fix: move flash components into top-level

modeling_bert.py

@@ -29,17 +29,17 @@ from transformers.models.bert.modeling_bert import (
     BaseModelOutputWithPoolingAndCrossAttentions,
     BertForPreTrainingOutput,
 )
-from flash_components.bert_padding import (
+from .bert_padding import (
     index_first_axis,
     index_first_axis_residual,
     pad_input,
     unpad_input,
 )
 
-from flash_components.block import Block
-from flash_components.embedding import BertEmbeddings
-from flash_components.mha import MHA
-from flash_components.mlp import FusedMLP, Mlp
+from .block import Block
+from .embedding import BertEmbeddings
+from .mha import MHA
+from .mlp import FusedMLP, Mlp
 
 try:
     from flash_attn.ops.fused_dense import FusedDense