add RoPE attention module as this is shown to help training dynamics and generation quality for DiTs

src/lerobot/policies/multi_task_dit/configuration_multi_task_dit.py

@@ -167,9 +167,13 @@ class TransformerConfig:
     num_layers: int = 6  # Number of transformer layers
     num_heads: int = 8  # Number of attention heads
     dropout: float = 0.1  # Dropout rate
-    use_positional_encoding: bool = True  # Whether to use positional encoding
+    use_positional_encoding: bool = False  # Whether to use absolute positional encoding
     diffusion_step_embed_dim: int = 256  # Timestep embedding size
 
+    # RoPE (Rotary Position Embedding) configuration
+    use_rope: bool = True  # Whether to use Rotary Position Embedding in attention (baseline is True)
+    rope_base: float = 10000.0  # Base frequency for RoPE computation
+
     def __post_init__(self):
         """Validate Transformer-specific parameters."""
         if self.hidden_dim <= 0:

src/lerobot/policies/multi_task_dit/modules/transformer.py

@@ -71,6 +71,146 @@ class SinusoidalPosEmb(nn.Module):
         return emb
 
 
+class RotaryPositionalEmbedding(nn.Module):
+    """Rotary Position Embedding (RoPE) for transformers.
+
+    RoPE encodes position information by rotating query and key vectors,
+    which naturally captures relative positions through the dot product.
+    Applied at every attention layer rather than once at input.
+
+    To do this, we need to reimplement the attention mechanism to apply RoPE
+    to Q and K before computing the attention scores, so we cannot use the
+    the built-in MultiheadAttention module.
+
+    Original RoPE Paper: https://arxiv.org/abs/2104.09864 (RoFormer)
+    """
+
+    def __init__(self, head_dim: int, max_seq_len: int = 512, base: float = 10000.0):
+        super().__init__()
+        assert head_dim % 2 == 0, "head_dim must be even for RoPE"
+
+        self.head_dim = head_dim
+        self.max_seq_len = max_seq_len
+        self.base = base
+
+        # Precompute inverse frequencies: theta_i = 1 / (base^(2i/d))
+        inv_freq = 1.0 / (base ** (torch.arange(0, head_dim, 2).float() / head_dim))
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+        self._precompute_cache(max_seq_len)
+
+    def _precompute_cache(self, seq_len: int):
+        t = torch.arange(seq_len, dtype=self.inv_freq.dtype)
+
+        freqs = torch.outer(t, self.inv_freq)
+
+        emb = torch.cat((freqs, freqs), dim=-1)
+
+        self.register_buffer("_cos_cached", emb.cos()[None, None, :, :], persistent=False)
+        self.register_buffer("_sin_cached", emb.sin()[None, None, :, :], persistent=False)
+
+    def _rotate_half(self, x: Tensor) -> Tensor:
+        """Rotate half the hidden dims of the input.
+
+        For x = [x1, x2], returns [-x2, x1]
+        """
+        x1 = x[..., : x.shape[-1] // 2]
+        x2 = x[..., x.shape[-1] // 2 :]
+        return torch.cat((-x2, x1), dim=-1)
+
+    def forward(self, q: Tensor, k: Tensor) -> tuple[Tensor, Tensor]:
+        """Apply rotary embeddings to query and key tensors."""
+        seq_len = q.shape[2]
+
+        if seq_len > self.max_seq_len:
+            raise ValueError(
+                f"Sequence length {seq_len} exceeds max_seq_len {self.max_seq_len}. "
+                f"Increase max_seq_len in RoPE config."
+            )
+
+        # Slice precomputed cache to actual sequence length
+        cos = self._cos_cached[:, :, :seq_len, :].to(q.dtype)
+        sin = self._sin_cached[:, :, :seq_len, :].to(q.dtype)
+
+        # Apply rotation: q_rot = q * cos + rotate_half(q) * sin
+        q_rotated = (q * cos) + (self._rotate_half(q) * sin)
+        k_rotated = (k * cos) + (self._rotate_half(k) * sin)
+
+        return q_rotated, k_rotated
+
+
+class RoPEAttention(nn.Module):
+    """Multi-head self-attention with Rotary Position Embedding (RoPE).
+
+    Custom attention implementation that applies RoPE to Q and K before
+    computing attention scores. This allows position information to be
+    encoded at every attention layer.
+    """
+
+    def __init__(
+        self,
+        hidden_size: int,
+        num_heads: int,
+        dropout: float = 0.0,
+        max_seq_len: int = 512,
+        rope_base: float = 10000.0,
+    ):
+        """
+        Args:
+            hidden_size: Total hidden dimension
+            num_heads: Number of attention heads
+            dropout: Attention dropout rate
+            max_seq_len: Maximum sequence length for RoPE cache
+            rope_base: Base for RoPE frequency computation
+        """
+        super().__init__()
+        assert hidden_size % num_heads == 0, "hidden_size must be divisible by num_heads"
+
+        self.hidden_size = hidden_size
+        self.num_heads = num_heads
+        self.head_dim = hidden_size // num_heads
+        self.scale = self.head_dim**-0.5
+
+        self.qkv_proj = nn.Linear(hidden_size, 3 * hidden_size, bias=True)
+        self.out_proj = nn.Linear(hidden_size, hidden_size, bias=True)
+        self.dropout = nn.Dropout(dropout) if dropout > 0 else nn.Identity()
+        self.rope = RotaryPositionalEmbedding(head_dim=self.head_dim, max_seq_len=max_seq_len, base=rope_base)
+
+    def forward(self, x: Tensor) -> Tensor:
+        """
+        Args:
+            x: (B, T, hidden_size) input sequence
+
+        Returns:
+            (B, T, hidden_size) attention output
+        """
+        B, T, _ = x.shape  # noqa: N806
+
+        # Compute Q, K, V
+        qkv = self.qkv_proj(x)  # (B, T, 3 * hidden_size)
+        qkv = qkv.reshape(B, T, 3, self.num_heads, self.head_dim)
+        qkv = qkv.permute(2, 0, 3, 1, 4)  # (3, B, num_heads, T, head_dim)
+        q, k, v = qkv[0], qkv[1], qkv[2]  # Each: (B, num_heads, T, head_dim)
+
+        # Apply RoPE to Q and K
+        q, k = self.rope(q, k)
+
+        # Scaled dot-product attention
+        # Using PyTorch's efficient attention when available
+        attn_out = torch.nn.functional.scaled_dot_product_attention(
+            q,
+            k,
+            v,
+            dropout_p=self.dropout.p if isinstance(self.dropout, nn.Dropout) and self.training else 0.0,
+        )  # (B, num_heads, T, head_dim)
+
+        # Reshape and project output
+        attn_out = attn_out.transpose(1, 2).reshape(B, T, self.hidden_size)  # (B, T, hidden_size)
+        output = self.out_proj(attn_out)
+
+        return output
+
+
 class TransformerBlock(nn.Module):
     """DiT-style transformer block with AdaLN-Zero.
 
@@ -78,11 +218,20 @@ class TransformerBlock(nn.Module):
     - shift_msa, scale_msa, gate_msa: for attention block
     - shift_mlp, scale_mlp, gate_mlp: for MLP block
 
+    Supports both standard attention and RoPE attention.
+
     Reference: https://github.com/facebookresearch/DiT
     """
 
     def __init__(
-        self, hidden_size: int = 128, num_heads: int = 4, num_features: int = 128, dropout: float = 0.0
+        self,
+        hidden_size: int = 128,
+        num_heads: int = 4,
+        num_features: int = 128,
+        dropout: float = 0.0,
+        use_rope: bool = False,
+        max_seq_len: int = 512,
+        rope_base: float = 10000.0,
     ):
         """
         Args:
@@ -90,12 +239,26 @@ class TransformerBlock(nn.Module):
             num_heads: Number of attention heads
             num_features: Size of conditioning features
             dropout: Dropout rate
+            use_rope: Whether to use Rotary Position Embedding
+            max_seq_len: Maximum sequence length (for RoPE cache)
+            rope_base: Base frequency for RoPE
         """
         super().__init__()
 
-        self.multihead_attn = nn.MultiheadAttention(
-            hidden_size, num_heads=num_heads, batch_first=True, dropout=dropout
-        )
+        self.use_rope = use_rope
+
+        if use_rope:
+            self.attn = RoPEAttention(
+                hidden_size=hidden_size,
+                num_heads=num_heads,
+                dropout=dropout,
+                max_seq_len=max_seq_len,
+                rope_base=rope_base,
+            )
+        else:
+            self.multihead_attn = nn.MultiheadAttention(
+                hidden_size, num_heads=num_heads, batch_first=True, dropout=dropout
+            )
 
         # Layer normalizations (no learnable affine parameters, all adaptation via conditioning)
         self.norm1 = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
@@ -128,7 +291,12 @@ class TransformerBlock(nn.Module):
         # Attention block: norm → modulate → attn → gate × output → residual
         # modulate requires unsqueeze(1) to add sequence dimension for broadcasting
         attn_input = modulate(self.norm1(x), shift_msa.unsqueeze(1), scale_msa.unsqueeze(1))
-        attn_out, _ = self.multihead_attn(attn_input, attn_input, attn_input)
+
+        if self.use_rope:
+            attn_out = self.attn(attn_input)
+        else:
+            attn_out, _ = self.multihead_attn(attn_input, attn_input, attn_input)
+
         x = x + gate_msa.unsqueeze(1) * attn_out
 
         # MLP block: norm → modulate → mlp → gate × output → residual
@@ -163,6 +331,7 @@ class DiffusionTransformer(nn.Module):
         self.num_layers = self.transformer_config.num_layers
         self.num_heads = self.transformer_config.num_heads
         self.dropout = self.transformer_config.dropout
+        self.use_rope = self.transformer_config.use_rope
 
         self.timestep_embed_dim = self.transformer_config.diffusion_step_embed_dim
         self.time_mlp = nn.Sequential(
@@ -179,7 +348,7 @@ class DiffusionTransformer(nn.Module):
         self.input_proj = nn.Linear(self.action_dim, self.hidden_size)
 
         if self.transformer_config.use_positional_encoding:
-            # Learnable positional embeddings for sequence positions
+            # Learnable positional embeddings for sequence positions (absolute encoding)
             self.pos_embedding = nn.Parameter(
                 torch.empty(1, self.horizon, self.hidden_size).normal_(std=0.02)
             )
@@ -193,6 +362,9 @@ class DiffusionTransformer(nn.Module):
                     num_heads=self.num_heads,
                     num_features=self.cond_dim,
                     dropout=self.dropout,
+                    use_rope=self.use_rope,
+                    max_seq_len=self.horizon,  # This remains fixed because we aren't generating variable length sequences
+                    rope_base=getattr(self.transformer_config, "rope_base", 10000.0),
                 )
                 for _ in range(self.num_layers)
             ]