[E2E_baseline] Torchbench detectron2 series models accuracy test with RuntimeError: dets should have the same type as scores

[chuanqi129]

🐛 Describe the bug

Those detectron2 series models accuracy crash with RuntimeError: dets should have the same type as scores

Model	Precision	Mode
detectron2_fasterrcnn_r_101_c4	bf16/fp16	inference
detectron2_fasterrcnn_r_101_dc5	bf16/fp16	inference
detectron2_fasterrcnn_r_101_fpn	bf16/fp16	inference
detectron2_fasterrcnn_r_50_c4	bf16/fp16	inference
detectron2_fasterrcnn_r_50_dc5	bf16/fp16	inference
detectron2_fasterrcnn_r_50_fpn	bf16/fp16	inference
detectron2_maskrcnn	bf16/fp16	inference
detectron2_maskrcnn_r_101_c4	bf16/fp16	inference
detectron2_maskrcnn_r_101_fpn	bf16/fp16	inference
detectron2_maskrcnn_r_50_c4	bf16/fp16	inference
detectron2_maskrcnn_r_50_fpn	bf16/fp16	inference
detectron2_maskrcnn	bf16/fp16	training
detectron2_maskrcnn_r_50_c4	bf16/fp16	training

Versions

Pytorch: git clone -b e2e-baseline https://github.com/etaf/pytorch-inductor-xpu pytorch
Test script: inductor_xpu_test.sh

[chuanqi129]

Close it as we have refreshed baseline

[anmyachev]

I am still seeing this problem message in WARNING:common:current_device=xpu; error:dets should have the same type as scores.

I have a fix for it in facebookresearch/detectron2#5479 but I doubt that it will be merged, since the repository is not very active.