[Bug]: [GPU] Can't find fusable transpose format: 0, 3, 1, 2,

[Jackochiu]

OpenVINO Version

2024.6.0

Operating System

Windows System

Device used for inference

GPU

Framework

PyTorch

Model used

https://huggingface.co/microsoft/Florence-2-base-ft

Issue description

We tested the performance of running the Florence-2 model on OpenVINO using the GPU on two different notebooks.
The platforms are a Core Ultra 5 125H and a Core Ultra 5 226V. We've updated the Intel iGPU driver to version 32.0.101.6078 and are using OpenVINO 2024.6.0.
We're following this guide (https://docs.openvino.ai/2024/notebooks/florence2-with-output.html) to run the Florence-2 model.
We were able to successfully run the Florence-2 model on the iGPU with the Core Ultra 5 125H, but it fails on the Core Ultra 5 226V. The error message is as follows:

[ ERROR ] Exception from src\inference\src\cpp\core.cpp:107:
Exception from src\inference\src\dev\plugin.cpp:53:
Check 'false' failed at src\plugins\intel_gpu\src\plugin\program_builder.cpp:186:
[GPU] ProgramBuilder build failed!
Check 'shape_type == shape_types::dynamic_shape || node->selected_impl != nullptr' failed at src\plugins\intel_gpu\src\graph\graph_optimizer\compile_graph.cpp:58:
[GPU] Failed to select implementation for
name:gemm:__module.vision_tower.blocks.3.0.channel_block.channel_attn.fn/aten::transpose/Transpose_3
type: gemm
original_type: GemmCheck 'false' failed at src\plugins\intel_gpu\src\graph\impls\onednn\gemm_onednn.cpp:195:
[GPU] Can't find fusable transpose format: 0, 3, 1, 2,

We've tried different versions of the Intel GPU driver with different versions of OpenVINO, but we still couldn't get Florence-2 to run properly on that notebook. Any suggestions on how to resolve this issue?

Step-by-step reproduction

No response

Relevant log output

Issue submission checklist

• I'm reporting an issue. It's not a question.
• I checked the problem with the documentation, FAQ, open issues, Stack Overflow, etc., and have not found a solution.
• There is reproducer code and related data files such as images, videos, models, etc.

[zulkifli-halim]

Hi @Jackochiu
I tested on my side and received the same result. We are checking this out and will get back to you.

[DipayanDasgupta]

.take

[github-actions]

Thanks for being interested in this issue. It looks like this ticket is already assigned to a contributor. Please communicate with the assigned contributor to confirm the status of the issue.