Spaces:
Running on Zero
Running on Zero
File size: 4,262 Bytes
2dc3625 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 | """Wan self-attention with SCoPE Normalize-Gate-Inject encoding."""
import torch
from diffsynth.models.wan_video_dit import (
SelfAttention,
modulate,
rope_apply,
)
from scope.encoding import SightlineCoordinatePE
class SelfAttentionWithSCoPE(SelfAttention):
"""
Self-Attention with Normalize-Gate-Inject Plücker PE.
All layers get scale-gated Q/K PE. Layers with enable_cam_residual=True
also add a gated frame-uniform camera residual.
"""
def __init__(
self,
dim: int,
num_heads: int,
eps: float = 1e-6,
plucker_init: str = "zero",
plucker_init_scale: float = 0.01,
plucker_mlp_hidden: int = 0,
plucker_scale: float = 0.0,
gate_init_bias: float = -2.0,
disable_spatial_rope: bool = False,
enable_cam_residual: bool = True,
scale_gate_hidden: int = 0,
log_scale_aug_prob: float = 0.0,
log_scale_aug_range: tuple = (-1.2, 1.6),
):
super().__init__(dim, num_heads, eps)
self.disable_spatial_rope = disable_spatial_rope
self.plucker_pe = SightlineCoordinatePE(
dim=dim,
plucker_init=plucker_init,
plucker_init_scale=plucker_init_scale,
plucker_mlp_hidden=plucker_mlp_hidden,
plucker_scale=plucker_scale,
gate_init_bias=gate_init_bias,
enable_cam_residual=enable_cam_residual,
scale_gate_hidden=scale_gate_hidden,
log_scale_aug_prob=log_scale_aug_prob,
log_scale_aug_range=log_scale_aug_range,
)
def _mask_spatial_rope(self, freqs: torch.Tensor) -> torch.Tensor:
head_dim = self.dim // self.num_heads
half_head = head_dim // 2
f_dim = half_head - 2 * (half_head // 3)
masked = freqs.clone()
masked[..., f_dim:] = 1.0
return masked
def forward(
self,
x: torch.Tensor,
freqs: torch.Tensor | None = None,
control_camera_dit_input: dict | None = None,
) -> torch.Tensor:
q = self.norm_q(self.q(x))
k = self.norm_k(self.k(x))
v = self.v(x)
if freqs is not None:
if self.disable_spatial_rope:
freqs = self._mask_spatial_rope(freqs)
q = rope_apply(q, freqs, self.num_heads)
k = rope_apply(k, freqs, self.num_heads)
cam_residual = None
if control_camera_dit_input is not None and "plucker_6d" in control_camera_dit_input:
q, k, cam_residual = self.plucker_pe.apply_to_qk_and_output(
q,
k,
control_camera_dit_input["plucker_6d"],
num_frames=control_camera_dit_input.get("num_frames", 1),
)
if cam_residual is not None:
v = v + cam_residual
out = self.attn(q, k, v)
return self.o(out)
def create_scope_block_forward():
"""Return a DiT block forward method that passes SCoPE coordinates."""
def forward_scope(self, x, context, t_mod, freqs, control_camera_dit_input=None):
has_seq = t_mod.dim() == 4
chunk_dim = 2 if has_seq else 1
shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = (
self.modulation.to(dtype=t_mod.dtype, device=t_mod.device) + t_mod
).chunk(6, dim=chunk_dim)
if has_seq:
shift_msa = shift_msa.squeeze(2)
scale_msa = scale_msa.squeeze(2)
gate_msa = gate_msa.squeeze(2)
shift_mlp = shift_mlp.squeeze(2)
scale_mlp = scale_mlp.squeeze(2)
gate_mlp = gate_mlp.squeeze(2)
input_x = modulate(self.norm1(x), shift_msa, scale_msa)
residual = self.self_attn(
input_x,
freqs=freqs,
control_camera_dit_input=control_camera_dit_input,
)
x = self.gate(x, gate_msa, residual)
x = x + self.cross_attn(self.norm3(x), context)
input_x = modulate(self.norm2(x), shift_mlp, scale_mlp)
x = self.gate(x, gate_mlp, self.ffn(input_x))
return x
return forward_scope
|