Commit 1b15751a authored by Phil Wang's avatar Phil Wang
Browse files

fix length extrapolation for rotary embeddings in local attention package

parent a94e32f4
Loading
Loading
Loading
Loading
+1 −1
Original line number Diff line number Diff line
@@ -342,7 +342,7 @@ class LocalTransformerBlock(nn.Module):
        **kwargs
    ):
        super().__init__()
        self.attn = LocalMHA(dim = dim, qk_rmsnorm = True, **kwargs)
        self.attn = LocalMHA(dim = dim, qk_rmsnorm = True, use_xpos = True, **kwargs)
        self.ff = FeedForward(dim = dim)

    def forward(self, x):
+1 −1
Original line number Diff line number Diff line
__version__ = '0.18.0'
__version__ = '0.18.1'
+1 −1
Original line number Diff line number Diff line
@@ -26,7 +26,7 @@ setup(
    'fairseq',
    'joblib',
    'lion-pytorch',
    'local-attention>=1.6.0',
    'local-attention>=1.7.0',
    'scikit-learn',
    'sentencepiece',
    'torch>=1.12',