mirror of
https://github.com/ml-explore/mlx-examples.git
synced 2025-08-31 11:54:37 +08:00
chore: remove hardcoded rope_scaling_factor
This commit is contained in:
parent
e17e07002a
commit
6a62a8bca4
@ -105,7 +105,9 @@ class Attention(nn.Module):
|
||||
args.num_attention_heads * self.head_dim, args.hidden_size, bias=False
|
||||
)
|
||||
self.rope = LinearScalingRoPE(
|
||||
self.head_dim, rope_scaling_factor=4.0, base=args.rope_theta
|
||||
self.head_dim,
|
||||
rope_scaling_factor=args.rope_scaling_factor,
|
||||
base=args.rope_theta,
|
||||
)
|
||||
|
||||
def __call__(
|
||||
|
Loading…
Reference in New Issue
Block a user