Separate attention backends (#3005)
This commit is contained in:
@@ -10,7 +10,7 @@ from torch.nn import LayerNorm
|
||||
|
||||
from vllm.model_executor.input_metadata import InputMetadata
|
||||
from vllm.model_executor.layers.activation import SiluAndMul
|
||||
from vllm.model_executor.layers.attention import PagedAttention
|
||||
from vllm.model_executor.layers.attention import Attention
|
||||
from vllm.model_executor.layers.layernorm import RMSNorm
|
||||
from vllm.model_executor.layers.linear import (LinearMethodBase,
|
||||
MergedColumnParallelLinear,
|
||||
@@ -87,7 +87,7 @@ class GLMAttention(nn.Module):
|
||||
base=10000 * rope_ratio,
|
||||
is_neox_style=False,
|
||||
)
|
||||
self.attn = PagedAttention(
|
||||
self.attn = Attention(
|
||||
self.num_heads,
|
||||
self.head_dim,
|
||||
self.scaling,
|
||||
|
||||
Reference in New Issue
Block a user