Remove upstream fa checks (#29471)
Signed-off-by: mingyuanm <mingyuanm@nvidia.com> Signed-off-by: Roger Wang <hey@rogerw.io> Co-authored-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
co-authored by
Roger Wang
parent
e2f56c309d
commit
460d8bbf2d
@@ -11,7 +11,6 @@ from transformers.models.qwen2_vl import Qwen2VLProcessor
|
||||
|
||||
from vllm.attention.backends.registry import AttentionBackendEnum
|
||||
from vllm.attention.layer import (
|
||||
check_upstream_fa_availability,
|
||||
maybe_get_vit_flash_attn_backend,
|
||||
)
|
||||
from vllm.config import VllmConfig
|
||||
@@ -294,12 +293,10 @@ class DotsVisionAttention(nn.Module):
|
||||
torch.get_default_dtype(),
|
||||
attn_backend_override=attn_backend_override,
|
||||
)
|
||||
self.use_upstream_fa = False
|
||||
|
||||
self.attn_backend, self.flash_attn_varlen_func = (
|
||||
maybe_get_vit_flash_attn_backend(
|
||||
self.attn_backend,
|
||||
self.use_upstream_fa,
|
||||
attn_backend_override=attn_backend_override,
|
||||
)
|
||||
)
|
||||
@@ -569,11 +566,6 @@ class DotsVisionTransformer(nn.Module):
|
||||
dtype=torch.get_default_dtype(),
|
||||
attn_backend_override=attn_backend_override,
|
||||
)
|
||||
if (
|
||||
self.attn_backend != AttentionBackendEnum.FLASH_ATTN
|
||||
and check_upstream_fa_availability(torch.get_default_dtype())
|
||||
):
|
||||
self.attn_backend = AttentionBackendEnum.FLASH_ATTN
|
||||
self.out_hidden_size = config.hidden_size
|
||||
# Keep blocks for compatibility with other vision towers
|
||||
num_layers = (
|
||||
|
||||
Reference in New Issue
Block a user