [numpy.full][]."""
31
return array(VLLM_TOKEN_ID_ARRAY_TYPE, [token_id]) * count
32
33
vllm/v1/worker/utils.py
+1
-1
@@ -66,7 +66,7 @@ def gather_mm_placeholders(
66
"""
67
Reconstructs the embeddings from the placeholder tokens.
68
69
- This is the operation of {func}`scatter_mm_placeholders`.
69
+ This is the operation of [scatter_mm_placeholders][].
70
"""
71
if is_embed is None:
72
return placeholders