x-transformers 1.43.0__py3-none-any.whl → 1.43.1__py3-none-any.whl

x_transformers/x_transformers.py

@@ -870,6 +870,7 @@ class HyperConnection(Module):
         *,
         layer_index,
         num_residual_streams,
+        tanh = True,
         **kwargs
     ):
         """
@@ -878,6 +879,8 @@ class HyperConnection(Module):
         """
         super().__init__()
 
+        self.act = nn.Tanh() if tanh else nn.Identity()
+
         self.norm = nn.LayerNorm(dim, bias = False)
 
         self.num_residual_streams = num_residual_streams
@@ -901,11 +904,11 @@ class HyperConnection(Module):
 
         normed = self.norm(residuals)
 
-        wc_weight = (normed @ self.dynamic_alpha_fn).tanh()
+        wc_weight = self.act(normed @ self.dynamic_alpha_fn)
         dynamic_alpha = wc_weight * self.dynamic_alpha_scale
         alpha = dynamic_alpha + self.static_alpha
 
-        dc_weight = (normed @ self.dynamic_beta_fn).tanh()
+        dc_weight = self.act(normed @ self.dynamic_beta_fn)
         dynamic_beta = dc_weight * self.dynamic_beta_scale
         beta = dynamic_beta + self.static_beta
 
@@ -1650,9 +1653,10 @@ class AttentionLayers(Module):
         unet_skips = False,
         num_residual_streams = 1,
         reinject_input = False,              # seen first in DEQ paper https://arxiv.org/abs/1909.01377, but later used in a number of papers trying to achieve depthwise generalization https://arxiv.org/abs/2410.03020v1
-        add_value_residual = False,          # resformer from Zhou et al - https://arxiv.org/abs/2410.17897v1
+        add_value_residual = False,          # resformer from Zhou et al - https://arxiv.org/abs/2410.17897v1 - further corroboration by https://arxiv.org/abs/2412.15113 (faster emergence of ICL) - looks like this setting may becoming a necessity for every transformer soon
         learned_value_residual_mix = True,   # seeing big improvements when the value residual mix value is learned per token - credit goes to @faresobeid for taking the first step with learned scalar mix, then @Blinkdl for taking it a step further with data dependent. here we will use per token learned
         rel_pos_kwargs: dict = dict(),
+        residual_fn_kwargs: dict = dict(),
         **kwargs
     ):
         super().__init__()
@@ -1957,7 +1961,7 @@ class AttentionLayers(Module):
             else:
                 residual_fn = Residual
 
-            residual = residual_fn(dim, layer_index = ind, scale_residual = scale_residual, scale_residual_constant = scale_residual_constant)
+            residual = residual_fn(dim, layer_index = ind, scale_residual = scale_residual, scale_residual_constant = scale_residual_constant, **residual_fn_kwargs)
 
             # handle unet skip connection
 

{x_transformers-1.43.0.dist-info → x_transformers-1.43.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: x-transformers
-Version: 1.43.0
+Version: 1.43.1
 Summary: X-Transformers - Pytorch
 Home-page: https://github.com/lucidrains/x-transformers
 Author: Phil Wang

{x_transformers-1.43.0.dist-info → x_transformers-1.43.1.dist-info}/RECORD

@@ -6,11 +6,11 @@ x_transformers/dpo.py,sha256=xt4OuOWhU8pN3OKN2LZAaC2NC8iiEnchqqcrPWVqf0o,3521
 x_transformers/multi_input.py,sha256=tCh-fTJDj2ib4SMGtsa-AM8MxKzJAQSwqAXOu3HU2mg,9252
 x_transformers/neo_mlp.py,sha256=XCNnnop9WLarcxap1kGuYc1x8GHvwkZiDRnXOxSl3Po,3452
 x_transformers/nonautoregressive_wrapper.py,sha256=2NU58hYMgn-4Jzg3mie-mXb0XH_dCN7fjlzd3K1rLUY,10510
-x_transformers/x_transformers.py,sha256=wAY0lqZvFlXk-fmpr4Ot6yZ6ivzEjetFXTin7z7eA88,100075
+x_transformers/x_transformers.py,sha256=JG38kcXdhRBKT5_FHMhV5dQabSGrAHsuIQkHjPalDiI,100384
 x_transformers/xl_autoregressive_wrapper.py,sha256=CvZMJ6A6PA-Y_bQAhnORwjJBSl6Vjq2IdW5KTdk8NI8,4195
 x_transformers/xval.py,sha256=7S00kCuab4tWQa-vf-z-XfzADjVj48MoFIr7VSIvttg,8575
-x_transformers-1.43.0.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
-x_transformers-1.43.0.dist-info/METADATA,sha256=C6eRstMfzmbxQUxNeKnt1Mf-e9pJ45GKNJ8hsc_3uwo,738
-x_transformers-1.43.0.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
-x_transformers-1.43.0.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
-x_transformers-1.43.0.dist-info/RECORD,,
+x_transformers-1.43.1.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
+x_transformers-1.43.1.dist-info/METADATA,sha256=V57c6Bps0GjG0GLEBpxkHdbvxIWzXss2Xu5_KQJJXPc,738
+x_transformers-1.43.1.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
+x_transformers-1.43.1.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
+x_transformers-1.43.1.dist-info/RECORD,,