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 14, 05:30 UTC
[Bugfix] Fix `PixtralHFImagePixelInputs` dynamic shape check (#22827)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Isotr0py committed
Aug 14, 2025 at 17:35 UTC
7c3a0741c67007dd759f52d07f7aca854628b81b
2 files changed
+4
-3
tests/models/multimodal/test_tensor_schema.py
+1
-1
@@ -153,4 +153,4 @@ def test_model_tensor_schema(model_arch: str, vllm_runner: type[VllmRunner],
153
if hasattr(model, method_name):
154
getattr(model, method_name)(**mm_kwargs)
155
156
- vllm_model.apply_model(validate_model_input)
156
+ vllm_model.apply_model(validate_model_input)
\ No newline at end of file
vllm/model_executor/models/llava.py
+3
-2
@@ -72,8 +72,9 @@ class PixtralHFImagePixelInputs(TensorSchema):
72
in which case the data is passed as a list instead of a batched tensor.
73
"""
74
type: Literal["pixel_values_pixtral"] = "pixel_values_pixtral"
75
- pixel_values: Annotated[Union[torch.Tensor, list[torch.Tensor]],
76
- TensorShape("bn", "c", "h", "w")]
75
+ pixel_values: Annotated[
76
+ Union[torch.Tensor, list[torch.Tensor]],
77
+ TensorShape("bn", "c", "h", "w", dynamic_dims={"h", "w"})]
78
79
80
class LlavaImageEmbeddingInputs(TensorSchema):