more informative error message

robotic_transformer_pytorch/robotic_transformer_pytorch.py

@@ -1,8 +1,11 @@
+from __future__ import annotations
+
 import torch
+from torch.nn import Module, ModuleList
 import torch.nn.functional as F
 from torch import nn, einsum, Tensor
 
-from typing import List, Optional, Callable, Tuple
+from typing import Callable
 from beartype import beartype
 
 from einops import pack, unpack, repeat, reduce, rearrange
@@ -42,15 +45,15 @@ def posemb_sincos_1d(seq, dim, temperature = 10000, device = None, dtype = torch
 
 # helper classes
 
-class Residual(nn.Module):
+class Residual(Module):
     def __init__(self, fn):
         super().__init__()
         self.fn = fn
 
     def forward(self, x):
         return self.fn(x) + x
 
-class LayerNorm(nn.Module):
+class LayerNorm(Module):
     def __init__(self, dim):
         super().__init__()
         self.gamma = nn.Parameter(torch.ones(dim))
@@ -59,7 +62,7 @@ def __init__(self, dim):
     def forward(self, x):
         return F.layer_norm(x, x.shape[-1:], self.gamma, self.beta)
 
-class FeedForward(nn.Module):
+class FeedForward(Module):
     def __init__(self, dim, mult = 4, dropout = 0.):
         super().__init__()
         inner_dim = int(dim * mult)
@@ -83,7 +86,7 @@ def forward(self, x, cond_fn = None):
 
 # MBConv
 
-class SqueezeExcitation(nn.Module):
+class SqueezeExcitation(Module):
     def __init__(self, dim, shrinkage_rate = 0.25):
         super().__init__()
         hidden_dim = int(dim * shrinkage_rate)
@@ -101,7 +104,7 @@ def forward(self, x):
         return x * self.gate(x)
 
 
-class MBConvResidual(nn.Module):
+class MBConvResidual(Module):
     def __init__(self, fn, dropout = 0.):
         super().__init__()
         self.fn = fn
@@ -112,7 +115,7 @@ def forward(self, x):
         out = self.dropsample(out)
         return out + x
 
-class Dropsample(nn.Module):
+class Dropsample(Module):
     def __init__(self, prob = 0):
         super().__init__()
         self.prob = prob
@@ -157,7 +160,7 @@ def MBConv(
 
 # attention related classes
 
-class Attention(nn.Module):
+class Attention(Module):
     def __init__(
         self,
         dim,
@@ -259,7 +262,7 @@ def forward(self, x):
         out = self.to_out(out)
         return rearrange(out, '(b x y) ... -> b x y ...', x = height, y = width)
 
-class MaxViT(nn.Module):
+class MaxViT(Module):
     def __init__(
         self,
         *,
@@ -294,7 +297,7 @@ def __init__(
         dims = (dim_conv_stem, *dims)
         dim_pairs = tuple(zip(dims[:-1], dims[1:]))
 
-        self.layers = nn.ModuleList([])
+        self.layers = ModuleList([])
 
         # shorthand for window size for efficient block - grid like attention
 
@@ -349,8 +352,8 @@ def __init__(
     def forward(
         self,
         x,
-        texts: Optional[List[str]] = None,
-        cond_fns: Optional[Tuple[Callable, ...]] = None,
+        texts: list[str] | None = None,
+        cond_fns: tuple[Callable, ...] | None = None,
         cond_drop_prob = 0.,
         return_embeddings = False
     ):
@@ -373,7 +376,7 @@ def forward(
 
 # attention
 
-class TransformerAttention(nn.Module):
+class TransformerAttention(Module):
     def __init__(
         self,
         dim,
@@ -411,7 +414,7 @@ def forward(
         mask = None,
         attn_bias = None,
         attn_mask = None,
-        cond_fn: Optional[Callable] = None
+        cond_fn: Callable | None = None
     ):
         b = x.shape[0]
 
@@ -457,8 +460,8 @@ def forward(
         out = rearrange(out, 'b h n d -> b n (h d)')
         return self.to_out(out)
 
-@beartype
-class Transformer(nn.Module):
+class Transformer(Module):
+    @beartype
     def __init__(
         self,
         dim,
@@ -469,17 +472,18 @@ def __init__(
         ff_dropout = 0.
     ):
         super().__init__()
-        self.layers = nn.ModuleList([])
+        self.layers = ModuleList([])
         for _ in range(depth):
-            self.layers.append(nn.ModuleList([
+            self.layers.append(ModuleList([
                 TransformerAttention(dim = dim, heads =  heads, dropout = attn_dropout),
                 FeedForward(dim = dim, dropout = ff_dropout)
             ]))
 
+    @beartype
     def forward(
         self,
         x,
-        cond_fns: Optional[Tuple[Callable, ...]] = None,
+        cond_fns: tuple[Callable, ...] | None = None,
         attn_mask = None
     ):
         cond_fns = iter(default(cond_fns, []))
@@ -491,7 +495,7 @@ def forward(
 
 # token learner module
 
-class TokenLearner(nn.Module):
+class TokenLearner(Module):
     """
     https://arxiv.org/abs/2106.11297
     using the 1.1 version with the MLP (2 dense layers with gelu) for generating attention map
@@ -529,8 +533,8 @@ def forward(self, x):
 
 # Robotic Transformer
 
-@beartype
-class RT1(nn.Module):
+class RT1(Module):
+    @beartype
     def __init__(
         self,
         *,
@@ -587,18 +591,28 @@ def __init__(
             Rearrange('... (a b) -> ... a b', b = action_bins)
         )
 
-    def embed_texts(self, texts: List[str]):
+    @beartype
+    def embed_texts(self, texts: list[str]):
         return self.conditioner.embed_texts(texts)
 
     @classifier_free_guidance
+    @beartype
     def forward(
         self,
         video,
-        texts: Optional[List[str]] = None,
-        text_embeds: Optional[Tensor] = None,
+        texts: list[str] | None = None,
+        text_embeds: Tensor | None = None,
         cond_drop_prob = 0.
     ):
         assert exists(texts) ^ exists(text_embeds)
+
+        if exists(texts):
+            num_texts = len(texts)
+        elif exists(text_embeds):
+            num_texts = text_embeds.shape[0]
+
+        assert num_texts == video.shape[0], f'you only passed in {num_texts} strings for guiding the robot actions, but received batch size of {video.shape[0]} videos'
+
         cond_kwargs = dict(texts = texts, text_embeds = text_embeds)
 
         depth = self.transformer_depth

setup.py

@@ -3,7 +3,7 @@
 setup(
   name = 'robotic-transformer-pytorch',
   packages = find_packages(exclude=[]),
-  version = '0.2.1',
+  version = '0.2.2',
   license='MIT',
   description = 'Robotic Transformer - Pytorch',
   author = 'Phil Wang',
@@ -19,7 +19,7 @@
   ],
   install_requires=[
     'classifier-free-guidance-pytorch>=0.4.0',
-    'einops>=0.7',
+    'einops>=0.8',
     'torch>=2.0',
   ],
   classifiers=[