Skip to content

[Model] Define merge_by_field_config MM interface (R-T) #26260

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
+10 −9
Open

[Model] Define merge_by_field_config MM interface (R-T) #26260

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6e3d1d8
[Model] Use merge_by_field_config for MM models (R-T)
ayushsatyam146 Oct 6, 2025
af11f57
Revert
DarkLight1337 Oct 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 8 additions & 4 deletions vllm/model_executor/models/step3_vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,6 @@
from .utils import (
AutoWeightsLoader,
WeightsMapper,
flatten_bn,
init_vllm_registered_model,
maybe_prefix,
)
Expand Down Expand Up @@ -895,6 +894,8 @@ def forward(
dummy_inputs=Step3VLDummyInputsBuilder,
)
class Step3VLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP):
merge_by_field_config = True

hf_to_vllm_mapper = WeightsMapper(
orig_to_new_prefix={
"model.": "language_model.model.",
Expand Down Expand Up @@ -982,18 +983,21 @@ def _parse_and_validate_image_input(
return None

if pixel_values is not None:
pixel_values = flatten_bn(pixel_values, concat=True)
if pixel_values.dim() >= 3:
pixel_values = pixel_values.view(-1, *pixel_values.shape[-3:])
if patch_pixel_values is not None:
patch_pixel_values = flatten_bn(patch_pixel_values, concat=True)
patch_pixel_values = patch_pixel_values.view(
-1, *patch_pixel_values.shape[-3:]
)
# Handle empty patch_pixel_values by setting to None
if patch_pixel_values.shape[0] == 0:
patch_pixel_values = None
num_patches = flatten_bn(num_patches, concat=True).tolist()
if isinstance(num_patches, torch.Tensor):
num_patches = num_patches.tolist()
elif isinstance(num_patches, list):
num_patches = [
n.item() if isinstance(n, torch.Tensor) else n for n in num_patches
]

return Step3VLImagePixelInputs(
type="pixel_values",
Expand Down
6 changes: 1 addition & 5 deletions vllm/model_executor/models/tarsier.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,11 +47,7 @@
from .clip import CLIPVisionModel
from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
from .siglip import SiglipVisionModel
from .utils import (
AutoWeightsLoader,
init_vllm_registered_model,
maybe_prefix,
)
from .utils import AutoWeightsLoader, init_vllm_registered_model, maybe_prefix
from .vision import (
VisionEncoderInfo,
get_num_selected_vision_tokens,
Expand Down
1 change: 1 addition & 0 deletions vllm/model_executor/models/terratorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -226,6 +226,7 @@ def apply(
dummy_inputs=TerratorchInputBuilder,
)
class Terratorch(nn.Module, IsAttentionFree, SupportsMultiModal):
merge_by_field_config = True
supports_multimodal_raw_input_only = True
is_pooling_model = True

Expand Down