Commit 06602256 authored by Phil Wang's avatar Phil Wang
Browse files

allow for configuring xpos scale base, as @ilya16 reports that at about 3... 

allow for configuring xpos scale base, as @ilya16 reports that at about 3 minute mark, it breaks down
parent 3e4f582b

audiolm_pytorch/soundstream.py

+3 −1
Original line number Diff line number Diff line
@@ -360,7 +360,8 @@ class SoundStream(nn.Module): 
        attn_window_size = 128,

        attn_dim_head = 64,

        attn_heads = 8,

        attn_depth = 1

        attn_depth = 1,

        attn_xpos_scale_base = None

    ):

        super().__init__()
@@ -398,6 +399,7 @@ class SoundStream(nn.Module): 
            dim_head = attn_dim_head,

            heads = attn_heads,

            window_size = attn_window_size,

            xpos_scale_base = attn_xpos_scale_base,

            prenorm = True,

            causal = True

        )

audiolm_pytorch/version.py

+1 −1
Original line number Diff line number Diff line 
__version__ = '0.19.1'
 
__version__ = '0.19.2'

setup.py

+1 −1
Original line number Diff line number Diff line
@@ -26,7 +26,7 @@ setup( 
    'fairseq',

    'joblib',

    'lion-pytorch',

    'local-attention>=1.7.0',

    'local-attention>=1.7.1',

    'scikit-learn',

    'sentencepiece',

    'torch>=1.12',