mirror of
https://github.com/huggingface/transformers.git
synced 2025-07-04 05:10:06 +06:00
91be6a5eb2
* fix function imports * improve comment * Update modeling_switch_function.py * make checks more robust * improvement * rename * final test update
16 lines
704 B
Python
16 lines
704 B
Python
# Note that zamba does not have the `apply_rotary_pos_emb` function!
|
|
from transformers.models.llama.modeling_llama import apply_rotary_pos_emb
|
|
from transformers.models.zamba.modeling_zamba import ZambaAttention
|
|
|
|
|
|
# When following ZambaAttention dependencies, the function `apply_rotary_pos_emb` is not present
|
|
# by default as it is absent from the class definition (and the file altogether).
|
|
# Note that this syntax should be able to add both `apply_rotary_pos_emb` as imported directly, but
|
|
# `rotate_half` as well as a dependency from the imported function!!
|
|
class TestAttention(ZambaAttention):
|
|
def __init__(self):
|
|
pass
|
|
|
|
def forward(self):
|
|
_ = apply_rotary_pos_emb(1, 1, 1, 1)
|