vllm_flash_attn: Setup for vllm_kernels package (#71)

This import was failing when I was attempting to run tests so let's just
use relative imports instead of full package imports

Signed-off-by: Eli Uriegas <eliuriegas@meta.com>

Author: seemethere

vllm_flash_attn/layers/rotary.py

@@ -7,7 +7,7 @@
 
 import torch
 from einops import rearrange, repeat
-from vllm.vllm_flash_attn.ops.triton.rotary import apply_rotary   # modified from original
+from ..ops.triton.rotary import apply_rotary   # modified from original
 
 
 def rotate_half(x, interleaved=False):