[TEST] fix error in argmin UT when dtype=int16

[awayzjj]

PR Category

Type of Change

Description

Issue

1. The tl.min function will upcast an int16 input tensor to int32 on iluvatar. Therefore, the min_values should be initialized with dtype=int32; otherwise, the Triton IR cannot be generated.
2. add tune config of batchnorm

Progress

• Change is properly reviewed (1 reviewer required, 2 recommended).
• Change is responded to an issue.
• Change is fully covered by a UT.

Performance

[iclementine]

LGTM.

It is a temporary workaround to avoid promoting min_values to float32 when value is of type tl.int16.