[Nightly] Modify performance comparison

[mengfei25]

1. Use issue to track reference
2. Enable performance comparison manually

[mengfei25]

Comparison https://github.com/intel/torch-xpu-ops/actions/runs/14900177123

[mengfei25]

Distributed test timeout is expected, UT failures are not related with this change

[chuanqi129]

Please work with @RUIJIEZHONG66166 to skip the temp ut failures. And for the e2e performance, I noticed that there are some performance regression as below. Does it cause by machine difference?

Category	Model	Target eager	Target inductor	Inductor vs. Eager [Target]	Baseline eager	Baseline inductor	Inductor vs. Eager [Baseline]	Target vs. Baseline [Eager]	Target vs. Baseline [Inductor]
torchbench_bfloat16_training	vision_maskrcnn	41.289908	35.392219	1.166638	38.862280	30.328529	1.281377	0.941205	0.856926
torchbench_bfloat16_training	dcgan	1.914976	2.312614	0.828057	1.761685	1.995414	0.882867	0.919951	0.862839
torchbench_bfloat16_training	mnasnet1_0	25.409208	21.975949	1.156228	27.014366	19.572720	1.380205	1.063172	0.890643
torchbench_bfloat16_training	mobilenet_v3_large	27.558122	25.776550	1.069116	27.264004	23.698865	1.150435	0.989327	0.919396
torchbench_bfloat16_training	fastNLP_Bert	57.760067	44.603815	1.294958	59.994632	41.839274	1.433931	1.038687	0.938020
huggingface_bfloat16_training	BlenderbotForCausalLM	80.269073	78.515200	1.022338	77.331836	73.737783	1.048741	0.963408	0.939153
torchbench_bfloat16_training	BERT_pytorch	44.388574	29.732542	1.492929	44.020327	27.952662	1.574817	0.991704	0.940137

[mengfei25]

Please work with @RUIJIEZHONG66166 to skip the temp ut failures. And for the e2e performance, I noticed that there are some performance regression as below. Does it cause by machine difference?

Category Model Target eager Target inductor Inductor vs. Eager [Target] Baseline eager Baseline inductor Inductor vs. Eager [Baseline] Target vs. Baseline [Eager] Target vs. Baseline [Inductor]
torchbench_bfloat16_training vision_maskrcnn 41.289908 35.392219 1.166638 38.862280 30.328529 1.281377 0.941205 0.856926
torchbench_bfloat16_training dcgan 1.914976 2.312614 0.828057 1.761685 1.995414 0.882867 0.919951 0.862839
torchbench_bfloat16_training mnasnet1_0 25.409208 21.975949 1.156228 27.014366 19.572720 1.380205 1.063172 0.890643
torchbench_bfloat16_training mobilenet_v3_large 27.558122 25.776550 1.069116 27.264004 23.698865 1.150435 0.989327 0.919396
torchbench_bfloat16_training fastNLP_Bert 57.760067 44.603815 1.294958 59.994632 41.839274 1.433931 1.038687 0.938020
huggingface_bfloat16_training BlenderbotForCausalLM 80.269073 78.515200 1.022338 77.331836 73.737783 1.048741 0.963408 0.939153
torchbench_bfloat16_training BERT_pytorch 44.388574 29.732542 1.492929 44.020327 27.952662 1.574817 0.991704 0.940137

Should be caused by machine, need time to confirm. I will collect performance this weekend on all CI machine to check it