445 lines
17 KiB
Python
445 lines
17 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# Adapted from Matrix-Game-3.0: https://github.com/SkyworkAI/Matrix-Game/blob/main/Matrix-Game-3/wan/modules/action_module.py
|
|
|
|
from einops import rearrange
|
|
import torch
|
|
import torch.nn as nn
|
|
|
|
from fastvideo.attention import LocalAttention
|
|
from fastvideo.layers.linear import ReplicatedLinear
|
|
from fastvideo.layers.layernorm import FP32LayerNorm
|
|
from fastvideo.layers.rotary_embedding import (
|
|
get_nd_rotary_pos_embed as _fv_get_nd_rotary_pos_embed,
|
|
_apply_rotary_emb,
|
|
)
|
|
from fastvideo.platforms import AttentionBackendEnum
|
|
|
|
|
|
class WanRMSNorm(nn.Module):
|
|
|
|
def __init__(self, dim: int, eps: float = 1e-5):
|
|
super().__init__()
|
|
self.dim = dim
|
|
self.eps = eps
|
|
self.weight = nn.Parameter(torch.ones(dim))
|
|
|
|
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
|
return x * torch.rsqrt(x.pow(2).mean(dim=-1, keepdim=True) + self.eps)
|
|
|
|
|
|
def _get_nd_rotary_pos_embed_matrixgame(
|
|
rope_dim_list,
|
|
rope_sizes,
|
|
theta: float = 10000.0,
|
|
theta_rescale_factor: float = 1.0,
|
|
):
|
|
cos, sin = _fv_get_nd_rotary_pos_embed(
|
|
rope_dim_list,
|
|
rope_sizes,
|
|
theta=theta,
|
|
theta_rescale_factor=theta_rescale_factor,
|
|
dtype=torch.float32,
|
|
)
|
|
# convert from [S, D/2] to [S, D] format
|
|
cos = cos.repeat_interleave(2, dim=1)
|
|
sin = sin.repeat_interleave(2, dim=1)
|
|
return cos, sin
|
|
|
|
|
|
def _apply_rotary_emb_qk(
|
|
xq: torch.Tensor,
|
|
xk: torch.Tensor,
|
|
freqs_cos: torch.Tensor,
|
|
freqs_sin: torch.Tensor,
|
|
) -> tuple[torch.Tensor, torch.Tensor]:
|
|
seq_len = xq.shape[1]
|
|
cos = freqs_cos[:seq_len]
|
|
sin = freqs_sin[:seq_len]
|
|
cos_half = cos[:, ::2]
|
|
sin_half = sin[:, ::2]
|
|
xq_out = _apply_rotary_emb(xq, cos_half, sin_half, is_neox_style=False)
|
|
xk_out = _apply_rotary_emb(xk, cos_half, sin_half, is_neox_style=False)
|
|
return xq_out, xk_out
|
|
|
|
|
|
class MatrixGame3ActionModule(nn.Module):
|
|
"""
|
|
action module from https://arxiv.org/pdf/2501.08325
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
mouse_dim_in: int = 2,
|
|
keyboard_dim_in: int = 6,
|
|
hidden_size: int = 128,
|
|
img_hidden_size: int = 1536,
|
|
keyboard_hidden_dim: int = 1024,
|
|
mouse_hidden_dim: int = 1024,
|
|
vae_time_compression_ratio: int = 4,
|
|
windows_size: int = 3,
|
|
heads_num: int = 16,
|
|
patch_size: list | None = None,
|
|
qk_norm: bool = True,
|
|
qkv_bias: bool = False,
|
|
rope_dim_list: list | None = None,
|
|
rope_theta=256,
|
|
mouse_qk_dim_list: list | None = None,
|
|
enable_mouse=True,
|
|
enable_keyboard=True,
|
|
blocks: list | None = None,
|
|
):
|
|
super().__init__()
|
|
patch_size = patch_size if patch_size is not None else [1, 2, 2]
|
|
rope_dim_list = (rope_dim_list if rope_dim_list is not None else [8, 28, 28])
|
|
mouse_qk_dim_list = (mouse_qk_dim_list if mouse_qk_dim_list is not None else [8, 28, 28])
|
|
blocks = blocks if blocks is not None else []
|
|
self.enable_mouse = enable_mouse
|
|
self.enable_keyboard = enable_keyboard
|
|
|
|
self.rope_dim_list = rope_dim_list
|
|
self.rope_theta = rope_theta
|
|
if self.enable_keyboard:
|
|
self.keyboard_embed = nn.Sequential(
|
|
nn.Linear(keyboard_dim_in, hidden_size, bias=True),
|
|
nn.SiLU(),
|
|
nn.Linear(hidden_size, hidden_size, bias=True),
|
|
)
|
|
|
|
self.mouse_qk_dim_list = mouse_qk_dim_list
|
|
self.heads_num = heads_num
|
|
if self.enable_mouse:
|
|
c = mouse_hidden_dim
|
|
self.mouse_mlp = nn.Sequential(
|
|
nn.Linear(
|
|
mouse_dim_in * vae_time_compression_ratio * windows_size + img_hidden_size,
|
|
c,
|
|
bias=True,
|
|
),
|
|
nn.GELU(approximate="tanh"),
|
|
nn.Linear(c, c),
|
|
FP32LayerNorm(c, elementwise_affine=True),
|
|
)
|
|
|
|
head_dim = c // heads_num
|
|
self.t_qkv = ReplicatedLinear(c, c * 3, bias=qkv_bias)
|
|
self.img_attn_q_norm = (WanRMSNorm(head_dim, eps=1e-6) if qk_norm else nn.Identity())
|
|
self.img_attn_k_norm = (WanRMSNorm(head_dim, eps=1e-6) if qk_norm else nn.Identity())
|
|
self.proj_mouse = ReplicatedLinear(c, img_hidden_size, bias=qkv_bias)
|
|
|
|
if self.enable_keyboard:
|
|
head_dim_key = keyboard_hidden_dim // heads_num
|
|
self.key_attn_q_norm = (WanRMSNorm(head_dim_key, eps=1e-6) if qk_norm else nn.Identity())
|
|
self.key_attn_k_norm = (WanRMSNorm(head_dim_key, eps=1e-6) if qk_norm else nn.Identity())
|
|
|
|
self.mouse_attn_q = ReplicatedLinear(img_hidden_size, keyboard_hidden_dim, bias=qkv_bias)
|
|
self.keyboard_attn_kv = ReplicatedLinear(
|
|
hidden_size * windows_size * vae_time_compression_ratio,
|
|
keyboard_hidden_dim * 2,
|
|
bias=qkv_bias,
|
|
)
|
|
self.proj_keyboard = ReplicatedLinear(keyboard_hidden_dim, img_hidden_size, bias=qkv_bias)
|
|
|
|
self.mouse_attn_layer = (LocalAttention(
|
|
num_heads=heads_num,
|
|
head_size=mouse_hidden_dim // heads_num,
|
|
causal=False,
|
|
supported_attention_backends=(
|
|
AttentionBackendEnum.FLASH_ATTN,
|
|
AttentionBackendEnum.TORCH_SDPA,
|
|
),
|
|
) if self.enable_mouse else None)
|
|
|
|
self.keyboard_attn_layer = (LocalAttention(
|
|
num_heads=heads_num,
|
|
head_size=keyboard_hidden_dim // heads_num,
|
|
causal=False,
|
|
supported_attention_backends=(
|
|
AttentionBackendEnum.FLASH_ATTN,
|
|
AttentionBackendEnum.TORCH_SDPA,
|
|
),
|
|
) if self.enable_keyboard else None)
|
|
|
|
self.vae_time_compression_ratio = vae_time_compression_ratio
|
|
self.windows_size = windows_size
|
|
self.patch_size = patch_size
|
|
# Lazy initialization: freqs will be created on first forward pass
|
|
self._freqs_cos = None
|
|
self._freqs_sin = None
|
|
|
|
def patchify(self, x, patch_size):
|
|
"""
|
|
x : (N C T H W)
|
|
"""
|
|
pt, ph, pw = self.patch_size
|
|
t, h, w = x.shape[2] // pt, x.shape[3] // ph, x.shape[4] // pw
|
|
c = x.shape[1]
|
|
x = x.reshape(shape=(x.shape[0], c, t, pt, h, ph, w, pw))
|
|
x = torch.einsum("nctohpwq->nthwcopq", x)
|
|
x = x.reshape(shape=(x.shape[0], t * h * w, c * pt * ph * pw))
|
|
return x
|
|
|
|
def unpatchify(self, x, t, h, w, patch_size):
|
|
"""
|
|
x: (N, T, patch_size**2 * C)
|
|
imgs: (N, H, W, C)
|
|
"""
|
|
c = x.shape[2] // patch_size
|
|
pt, ph, pw = self.patch_size
|
|
assert t * h * w == x.shape[1]
|
|
|
|
x = x.reshape(shape=(x.shape[0], t, h, w, c, pt, ph, pw))
|
|
x = torch.einsum("nthwcopq->nctohpwq", x)
|
|
imgs = x.reshape(shape=(x.shape[0], c, t * pt, h * ph, w * pw))
|
|
|
|
return imgs
|
|
|
|
def get_rotary_pos_embed(
|
|
self,
|
|
video_length,
|
|
height,
|
|
width,
|
|
head_dim,
|
|
rope_dim_list=None,
|
|
):
|
|
target_ndim = 3
|
|
ndim = 5 - 2
|
|
|
|
latents_size = [video_length, height, width]
|
|
|
|
if isinstance(self.patch_size, int):
|
|
assert all(s % self.patch_size == 0 for s in latents_size), (
|
|
f"Latent size(last {ndim} dimensions) should be divisible by patch size({self.patch_size}), "
|
|
f"but got {latents_size}.")
|
|
rope_sizes = [s // self.patch_size for s in latents_size]
|
|
elif isinstance(self.patch_size, list):
|
|
assert all(s % self.patch_size[idx] == 0 for idx, s in enumerate(latents_size)), (
|
|
f"Latent size(last {ndim} dimensions) should be divisible by patch size({self.patch_size}), "
|
|
f"but got {latents_size}.")
|
|
rope_sizes = [s // self.patch_size[idx] for idx, s in enumerate(latents_size)]
|
|
|
|
if len(rope_sizes) != target_ndim:
|
|
rope_sizes = [1] * (target_ndim - len(rope_sizes)) + rope_sizes # time axis
|
|
|
|
if rope_dim_list is None:
|
|
rope_dim_list = [head_dim // target_ndim for _ in range(target_ndim)]
|
|
assert sum(rope_dim_list) == head_dim, ("sum(rope_dim_list) should equal to head_dim of attention layer")
|
|
# Use Matrix-Game wrapper for FastVideo's function
|
|
freqs_cos, freqs_sin = _get_nd_rotary_pos_embed_matrixgame(
|
|
rope_dim_list,
|
|
rope_sizes,
|
|
theta=self.rope_theta,
|
|
theta_rescale_factor=1,
|
|
)
|
|
return freqs_cos[-video_length * rope_sizes[1] * rope_sizes[2] //
|
|
self.patch_size[0]:], freqs_sin[-video_length * rope_sizes[1] * rope_sizes[2] //
|
|
self.patch_size[0]:]
|
|
|
|
def _forward_mouse(
|
|
self,
|
|
hidden_states: torch.Tensor,
|
|
mouse_condition: torch.Tensor,
|
|
mouse_cond_memory: torch.Tensor | None,
|
|
*,
|
|
pad_t: int,
|
|
freqs_cis: tuple[torch.Tensor, torch.Tensor],
|
|
N_feats: int,
|
|
B: int,
|
|
C: int,
|
|
th: int,
|
|
tw: int,
|
|
) -> torch.Tensor:
|
|
pad = mouse_condition[:, 0:1, :].expand(-1, pad_t, -1)
|
|
mouse_condition = torch.cat([pad, mouse_condition], dim=1)
|
|
group_mouse = [
|
|
mouse_condition[
|
|
:,
|
|
self.vae_time_compression_ratio * (i - self.windows_size) + pad_t:i * self.vae_time_compression_ratio +
|
|
pad_t,
|
|
:,
|
|
] for i in range(N_feats)
|
|
]
|
|
group_mouse = torch.stack(group_mouse, dim=1)
|
|
mem_len = 0
|
|
if mouse_cond_memory is not None:
|
|
mem_len = mouse_cond_memory.shape[1]
|
|
mem = mouse_cond_memory.to(device=group_mouse.device, dtype=group_mouse.dtype)
|
|
mem = mem.unsqueeze(2).expand(-1, -1, pad_t, -1)
|
|
group_mouse = torch.cat([mem, group_mouse], dim=1)
|
|
actual_num_frames = group_mouse.shape[1]
|
|
|
|
S = th * tw
|
|
group_mouse = group_mouse.unsqueeze(-1).expand(B, actual_num_frames, pad_t, C, S)
|
|
group_mouse = group_mouse.permute(0, 4, 1, 2, 3).reshape(B * S, actual_num_frames, pad_t * C)
|
|
|
|
group_mouse = torch.cat([hidden_states, group_mouse], dim=-1)
|
|
group_mouse = self.mouse_mlp(group_mouse)
|
|
mouse_qkv, _ = self.t_qkv(group_mouse)
|
|
q, k, v = rearrange(mouse_qkv, "B L (K H D) -> K B L H D", K=3, H=self.heads_num)
|
|
q = self.img_attn_q_norm(q).to(v)
|
|
k = self.img_attn_k_norm(k).to(v)
|
|
if mem_len > 0:
|
|
q_mem, k_mem = _apply_rotary_emb_qk(q[:, :mem_len], k[:, :mem_len], freqs_cis[0], freqs_cis[1])
|
|
q_pred, k_pred = _apply_rotary_emb_qk(q[:, mem_len:], k[:, mem_len:], freqs_cis[0], freqs_cis[1])
|
|
q = torch.cat([q_mem, q_pred], dim=1)
|
|
k = torch.cat([k_mem, k_pred], dim=1)
|
|
else:
|
|
q, k = _apply_rotary_emb_qk(q, k, freqs_cis[0], freqs_cis[1])
|
|
|
|
attn = self.mouse_attn_layer(q, k, v)
|
|
attn = rearrange(attn, "(b S) T h d -> b (T S) (h d)", b=B)
|
|
attn, _ = self.proj_mouse(attn)
|
|
return attn
|
|
|
|
def _forward_keyboard(
|
|
self,
|
|
hidden_states: torch.Tensor,
|
|
keyboard_condition: torch.Tensor,
|
|
keyboard_cond_memory: torch.Tensor | None,
|
|
*,
|
|
pad_t: int,
|
|
freqs_cis: tuple[torch.Tensor, torch.Tensor],
|
|
N_feats: int,
|
|
B: int,
|
|
th: int,
|
|
tw: int,
|
|
) -> torch.Tensor:
|
|
pad = keyboard_condition[:, 0:1, :].expand(-1, pad_t, -1)
|
|
keyboard_condition = torch.cat([pad, keyboard_condition], dim=1)
|
|
keyboard_condition = self.keyboard_embed(keyboard_condition)
|
|
group_keyboard = [
|
|
keyboard_condition[
|
|
:,
|
|
self.vae_time_compression_ratio * (i - self.windows_size) + pad_t:i * self.vae_time_compression_ratio +
|
|
pad_t,
|
|
:,
|
|
] for i in range(N_feats)
|
|
]
|
|
group_keyboard = torch.stack(group_keyboard, dim=1)
|
|
mem_len = 0
|
|
if keyboard_cond_memory is not None:
|
|
mem_len = keyboard_cond_memory.shape[1]
|
|
k_mem = keyboard_cond_memory.to(device=group_keyboard.device, dtype=group_keyboard.dtype)
|
|
k_mem = self.keyboard_embed(k_mem)
|
|
k_mem = k_mem.unsqueeze(2).expand(-1, -1, pad_t, -1)
|
|
group_keyboard = torch.cat([k_mem, group_keyboard], dim=1)
|
|
group_keyboard = group_keyboard.reshape(shape=(group_keyboard.shape[0], group_keyboard.shape[1], -1))
|
|
|
|
mouse_q, _ = self.mouse_attn_q(hidden_states)
|
|
keyboard_kv, _ = self.keyboard_attn_kv(group_keyboard)
|
|
|
|
B, L, HD = mouse_q.shape
|
|
D = HD // self.heads_num
|
|
q = mouse_q.view(B, L, self.heads_num, D)
|
|
|
|
B, L, KHD = keyboard_kv.shape
|
|
k, v = keyboard_kv.view(B, L, 2, self.heads_num, D).permute(2, 0, 1, 3, 4)
|
|
|
|
q = self.key_attn_q_norm(q).to(v)
|
|
k = self.key_attn_k_norm(k).to(v)
|
|
S = th * tw
|
|
|
|
B, TS, H, D = q.shape
|
|
T_ = TS // S
|
|
q = q.view(B, T_, S, H, D).transpose(1, 2).reshape(B * S, T_, H, D)
|
|
if mem_len > 0:
|
|
q_mem, k_mem_r = _apply_rotary_emb_qk(q[:, :mem_len], k[:, :mem_len], freqs_cis[0], freqs_cis[1])
|
|
q_pred, k_pred_r = _apply_rotary_emb_qk(q[:, mem_len:], k[:, mem_len:], freqs_cis[0], freqs_cis[1])
|
|
q = torch.cat([q_mem, q_pred], dim=1)
|
|
k = torch.cat([k_mem_r, k_pred_r], dim=1)
|
|
else:
|
|
q, k = _apply_rotary_emb_qk(q, k, freqs_cis[0], freqs_cis[1])
|
|
|
|
k = k.repeat_interleave(S, dim=0)
|
|
v = v.repeat_interleave(S, dim=0)
|
|
|
|
attn = self.keyboard_attn_layer(q, k, v)
|
|
attn = rearrange(attn, "(B S) T H D -> B (T S) (H D)", S=S)
|
|
attn, _ = self.proj_keyboard(attn)
|
|
return attn
|
|
|
|
def forward(
|
|
self,
|
|
x: torch.Tensor,
|
|
tt: int,
|
|
th: int,
|
|
tw: int,
|
|
mouse_condition: torch.Tensor | None = None,
|
|
keyboard_condition: torch.Tensor | None = None,
|
|
mouse_cond_memory: torch.Tensor | None = None,
|
|
keyboard_cond_memory: torch.Tensor | None = None,
|
|
):
|
|
"""
|
|
hidden_states: B, tt*th*tw, C
|
|
mouse_condition: B, N_frames, C1
|
|
keyboard_condition: B, N_frames, C2
|
|
"""
|
|
target_device = x.device
|
|
target_dtype = x.dtype
|
|
if mouse_condition is not None:
|
|
mouse_condition = mouse_condition.to(device=target_device, dtype=target_dtype)
|
|
if keyboard_condition is not None:
|
|
keyboard_condition = keyboard_condition.to(device=target_device, dtype=target_dtype)
|
|
else:
|
|
return x
|
|
|
|
B, N_frames, C = keyboard_condition.shape
|
|
assert tt * th * tw == x.shape[1]
|
|
assert (((N_frames - 1) + self.vae_time_compression_ratio) % self.vae_time_compression_ratio == 0
|
|
or (N_frames % self.vae_time_compression_ratio == 0))
|
|
if ((N_frames - 1) + self.vae_time_compression_ratio) % self.vae_time_compression_ratio == 0:
|
|
N_feats = int((N_frames - 1) / self.vae_time_compression_ratio) + 1
|
|
else:
|
|
N_feats = N_frames // self.vae_time_compression_ratio
|
|
|
|
# Lazy initialization of freqs
|
|
if (self._freqs_cos is None or self._freqs_sin is None or self._freqs_cos.device != x.device):
|
|
fc, fs = self.get_rotary_pos_embed(
|
|
7500,
|
|
self.patch_size[1],
|
|
self.patch_size[2],
|
|
64,
|
|
self.mouse_qk_dim_list,
|
|
)
|
|
self._freqs_cos = fc.to(x.device)
|
|
self._freqs_sin = fs.to(x.device)
|
|
|
|
freqs_cis = (self._freqs_cos, self._freqs_sin)
|
|
|
|
if self.enable_mouse and mouse_condition is not None:
|
|
hidden_states = rearrange(x, "B (T S) C -> (B S) T C", T=tt, S=th * tw)
|
|
B, N_frames, C = mouse_condition.shape
|
|
else:
|
|
hidden_states = x
|
|
|
|
pad_t = self.vae_time_compression_ratio * self.windows_size
|
|
if self.enable_mouse and mouse_condition is not None:
|
|
attn = self._forward_mouse(
|
|
hidden_states,
|
|
mouse_condition,
|
|
mouse_cond_memory,
|
|
pad_t=pad_t,
|
|
freqs_cis=freqs_cis,
|
|
N_feats=N_feats,
|
|
B=B,
|
|
C=C,
|
|
th=th,
|
|
tw=tw,
|
|
)
|
|
hidden_states = rearrange(x, "(B S) T C -> B (T S) C", B=B)
|
|
hidden_states = hidden_states + attn
|
|
|
|
if self.enable_keyboard and keyboard_condition is not None:
|
|
attn = self._forward_keyboard(
|
|
hidden_states,
|
|
keyboard_condition,
|
|
keyboard_cond_memory,
|
|
pad_t=pad_t,
|
|
freqs_cis=freqs_cis,
|
|
N_feats=N_feats,
|
|
B=B,
|
|
th=th,
|
|
tw=tw,
|
|
)
|
|
hidden_states = hidden_states + attn
|
|
return hidden_states
|