Different batch sizes lead to different evalution results for LongVIT

[HHHedo]

Hi, Thanks for your great work. When I use the LongVIT for survival prediction, I find that different batch sizes will lead to different C-index results. After I print the risk for each person, I also find that some risks are different. The followings are some examples

bs = 8	bs =1	bs =2
TCGA-06-0168 -1.2465172	TCGA-06-0168 -1.2457247	TCGA-06-0168 -1.2465172
TCGA-06-0645 -3.1927319	TCGA-06-0645 -3.193125	TCGA-06-0645 -3.1927319
TCGA-06-0646 -2.8353539	TCGA-06-0646 -2.8353539	TCGA-06-0646 -2.8353539
TCGA-06-0750 -0.9242315	TCGA-06-0750 -0.924083	TCGA-06-0750 -0.9242315
TCGA-06-5410 -1.7878933	TCGA-06-5410 -1.7888575	TCGA-06-5410 -1.7877667
TCGA-06-5411 -0.7950437	TCGA-06-5411 -0.7950437	TCGA-06-5411 -0.7950437
TCGA-06-5413 -0.8715491	TCGA-06-5413 -0.8715491	TCGA-06-5413 -0.87227124
TCGA-06-5856 -1.512358	TCGA-06-5856 -1.512358	TCGA-06-5856 -1.512358
TCGA-12-0618 -3.3228474	TCGA-12-0618 -3.3228474	TCGA-12-0618 -3.3228474
TCGA-12-0821 -3.0764904	TCGA-12-0821 -3.0764904	TCGA-12-0821 -3.0764904
TCGA-12-3653 -2.5728168	TCGA-12-3653 -2.5728168	TCGA-12-3653 -2.5728168
TCGA-19-2629 -2.5467625	TCGA-19-2629 -2.5467625	TCGA-19-2629 -2.5467625
TCGA-19-5960 -2.1180227	TCGA-19-5960 -2.1180227	TCGA-19-5960 -2.1180227
TCGA-26-1442 -1.2899182	TCGA-26-1442 -1.2899779	TCGA-26-1442 -1.2899182
TCGA-26-5133 -2.0898883	TCGA-26-5133 -2.0900764	TCGA-26-5133 -2.0894563
TCGA-27-1830 -1.2196785	TCGA-27-1830 -1.2196785	TCGA-27-1830 -1.2196785
TCGA-27-1831 -2.6002393	TCGA-27-1831 -2.599653	TCGA-27-1831 -2.599653

Looking forward to your help!