Add Final annotation to attn_fas to avoid symbol lookup of new scaled_dot_product_attn fn on old PyTorch in jit

Author: rwightman

timm/models/maxxvit.py

@@ -42,6 +42,7 @@
 
 import torch
 from torch import nn
+from torch.jit import Final
 
 from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 from timm.layers import Mlp, ConvMlp, DropPath, LayerNorm, ClassifierHead, NormMlpClassifierHead
@@ -140,6 +141,8 @@ class MaxxVitCfg:
 
 
 class Attention2d(nn.Module):
+    fast_attn: Final[bool]
+
     """ multi-head attention for 2D NCHW tensors"""
     def __init__(
             self,
@@ -208,6 +211,8 @@ def forward(self, x, shared_rel_pos: Optional[torch.Tensor] = None):
 
 class AttentionCl(nn.Module):
     """ Channels-last multi-head attention (B, ..., C) """
+    fast_attn: Final[bool]
+
     def __init__(
             self,
             dim: int,

timm/models/vision_transformer.py

@@ -33,6 +33,7 @@
 import torch.nn as nn
 import torch.nn.functional as F
 import torch.utils.checkpoint
+from torch.jit import Final
 
 from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD, \
     OPENAI_CLIP_MEAN, OPENAI_CLIP_STD
@@ -51,6 +52,8 @@
 
 
 class Attention(nn.Module):
+    fast_attn: Final[bool]
+
     def __init__(
             self,
             dim,

timm/models/vision_transformer_relpos.py

@@ -11,6 +11,7 @@
 
 import torch
 import torch.nn as nn
+from torch.jit import Final
 from torch.utils.checkpoint import checkpoint
 
 from timm.data import IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
@@ -25,6 +26,8 @@
 
 
 class RelPosAttention(nn.Module):
+    fast_attn: Final[bool]
+
     def __init__(
             self,
             dim,