transformer/causal_model.py

from .attention import attention
from .model import (
    MatrixGameWanRMSNorm,
    rope_apply,
    MatrixGameWanLayerNorm,
    MatrixGameWan_CROSSATTENTION_CLASSES,
    rope_params,
    MLPProj,
    sinusoidal_embedding_1d,
)
from diffusers.loaders import FromOriginalModelMixin, PeftAdapterMixin
from torch.nn.attention.flex_attention import create_block_mask, flex_attention
from diffusers.configuration_utils import ConfigMixin, register_to_config
from torch.nn.attention.flex_attention import BlockMask
from diffusers.models.modeling_utils import ModelMixin
import torch.nn as nn
import torch
import math
import torch.distributed as dist
from .action_module import ActionModule


def causal_rope_apply(x, grid_sizes, freqs, start_frame=0):
    n, c = x.size(2), x.size(3) // 2

    # split freqs
    freqs = freqs.split([c - 2 * (c // 3), c // 3, c // 3], dim=1)

    # loop over samples
    output = []
    f, h, w = grid_sizes.tolist()

    for i in range(len(x)):
        seq_len = f * h * w

        # precompute multipliers
        x_i = torch.view_as_complex(
            x[i, :seq_len].to(torch.float64).reshape(seq_len, n, -1, 2)
        )
        freqs_i = torch.cat(
            [
                freqs[0][start_frame : start_frame + f]
                .view(f, 1, 1, -1)
                .expand(f, h, w, -1),
                freqs[1][:h].view(1, h, 1, -1).expand(f, h, w, -1),
                freqs[2][:w].view(1, 1, w, -1).expand(f, h, w, -1),
            ],
            dim=-1,
        ).reshape(seq_len, 1, -1)

        # apply rotary embedding
        x_i = torch.view_as_real(x_i * freqs_i).flatten(2)
        x_i = torch.cat([x_i, x[i, seq_len:]])

        # append to collection
        output.append(x_i)
    return torch.stack(output).type_as(x)


class MatrixGameWanCausalSelfAttention(nn.Module):
    def __init__(
        self, dim, num_heads, local_attn_size=-1, sink_size=0, qk_norm=True, eps=1e-6
    ):
        assert dim % num_heads == 0
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.head_dim = dim // num_heads
        self.local_attn_size = local_attn_size
        self.sink_size = sink_size
        self.qk_norm = qk_norm
        self.eps = eps
        self.max_attention_size = (
            15 * 1 * 880 if local_attn_size == -1 else local_attn_size * 880
        )
        # layers
        self.q = nn.Linear(dim, dim)
        self.k = nn.Linear(dim, dim)
        self.v = nn.Linear(dim, dim)
        self.o = nn.Linear(dim, dim)
        self.norm_q = MatrixGameWanRMSNorm(dim, eps=eps) if qk_norm else nn.Identity()
        self.norm_k = MatrixGameWanRMSNorm(dim, eps=eps) if qk_norm else nn.Identity()

    def forward(
        self,
        x,
        seq_lens,
        grid_sizes,
        freqs,
        block_mask,
        kv_cache=None,
        current_start=0,
        cache_start=None,
    ):
        r"""
        Args:
            x(Tensor): Shape [B, L, C] # num_heads, C / num_heads]
            seq_lens(Tensor): Shape [B]
            grid_sizes(Tensor): Shape [B, 3], the second dimension contains (F, H, W)
            freqs(Tensor): Rope freqs, shape [1024, C / num_heads / 2]
            block_mask (BlockMask)
        """
        b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim
        if cache_start is None:
            cache_start = current_start

        # query, key, value function
        def qkv_fn(x):
            q = self.norm_q(self.q(x)).view(b, s, n, d)
            k = self.norm_k(self.k(x)).view(b, s, n, d)
            v = self.v(x).view(b, s, n, d)
            return q, k, v

        q, k, v = qkv_fn(x)  # B, F, HW, C

        if kv_cache is None:
            roped_query = rope_apply(q, grid_sizes, freqs).type_as(v)
            roped_key = rope_apply(k, grid_sizes, freqs).type_as(v)

            padded_length = math.ceil(q.shape[1] / 128) * 128 - q.shape[1]
            padded_roped_query = torch.cat(
                [
                    roped_query,
                    torch.zeros(
                        [q.shape[0], padded_length, q.shape[2], q.shape[3]],
                        device=q.device,
                        dtype=v.dtype,
                    ),
                ],
                dim=1,
            )

            padded_roped_key = torch.cat(
                [
                    roped_key,
                    torch.zeros(
                        [k.shape[0], padded_length, k.shape[2], k.shape[3]],
                        device=k.device,
                        dtype=v.dtype,
                    ),
                ],
                dim=1,
            )

            padded_v = torch.cat(
                [
                    v,
                    torch.zeros(
                        [v.shape[0], padded_length, v.shape[2], v.shape[3]],
                        device=v.device,
                        dtype=v.dtype,
                    ),
                ],
                dim=1,
            )

            x = flex_attention(
                query=padded_roped_query.transpose(2, 1),  # after: B, HW, F, C
                key=padded_roped_key.transpose(2, 1),
                value=padded_v.transpose(2, 1),
                block_mask=block_mask,
            )[:, :, :-padded_length].transpose(2, 1)
        else:
            assert grid_sizes.ndim == 1
            frame_seqlen = math.prod(grid_sizes[1:]).item()
            current_start_frame = current_start // frame_seqlen
            roped_query = causal_rope_apply(
                q, grid_sizes, freqs, start_frame=current_start_frame
            ).type_as(v)
            roped_key = causal_rope_apply(
                k, grid_sizes, freqs, start_frame=current_start_frame
            ).type_as(v)

            current_end = current_start + roped_query.shape[1]
            sink_tokens = self.sink_size * frame_seqlen

            kv_cache_size = kv_cache["k"].shape[1]
            num_new_tokens = roped_query.shape[1]

            if (current_end > kv_cache["global_end_index"].item()) and (
                num_new_tokens + kv_cache["local_end_index"].item() > kv_cache_size
            ):
                num_evicted_tokens = (
                    num_new_tokens + kv_cache["local_end_index"].item() - kv_cache_size
                )
                num_rolled_tokens = (
                    kv_cache["local_end_index"].item()
                    - num_evicted_tokens
                    - sink_tokens
                )
                kv_cache["k"][:, sink_tokens : sink_tokens + num_rolled_tokens] = (
                    kv_cache["k"][
                        :,
                        sink_tokens + num_evicted_tokens : sink_tokens
                        + num_evicted_tokens
                        + num_rolled_tokens,
                    ].clone()
                )
                kv_cache["v"][:, sink_tokens : sink_tokens + num_rolled_tokens] = (
                    kv_cache["v"][
                        :,
                        sink_tokens + num_evicted_tokens : sink_tokens
                        + num_evicted_tokens
                        + num_rolled_tokens,
                    ].clone()
                )
                # Insert the new keys/values at the end
                local_end_index = (
                    kv_cache["local_end_index"].item()
                    + current_end
                    - kv_cache["global_end_index"].item()
                    - num_evicted_tokens
                )
                local_start_index = local_end_index - num_new_tokens
                kv_cache["k"][:, local_start_index:local_end_index] = roped_key
                kv_cache["v"][:, local_start_index:local_end_index] = v
            else:
                # Assign new keys/values directly up to current_end
                local_end_index = (
                    kv_cache["local_end_index"].item()
                    + current_end
                    - kv_cache["global_end_index"].item()
                )
                local_start_index = local_end_index - num_new_tokens

                kv_cache["k"][:, local_start_index:local_end_index] = roped_key
                kv_cache["v"][:, local_start_index:local_end_index] = v
            x = attention(
                roped_query,
                kv_cache["k"][
                    :,
                    max(0, local_end_index - self.max_attention_size) : local_end_index,
                ],
                kv_cache["v"][
                    :,
                    max(0, local_end_index - self.max_attention_size) : local_end_index,
                ],
            )
            kv_cache["global_end_index"].fill_(current_end)
            kv_cache["local_end_index"].fill_(local_end_index)

        # output
        x = x.flatten(2)
        x = self.o(x)
        return x


class MatrixGameWanCausalAttentionBlock(nn.Module):
    def __init__(
        self,
        cross_attn_type,
        dim,
        ffn_dim,
        num_heads,
        local_attn_size=-1,
        sink_size=0,
        qk_norm=True,
        cross_attn_norm=False,
        action_config={},
        block_idx=0,
        eps=1e-6,
    ):
        super().__init__()
        self.dim = dim
        self.ffn_dim = ffn_dim
        self.num_heads = num_heads
        self.local_attn_size = local_attn_size
        self.qk_norm = qk_norm
        self.cross_attn_norm = cross_attn_norm
        self.eps = eps
        if len(action_config) != 0 and block_idx in action_config["blocks"]:
            self.action_model = ActionModule(
                **action_config, local_attn_size=self.local_attn_size
            )
        else:
            self.action_model = None
        # layers
        self.norm1 = MatrixGameWanLayerNorm(dim, eps)
        self.self_attn = MatrixGameWanCausalSelfAttention(
            dim, num_heads, local_attn_size, sink_size, qk_norm, eps
        )
        self.norm3 = (
            MatrixGameWanLayerNorm(dim, eps, elementwise_affine=True)
            if cross_attn_norm
            else nn.Identity()
        )
        self.cross_attn = MatrixGameWan_CROSSATTENTION_CLASSES[cross_attn_type](
            dim, num_heads, (-1, -1), qk_norm, eps
        )
        self.norm2 = MatrixGameWanLayerNorm(dim, eps)
        self.ffn = nn.Sequential(
            nn.Linear(dim, ffn_dim),
            nn.GELU(approximate="tanh"),
            nn.Linear(ffn_dim, dim),
        )

        # modulation
        self.modulation = nn.Parameter(torch.randn(1, 6, dim) / dim**0.5)

    def forward(
        self,
        x,
        e,
        seq_lens,
        grid_sizes,
        freqs,
        context,
        block_mask,
        block_mask_mouse,
        block_mask_keyboard,
        num_frame_per_block=3,
        use_rope_keyboard=False,
        mouse_cond=None,
        keyboard_cond=None,
        kv_cache=None,
        kv_cache_mouse=None,
        kv_cache_keyboard=None,
        crossattn_cache=None,
        current_start=0,
        cache_start=None,
        context_lens=None,
    ):
        r"""
        Args:
            x(Tensor): Shape [B, L, C]
            e(Tensor): Shape [B, F, 6, C]
            seq_lens(Tensor): Shape [B], length of each sequence in batch
            grid_sizes(Tensor): Shape [B, 3], the second dimension contains (F, H, W)
            freqs(Tensor): Rope freqs, shape [1024, C / num_heads / 2]
        """
        assert e.ndim == 4
        num_frames, frame_seqlen = e.shape[1], x.shape[1] // e.shape[1]

        e = (self.modulation.unsqueeze(1) + e).chunk(6, dim=2)

        y = self.self_attn(
            (
                self.norm1(x).unflatten(dim=1, sizes=(num_frames, frame_seqlen))
                * (1 + e[1])
                + e[0]
            ).flatten(1, 2),
            seq_lens,
            grid_sizes,
            freqs,
            block_mask,
            kv_cache,
            current_start,
            cache_start,
        )

        x = x + (y.unflatten(dim=1, sizes=(num_frames, frame_seqlen)) * e[2]).flatten(
            1, 2
        )

        # cross-attention & ffn function
        def cross_attn_ffn(
            x,
            context,
            e,
            mouse_cond,
            keyboard_cond,
            block_mask_mouse,
            block_mask_keyboard,
            kv_cache_mouse=None,
            kv_cache_keyboard=None,
            crossattn_cache=None,
            start_frame=0,
            use_rope_keyboard=False,
            num_frame_per_block=3,
        ):
            x = x + self.cross_attn(
                self.norm3(x.to(context.dtype)),
                context,
                crossattn_cache=crossattn_cache,
            )
            if self.action_model is not None:
                assert mouse_cond is not None or keyboard_cond is not None
                x = self.action_model(
                    x.to(context.dtype),
                    grid_sizes[0],
                    grid_sizes[1],
                    grid_sizes[2],
                    mouse_cond,
                    keyboard_cond,
                    block_mask_mouse,
                    block_mask_keyboard,
                    is_causal=True,
                    kv_cache_mouse=kv_cache_mouse,
                    kv_cache_keyboard=kv_cache_keyboard,
                    start_frame=start_frame,
                    use_rope_keyboard=use_rope_keyboard,
                    num_frame_per_block=num_frame_per_block,
                )

            y = self.ffn(
                (
                    self.norm2(x).unflatten(dim=1, sizes=(num_frames, frame_seqlen))
                    * (1 + e[4])
                    + e[3]
                ).flatten(1, 2)
            )

            x = x + (
                y.unflatten(dim=1, sizes=(num_frames, frame_seqlen)) * e[5]
            ).flatten(1, 2)
            return x

        assert grid_sizes.ndim == 1
        x = cross_attn_ffn(
            x,
            context,
            e,
            mouse_cond,
            keyboard_cond,
            block_mask_mouse,
            block_mask_keyboard,
            kv_cache_mouse,
            kv_cache_keyboard,
            crossattn_cache,
            start_frame=current_start // math.prod(grid_sizes[1:]).item(),
            use_rope_keyboard=use_rope_keyboard,
            num_frame_per_block=num_frame_per_block,
        )
        return x


class CausalHead(nn.Module):
    def __init__(self, dim, out_dim, patch_size, eps=1e-6):
        super().__init__()
        self.dim = dim
        self.out_dim = out_dim
        self.patch_size = patch_size
        self.eps = eps

        # layers
        out_dim = math.prod(patch_size) * out_dim
        self.norm = MatrixGameWanLayerNorm(dim, eps)
        self.head = nn.Linear(dim, out_dim)

        # modulation
        self.modulation = nn.Parameter(torch.randn(1, 2, dim) / dim**0.5)

    def forward(self, x, e):
        r"""
        Args:
            x(Tensor): Shape [B, L1, C]
            e(Tensor): Shape [B, F, 1, C]
        """

        num_frames, frame_seqlen = e.shape[1], x.shape[1] // e.shape[1]
        e = (self.modulation.unsqueeze(1) + e).chunk(2, dim=2)
        x = self.head(
            self.norm(x).unflatten(dim=1, sizes=(num_frames, frame_seqlen)) * (1 + e[1])
            + e[0]
        )
        return x


class MatrixGameWanCausalModel(ModelMixin, ConfigMixin, FromOriginalModelMixin, PeftAdapterMixin):
    r"""
    MatrixGameWan diffusion backbone supporting both text-to-video and image-to-video.
    """

    ignore_for_config = ["patch_size", "cross_attn_norm", "qk_norm", "text_dim"]
    _no_split_modules = ["MatrixGameWanAttentionBlock"]
    _supports_gradient_checkpointing = True

    @register_to_config
    def __init__(
        self,
        model_type="t2v",
        patch_size=(1, 2, 2),
        text_len=512,
        in_dim=36,
        dim=1536,
        ffn_dim=8960,
        freq_dim=256,
        text_dim=4096,
        out_dim=16,
        num_heads=12,
        num_layers=30,
        local_attn_size=-1,
        sink_size=0,
        qk_norm=True,
        cross_attn_norm=True,
        action_config={},
        eps=1e-6,
    ):
        r"""
        Initialize the diffusion model backbone.

        Args:
            model_type (`str`, *optional*, defaults to 't2v'):
                Model variant - 't2v' (text-to-video) or 'i2v' (image-to-video)
            patch_size (`tuple`, *optional*, defaults to (1, 2, 2)):
                3D patch dimensions for video embedding (t_patch, h_patch, w_patch)
            text_len (`int`, *optional*, defaults to 512):
                Fixed length for text embeddings
            in_dim (`int`, *optional*, defaults to 16):
                Input video channels (C_in)
            dim (`int`, *optional*, defaults to 2048):
                Hidden dimension of the transformer
            ffn_dim (`int`, *optional*, defaults to 8192):
                Intermediate dimension in feed-forward network
            freq_dim (`int`, *optional*, defaults to 256):
                Dimension for sinusoidal time embeddings
            text_dim (`int`, *optional*, defaults to 4096):
                Input dimension for text embeddings
            out_dim (`int`, *optional*, defaults to 16):
                Output video channels (C_out)
            num_heads (`int`, *optional*, defaults to 16):
                Number of attention heads
            num_layers (`int`, *optional*, defaults to 32):
                Number of transformer blocks
            local_attn_size (`int`, *optional*, defaults to -1):
                Window size for temporal local attention (-1 indicates global attention)
            sink_size (`int`, *optional*, defaults to 0):
                Size of the attention sink, we keep the first `sink_size` frames unchanged when rolling the KV cache
            qk_norm (`bool`, *optional*, defaults to True):
                Enable query/key normalization
            cross_attn_norm (`bool`, *optional*, defaults to False):
                Enable cross-attention normalization
            eps (`float`, *optional*, defaults to 1e-6):
                Epsilon value for normalization layers
        """

        super().__init__()

        assert model_type in ["i2v"]
        self.model_type = model_type
        self.use_action_module = len(action_config) > 0
        self.patch_size = patch_size
        self.text_len = text_len
        self.in_dim = in_dim
        self.dim = dim
        self.ffn_dim = ffn_dim
        self.freq_dim = freq_dim
        self.text_dim = text_dim
        self.out_dim = out_dim
        self.num_heads = num_heads
        self.num_layers = num_layers
        self.local_attn_size = local_attn_size
        self.qk_norm = qk_norm
        self.cross_attn_norm = cross_attn_norm
        self.eps = eps

        # embeddings
        self.patch_embedding = nn.Conv3d(
            in_dim, dim, kernel_size=patch_size, stride=patch_size
        )

        self.time_embedding = nn.Sequential(
            nn.Linear(freq_dim, dim), nn.SiLU(), nn.Linear(dim, dim)
        )
        self.time_projection = nn.Sequential(nn.SiLU(), nn.Linear(dim, dim * 6))

        # blocks
        cross_attn_type = "i2v_cross_attn"
        self.blocks = nn.ModuleList(
            [
                MatrixGameWanCausalAttentionBlock(
                    cross_attn_type,
                    dim,
                    ffn_dim,
                    num_heads,
                    local_attn_size,
                    sink_size,
                    qk_norm,
                    cross_attn_norm,
                    action_config=action_config,
                    eps=eps,
                    block_idx=idx,
                )
                for idx in range(num_layers)
            ]
        )

        # head
        self.head = CausalHead(dim, out_dim, patch_size, eps)

        # buffers (don't use register_buffer otherwise dtype will be changed in to())
        assert (dim % num_heads) == 0 and (dim // num_heads) % 2 == 0
        d = dim // num_heads
        self.freqs = torch.cat(
            [
                rope_params(1024, d - 4 * (d // 6)),
                rope_params(1024, 2 * (d // 6)),
                rope_params(1024, 2 * (d // 6)),
            ],
            dim=1,
        )

        if model_type == "i2v":
            self.img_emb = MLPProj(1280, dim)

        self.gradient_checkpointing = False

        self.block_mask = None
        self.block_mask_keyboard = None
        self.block_mask_mouse = None
        self.use_rope_keyboard = True

    def _set_gradient_checkpointing(self, module, value=False):
        self.gradient_checkpointing = value

    @staticmethod
    def _prepare_blockwise_causal_attn_mask(
        device: torch.device | str,
        num_frames: int = 9,
        frame_seqlen: int = 880,
        num_frame_per_block=1,
        local_attn_size=-1,
    ) -> BlockMask:
        """
        we will divide the token sequence into the following format
        [1 latent frame] [1 latent frame] ... [1 latent frame]
        We use flexattention to construct the attention mask
        """
        total_length = num_frames * frame_seqlen

        # we do right padding to get to a multiple of 128
        padded_length = math.ceil(total_length / 128) * 128 - total_length

        ends = torch.zeros(
            total_length + padded_length, device=device, dtype=torch.long
        )

        # Block-wise causal mask will attend to all elements that are before the end of the current chunk
        frame_indices = torch.arange(
            start=0,
            end=total_length,
            step=frame_seqlen * num_frame_per_block,
            device=device,
        )

        for tmp in frame_indices:
            ends[tmp : tmp + frame_seqlen * num_frame_per_block] = (
                tmp + frame_seqlen * num_frame_per_block
            )

        def attention_mask(b, h, q_idx, kv_idx):
            if local_attn_size == -1:
                return (kv_idx < ends[q_idx]) | (q_idx == kv_idx)
            else:
                return (
                    (kv_idx < ends[q_idx])
                    & (kv_idx >= (ends[q_idx] - local_attn_size * frame_seqlen))
                ) | (q_idx == kv_idx)
            # return ((kv_idx < total_length) & (q_idx < total_length))  | (q_idx == kv_idx) # bidirectional mask

        block_mask = create_block_mask(
            attention_mask,
            B=None,
            H=None,
            Q_LEN=total_length + padded_length,
            KV_LEN=total_length + padded_length,
            _compile=False,
            device=device,
        )

        import torch.distributed as dist

        if not dist.is_initialized() or dist.get_rank() == 0:
            print(
                f" cache a block wise causal mask with block size of {num_frame_per_block} frames"
            )

        return block_mask

    @staticmethod
    def _prepare_blockwise_causal_attn_mask_keyboard(
        device: torch.device | str,
        num_frames: int = 9,
        frame_seqlen: int = 880,
        num_frame_per_block=1,
        local_attn_size=-1,
    ) -> BlockMask:
        """
        we will divide the token sequence into the following format
        [1 latent frame] [1 latent frame] ... [1 latent frame]
        We use flexattention to construct the attention mask
        """
        total_length2 = num_frames * frame_seqlen

        # we do right padding to get to a multiple of 128
        padded_length2 = math.ceil(total_length2 / 32) * 32 - total_length2
        padded_length_kv2 = math.ceil(num_frames / 32) * 32 - num_frames
        ends2 = torch.zeros(
            total_length2 + padded_length2, device=device, dtype=torch.long
        )

        # Block-wise causal mask will attend to all elements that are before the end of the current chunk
        frame_indices2 = torch.arange(
            start=0,
            end=total_length2,
            step=frame_seqlen * num_frame_per_block,
            device=device,
        )
        cnt = num_frame_per_block
        for tmp in frame_indices2:
            ends2[tmp : tmp + frame_seqlen * num_frame_per_block] = cnt
            cnt += num_frame_per_block

        def attention_mask2(b, h, q_idx, kv_idx):
            if local_attn_size == -1:
                return (kv_idx < ends2[q_idx]) | (q_idx == kv_idx)
            else:
                return (
                    (kv_idx < ends2[q_idx])
                    & (kv_idx >= (ends2[q_idx] - local_attn_size))
                ) | (q_idx == kv_idx)
            # return ((kv_idx < total_length) & (q_idx < total_length))  | (q_idx == kv_idx) # bidirectional mask

        block_mask2 = create_block_mask(
            attention_mask2,
            B=None,
            H=None,
            Q_LEN=total_length2 + padded_length2,
            KV_LEN=num_frames + padded_length_kv2,
            _compile=False,
            device=device,
        )

        import torch.distributed as dist

        if not dist.is_initialized() or dist.get_rank() == 0:
            print(
                f" cache a block wise causal mask with block size of {num_frame_per_block} frames"
            )

        return block_mask2

    @staticmethod
    def _prepare_blockwise_causal_attn_mask_action(
        device: torch.device | str,
        num_frames: int = 9,
        frame_seqlen: int = 1,
        num_frame_per_block=1,
        local_attn_size=-1,
    ) -> BlockMask:
        """
        we will divide the token sequence into the following format
        [1 latent frame] [1 latent frame] ... [1 latent frame]
        We use flexattention to construct the attention mask
        """
        total_length2 = num_frames * frame_seqlen

        # we do right padding to get to a multiple of 128
        padded_length2 = math.ceil(total_length2 / 32) * 32 - total_length2
        padded_length_kv2 = math.ceil(num_frames / 32) * 32 - num_frames
        ends2 = torch.zeros(
            total_length2 + padded_length2, device=device, dtype=torch.long
        )

        # Block-wise causal mask will attend to all elements that are before the end of the current chunk
        frame_indices2 = torch.arange(
            start=0,
            end=total_length2,
            step=frame_seqlen * num_frame_per_block,
            device=device,
        )
        cnt = num_frame_per_block
        for tmp in frame_indices2:
            ends2[tmp : tmp + frame_seqlen * num_frame_per_block] = cnt
            cnt += num_frame_per_block

        def attention_mask2(b, h, q_idx, kv_idx):
            if local_attn_size == -1:
                return (kv_idx < ends2[q_idx]) | (q_idx == kv_idx)
            else:
                return (
                    (kv_idx < ends2[q_idx])
                    & (kv_idx >= (ends2[q_idx] - local_attn_size))
                ) | (q_idx == kv_idx)
            # return ((kv_idx < total_length) & (q_idx < total_length))  | (q_idx == kv_idx) # bidirectional mask

        block_mask2 = create_block_mask(
            attention_mask2,
            B=None,
            H=None,
            Q_LEN=total_length2 + padded_length2,
            KV_LEN=num_frames + padded_length_kv2,
            _compile=False,
            device=device,
        )

        import torch.distributed as dist

        if not dist.is_initialized() or dist.get_rank() == 0:
            print(
                f" cache a block wise causal mask with block size of {num_frame_per_block} frames"
            )

        return block_mask2

    def _forward_inference(
        self,
        x,
        t,
        visual_context,
        cond_concat,
        mouse_cond=None,
        keyboard_cond=None,
        kv_cache: dict = None,
        kv_cache_mouse=None,
        kv_cache_keyboard=None,
        crossattn_cache: dict = None,
        current_start: int = 0,
        cache_start: int = 0,
        num_frames_per_block=3,
    ):
        r"""
        Run the diffusion model with kv caching.
        See Algorithm 2 of CausVid paper https://arxiv.org/abs/2412.07772 for details.
        This function will be run for num_frame times.
        Process the latent frames one by one (1560 tokens each)

        Args:
            x (List[Tensor]):
                List of input video tensors, each with shape [C_in, F, H, W]
            t (Tensor):
                Diffusion timesteps tensor of shape [B]
            context (List[Tensor]):
                List of text embeddings each with shape [L, C]
            seq_len (`int`):
                Maximum sequence length for positional encoding
            clip_fea (Tensor, *optional*):
                CLIP image features for image-to-video mode
            y (List[Tensor], *optional*):
                Conditional video inputs for image-to-video mode, same shape as x

        Returns:
            List[Tensor]:
                List of denoised video tensors with original input shapes [C_out, F, H / 8, W / 8]
        """

        if mouse_cond is not None or keyboard_cond is not None:
            assert self.use_action_module == True
        # params
        device = self.patch_embedding.weight.device
        if self.freqs.device != device:
            self.freqs = self.freqs.to(device)

        x = torch.cat([x, cond_concat], dim=1)  # B C' F H W

        # embeddings
        x = self.patch_embedding(x)
        grid_sizes = torch.tensor(x.shape[2:], dtype=torch.long)

        x = x.flatten(2).transpose(1, 2)  # B FHW C'
        seq_lens = torch.tensor([u.size(0) for u in x], dtype=torch.long)
        assert seq_lens[0] <= 15 * 1 * 880

        e = self.time_embedding(
            sinusoidal_embedding_1d(self.freq_dim, t.flatten()).type_as(x)
        )
        e0 = (
            self.time_projection(e)
            .unflatten(1, (6, self.dim))
            .unflatten(dim=0, sizes=t.shape)
        )
        # context
        context_lens = None
        context = self.img_emb(visual_context)
        # arguments
        kwargs = dict(
            e=e0,
            seq_lens=seq_lens,
            grid_sizes=grid_sizes,
            freqs=self.freqs,
            context=context,
            mouse_cond=mouse_cond,
            context_lens=context_lens,
            keyboard_cond=keyboard_cond,
            block_mask=self.block_mask,
            block_mask_mouse=self.block_mask_mouse,
            block_mask_keyboard=self.block_mask_keyboard,
            use_rope_keyboard=self.use_rope_keyboard,
            num_frame_per_block=num_frames_per_block,
        )

        def create_custom_forward(module):
            def custom_forward(*inputs, **kwargs):
                return module(*inputs, **kwargs)

            return custom_forward

        for block_index, block in enumerate(self.blocks):
            if torch.is_grad_enabled() and self.gradient_checkpointing:
                kwargs.update(
                    {
                        "kv_cache": kv_cache[block_index],
                        "kv_cache_mouse": kv_cache_mouse[block_index],
                        "kv_cache_keyboard": kv_cache_keyboard[block_index],
                        "current_start": current_start,
                        "cache_start": cache_start,
                    }
                )
                x = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(block),
                    x,
                    **kwargs,
                    use_reentrant=False,
                )
            else:
                kwargs.update(
                    {
                        "kv_cache": kv_cache[block_index],
                        "kv_cache_mouse": kv_cache_mouse[block_index],
                        "kv_cache_keyboard": kv_cache_keyboard[block_index],
                        "crossattn_cache": crossattn_cache[block_index],
                        "current_start": current_start,
                        "cache_start": cache_start,
                    }
                )
                x = block(x, **kwargs)

        # head
        x = self.head(x, e.unflatten(dim=0, sizes=t.shape).unsqueeze(2))
        # unpatchify
        x = self.unpatchify(x, grid_sizes)
        return x

    def forward(self, *args, **kwargs):
        return self._forward_inference(*args, **kwargs)

    def unpatchify(self, x, grid_sizes):
        r"""
        Reconstruct video tensors from patch embeddings.

        Args:
            x (List[Tensor]):
                List of patchified features, each with shape [L, C_out * prod(patch_size)]
            grid_sizes (Tensor):
                Original spatial-temporal grid dimensions before patching,
                    shape [3] (3 dimensions correspond to F_patches, H_patches, W_patches)

        Returns:
            List[Tensor]:
                Reconstructed video tensors with shape [C_out, F, H / 8, W / 8]
        """

        c = self.out_dim
        bs = x.shape[0]
        x = x.view(bs, *grid_sizes, *self.patch_size, c)
        x = torch.einsum("bfhwpqrc->bcfphqwr", x)
        x = x.reshape(bs, c, *[i * j for i, j in zip(grid_sizes, self.patch_size)])
        return x