xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 03:54:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

Migrate InternVLImagePixelInputs (in nemotron_vl.py) to TensorSchema (#22023)

Browse Source 
Signed-off-by: Benji Beck <benjibeck@meta.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
Benji Beck 2025-08-18 22:48:26 -07:00 committed by GitHub
parent 78dba404ad
commit e75f342261
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
vllm/model_executor/models/nemotron_vl.py
View File

@ -458,27 +458,6 @@ class LlamaNemotronVLChatModel(nn.Module, SupportsMultiModal, SupportsPP,
vit_embeds = self.mlp1(vit_embeds)
return vit_embeds
def _validate_pixel_values(self, data: torch.Tensor) -> torch.Tensor:
#use force_image_size to get image_size
h = w = self.config.force_image_size
expected_dims = (3, h, w)
def _validate_shape(d: torch.Tensor):
actual_dims = tuple(d.shape)
if actual_dims != expected_dims:
expected_expr = str(expected_dims)
raise ValueError(
"The expected shape of pixel values per image per batch "
f" per patch is {expected_expr}. "
f"You supplied {tuple(d.shape)}.")
for d in data:
_validate_shape(d)
return data
def _parse_and_validate_image_input(
self, **kwargs: object) -> Optional[InternVLImageInputs]:
pixel_values_flat = kwargs.pop("pixel_values_flat", None)
@ -516,9 +495,12 @@ class LlamaNemotronVLChatModel(nn.Module, SupportsMultiModal, SupportsPP,
return InternVLImagePixelInputs(
type="pixel_values",
pixel_values_flat=self._validate_pixel_values(
pixel_values_flat),
pixel_values_flat=pixel_values_flat,
num_patches=image_num_patches,
resolve_bindings={
"h": self.config.force_image_size,
"w": self.config.force_image_size
},
)
raise AssertionError("This line should be unreachable.")