add FLUX.1 LoRA training

library/flux_models.py

@@ -0,0 +1,920 @@
+# copy from FLUX repo: https://github.com/black-forest-labs/flux
+# license: Apache-2.0 License
+
+
+from dataclasses import dataclass
+import math
+
+import torch
+from einops import rearrange
+from torch import Tensor, nn
+from torch.utils.checkpoint import checkpoint
+
+# USE_REENTRANT = True
+
+
+@dataclass
+class FluxParams:
+    in_channels: int
+    vec_in_dim: int
+    context_in_dim: int
+    hidden_size: int
+    mlp_ratio: float
+    num_heads: int
+    depth: int
+    depth_single_blocks: int
+    axes_dim: list[int]
+    theta: int
+    qkv_bias: bool
+    guidance_embed: bool
+
+
+# region autoencoder
+
+
+@dataclass
+class AutoEncoderParams:
+    resolution: int
+    in_channels: int
+    ch: int
+    out_ch: int
+    ch_mult: list[int]
+    num_res_blocks: int
+    z_channels: int
+    scale_factor: float
+    shift_factor: float
+
+
+def swish(x: Tensor) -> Tensor:
+    return x * torch.sigmoid(x)
+
+
+class AttnBlock(nn.Module):
+    def __init__(self, in_channels: int):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)
+
+        self.q = nn.Conv2d(in_channels, in_channels, kernel_size=1)
+        self.k = nn.Conv2d(in_channels, in_channels, kernel_size=1)
+        self.v = nn.Conv2d(in_channels, in_channels, kernel_size=1)
+        self.proj_out = nn.Conv2d(in_channels, in_channels, kernel_size=1)
+
+    def attention(self, h_: Tensor) -> Tensor:
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        b, c, h, w = q.shape
+        q = rearrange(q, "b c h w -> b 1 (h w) c").contiguous()
+        k = rearrange(k, "b c h w -> b 1 (h w) c").contiguous()
+        v = rearrange(v, "b c h w -> b 1 (h w) c").contiguous()
+        h_ = nn.functional.scaled_dot_product_attention(q, k, v)
+
+        return rearrange(h_, "b 1 (h w) c -> b c h w", h=h, w=w, c=c, b=b)
+
+    def forward(self, x: Tensor) -> Tensor:
+        return x + self.proj_out(self.attention(x))
+
+
+class ResnetBlock(nn.Module):
+    def __init__(self, in_channels: int, out_channels: int):
+        super().__init__()
+        self.in_channels = in_channels
+        out_channels = in_channels if out_channels is None else out_channels
+        self.out_channels = out_channels
+
+        self.norm1 = nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)
+        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1, padding=1)
+        self.norm2 = nn.GroupNorm(num_groups=32, num_channels=out_channels, eps=1e-6, affine=True)
+        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding=1)
+        if self.in_channels != self.out_channels:
+            self.nin_shortcut = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0)
+
+    def forward(self, x):
+        h = x
+        h = self.norm1(h)
+        h = swish(h)
+        h = self.conv1(h)
+
+        h = self.norm2(h)
+        h = swish(h)
+        h = self.conv2(h)
+
+        if self.in_channels != self.out_channels:
+            x = self.nin_shortcut(x)
+
+        return x + h
+
+
+class Downsample(nn.Module):
+    def __init__(self, in_channels: int):
+        super().__init__()
+        # no asymmetric padding in torch conv, must do it ourselves
+        self.conv = nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=2, padding=0)
+
+    def forward(self, x: Tensor):
+        pad = (0, 1, 0, 1)
+        x = nn.functional.pad(x, pad, mode="constant", value=0)
+        x = self.conv(x)
+        return x
+
+
+class Upsample(nn.Module):
+    def __init__(self, in_channels: int):
+        super().__init__()
+        self.conv = nn.Conv2d(in_channels, in_channels, kernel_size=3, stride=1, padding=1)
+
+    def forward(self, x: Tensor):
+        x = nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
+        x = self.conv(x)
+        return x
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        resolution: int,
+        in_channels: int,
+        ch: int,
+        ch_mult: list[int],
+        num_res_blocks: int,
+        z_channels: int,
+    ):
+        super().__init__()
+        self.ch = ch
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+        # downsampling
+        self.conv_in = nn.Conv2d(in_channels, self.ch, kernel_size=3, stride=1, padding=1)
+
+        curr_res = resolution
+        in_ch_mult = (1,) + tuple(ch_mult)
+        self.in_ch_mult = in_ch_mult
+        self.down = nn.ModuleList()
+        block_in = self.ch
+        for i_level in range(self.num_resolutions):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_in = ch * in_ch_mult[i_level]
+            block_out = ch * ch_mult[i_level]
+            for _ in range(self.num_res_blocks):
+                block.append(ResnetBlock(in_channels=block_in, out_channels=block_out))
+                block_in = block_out
+            down = nn.Module()
+            down.block = block
+            down.attn = attn
+            if i_level != self.num_resolutions - 1:
+                down.downsample = Downsample(block_in)
+                curr_res = curr_res // 2
+            self.down.append(down)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(in_channels=block_in, out_channels=block_in)
+        self.mid.attn_1 = AttnBlock(block_in)
+        self.mid.block_2 = ResnetBlock(in_channels=block_in, out_channels=block_in)
+
+        # end
+        self.norm_out = nn.GroupNorm(num_groups=32, num_channels=block_in, eps=1e-6, affine=True)
+        self.conv_out = nn.Conv2d(block_in, 2 * z_channels, kernel_size=3, stride=1, padding=1)
+
+    def forward(self, x: Tensor) -> Tensor:
+        # downsampling
+        hs = [self.conv_in(x)]
+        for i_level in range(self.num_resolutions):
+            for i_block in range(self.num_res_blocks):
+                h = self.down[i_level].block[i_block](hs[-1])
+                if len(self.down[i_level].attn) > 0:
+                    h = self.down[i_level].attn[i_block](h)
+                hs.append(h)
+            if i_level != self.num_resolutions - 1:
+                hs.append(self.down[i_level].downsample(hs[-1]))
+
+        # middle
+        h = hs[-1]
+        h = self.mid.block_1(h)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h)
+        # end
+        h = self.norm_out(h)
+        h = swish(h)
+        h = self.conv_out(h)
+        return h
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        ch: int,
+        out_ch: int,
+        ch_mult: list[int],
+        num_res_blocks: int,
+        in_channels: int,
+        resolution: int,
+        z_channels: int,
+    ):
+        super().__init__()
+        self.ch = ch
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+        self.ffactor = 2 ** (self.num_resolutions - 1)
+
+        # compute in_ch_mult, block_in and curr_res at lowest res
+        block_in = ch * ch_mult[self.num_resolutions - 1]
+        curr_res = resolution // 2 ** (self.num_resolutions - 1)
+        self.z_shape = (1, z_channels, curr_res, curr_res)
+
+        # z to block_in
+        self.conv_in = nn.Conv2d(z_channels, block_in, kernel_size=3, stride=1, padding=1)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(in_channels=block_in, out_channels=block_in)
+        self.mid.attn_1 = AttnBlock(block_in)
+        self.mid.block_2 = ResnetBlock(in_channels=block_in, out_channels=block_in)
+
+        # upsampling
+        self.up = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch * ch_mult[i_level]
+            for _ in range(self.num_res_blocks + 1):
+                block.append(ResnetBlock(in_channels=block_in, out_channels=block_out))
+                block_in = block_out
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in)
+                curr_res = curr_res * 2
+            self.up.insert(0, up)  # prepend to get consistent order
+
+        # end
+        self.norm_out = nn.GroupNorm(num_groups=32, num_channels=block_in, eps=1e-6, affine=True)
+        self.conv_out = nn.Conv2d(block_in, out_ch, kernel_size=3, stride=1, padding=1)
+
+    def forward(self, z: Tensor) -> Tensor:
+        # z to block_in
+        h = self.conv_in(z)
+
+        # middle
+        h = self.mid.block_1(h)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks + 1):
+                h = self.up[i_level].block[i_block](h)
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h)
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        h = self.norm_out(h)
+        h = swish(h)
+        h = self.conv_out(h)
+        return h
+
+
+class DiagonalGaussian(nn.Module):
+    def __init__(self, sample: bool = True, chunk_dim: int = 1):
+        super().__init__()
+        self.sample = sample
+        self.chunk_dim = chunk_dim
+
+    def forward(self, z: Tensor) -> Tensor:
+        mean, logvar = torch.chunk(z, 2, dim=self.chunk_dim)
+        if self.sample:
+            std = torch.exp(0.5 * logvar)
+            return mean + std * torch.randn_like(mean)
+        else:
+            return mean
+
+
+class AutoEncoder(nn.Module):
+    def __init__(self, params: AutoEncoderParams):
+        super().__init__()
+        self.encoder = Encoder(
+            resolution=params.resolution,
+            in_channels=params.in_channels,
+            ch=params.ch,
+            ch_mult=params.ch_mult,
+            num_res_blocks=params.num_res_blocks,
+            z_channels=params.z_channels,
+        )
+        self.decoder = Decoder(
+            resolution=params.resolution,
+            in_channels=params.in_channels,
+            ch=params.ch,
+            out_ch=params.out_ch,
+            ch_mult=params.ch_mult,
+            num_res_blocks=params.num_res_blocks,
+            z_channels=params.z_channels,
+        )
+        self.reg = DiagonalGaussian()
+
+        self.scale_factor = params.scale_factor
+        self.shift_factor = params.shift_factor
+
+    @property
+    def device(self) -> torch.device:
+        return next(self.parameters()).device
+
+    @property
+    def dtype(self) -> torch.dtype:
+        return next(self.parameters()).dtype
+
+    def encode(self, x: Tensor) -> Tensor:
+        z = self.reg(self.encoder(x))
+        z = self.scale_factor * (z - self.shift_factor)
+        return z
+
+    def decode(self, z: Tensor) -> Tensor:
+        z = z / self.scale_factor + self.shift_factor
+        return self.decoder(z)
+
+    def forward(self, x: Tensor) -> Tensor:
+        return self.decode(self.encode(x))
+
+
+# endregion
+# region config
+
+
+@dataclass
+class ModelSpec:
+    params: FluxParams
+    ae_params: AutoEncoderParams
+    ckpt_path: str | None
+    ae_path: str | None
+    # repo_id: str | None
+    # repo_flow: str | None
+    # repo_ae: str | None
+
+
+configs = {
+    "dev": ModelSpec(
+        # repo_id="black-forest-labs/FLUX.1-dev",
+        # repo_flow="flux1-dev.sft",
+        # repo_ae="ae.sft",
+        ckpt_path=None,  # os.getenv("FLUX_DEV"),
+        params=FluxParams(
+            in_channels=64,
+            vec_in_dim=768,
+            context_in_dim=4096,
+            hidden_size=3072,
+            mlp_ratio=4.0,
+            num_heads=24,
+            depth=19,
+            depth_single_blocks=38,
+            axes_dim=[16, 56, 56],
+            theta=10_000,
+            qkv_bias=True,
+            guidance_embed=True,
+        ),
+        ae_path=None,  # os.getenv("AE"),
+        ae_params=AutoEncoderParams(
+            resolution=256,
+            in_channels=3,
+            ch=128,
+            out_ch=3,
+            ch_mult=[1, 2, 4, 4],
+            num_res_blocks=2,
+            z_channels=16,
+            scale_factor=0.3611,
+            shift_factor=0.1159,
+        ),
+    ),
+    "schnell": ModelSpec(
+        # repo_id="black-forest-labs/FLUX.1-schnell",
+        # repo_flow="flux1-schnell.sft",
+        # repo_ae="ae.sft",
+        ckpt_path=None,  # os.getenv("FLUX_SCHNELL"),
+        params=FluxParams(
+            in_channels=64,
+            vec_in_dim=768,
+            context_in_dim=4096,
+            hidden_size=3072,
+            mlp_ratio=4.0,
+            num_heads=24,
+            depth=19,
+            depth_single_blocks=38,
+            axes_dim=[16, 56, 56],
+            theta=10_000,
+            qkv_bias=True,
+            guidance_embed=False,
+        ),
+        ae_path=None,  # os.getenv("AE"),
+        ae_params=AutoEncoderParams(
+            resolution=256,
+            in_channels=3,
+            ch=128,
+            out_ch=3,
+            ch_mult=[1, 2, 4, 4],
+            num_res_blocks=2,
+            z_channels=16,
+            scale_factor=0.3611,
+            shift_factor=0.1159,
+        ),
+    ),
+}
+
+
+# endregion
+
+# region math
+
+
+def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor) -> Tensor:
+    q, k = apply_rope(q, k, pe)
+
+    x = torch.nn.functional.scaled_dot_product_attention(q, k, v)
+    x = rearrange(x, "B H L D -> B L (H D)")
+
+    return x
+
+
+def rope(pos: Tensor, dim: int, theta: int) -> Tensor:
+    assert dim % 2 == 0
+    scale = torch.arange(0, dim, 2, dtype=torch.float64, device=pos.device) / dim
+    omega = 1.0 / (theta**scale)
+    out = torch.einsum("...n,d->...nd", pos, omega)
+    out = torch.stack([torch.cos(out), -torch.sin(out), torch.sin(out), torch.cos(out)], dim=-1)
+    out = rearrange(out, "b n d (i j) -> b n d i j", i=2, j=2)
+    return out.float()
+
+
+def apply_rope(xq: Tensor, xk: Tensor, freqs_cis: Tensor) -> tuple[Tensor, Tensor]:
+    xq_ = xq.float().reshape(*xq.shape[:-1], -1, 1, 2)
+    xk_ = xk.float().reshape(*xk.shape[:-1], -1, 1, 2)
+    xq_out = freqs_cis[..., 0] * xq_[..., 0] + freqs_cis[..., 1] * xq_[..., 1]
+    xk_out = freqs_cis[..., 0] * xk_[..., 0] + freqs_cis[..., 1] * xk_[..., 1]
+    return xq_out.reshape(*xq.shape).type_as(xq), xk_out.reshape(*xk.shape).type_as(xk)
+
+
+# endregion
+
+
+# region layers
+class EmbedND(nn.Module):
+    def __init__(self, dim: int, theta: int, axes_dim: list[int]):
+        super().__init__()
+        self.dim = dim
+        self.theta = theta
+        self.axes_dim = axes_dim
+
+    def forward(self, ids: Tensor) -> Tensor:
+        n_axes = ids.shape[-1]
+        emb = torch.cat(
+            [rope(ids[..., i], self.axes_dim[i], self.theta) for i in range(n_axes)],
+            dim=-3,
+        )
+
+        return emb.unsqueeze(1)
+
+
+def timestep_embedding(t: Tensor, dim, max_period=10000, time_factor: float = 1000.0):
+    """
+    Create sinusoidal timestep embeddings.
+    :param t: a 1-D Tensor of N indices, one per batch element.
+                      These may be fractional.
+    :param dim: the dimension of the output.
+    :param max_period: controls the minimum frequency of the embeddings.
+    :return: an (N, D) Tensor of positional embeddings.
+    """
+    t = time_factor * t
+    half = dim // 2
+    freqs = torch.exp(-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half).to(t.device)
+
+    args = t[:, None].float() * freqs[None]
+    embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
+    if dim % 2:
+        embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
+    if torch.is_floating_point(t):
+        embedding = embedding.to(t)
+    return embedding
+
+
+class MLPEmbedder(nn.Module):
+    def __init__(self, in_dim: int, hidden_dim: int):
+        super().__init__()
+        self.in_layer = nn.Linear(in_dim, hidden_dim, bias=True)
+        self.silu = nn.SiLU()
+        self.out_layer = nn.Linear(hidden_dim, hidden_dim, bias=True)
+
+        self.gradient_checkpointing = False
+
+    def enable_gradient_checkpointing(self):
+        self.gradient_checkpointing = True
+
+    def disable_gradient_checkpointing(self):
+        self.gradient_checkpointing = False
+
+    def _forward(self, x: Tensor) -> Tensor:
+        return self.out_layer(self.silu(self.in_layer(x)))
+
+    def forward(self, *args, **kwargs):
+        if self.training and self.gradient_checkpointing:
+            return checkpoint(self._forward, *args, use_reentrant=False, **kwargs)
+        else:
+            return self._forward(*args, **kwargs)
+
+    # def forward(self, x):
+    #     if self.training and self.gradient_checkpointing:
+    #         def create_custom_forward(func):
+    #             def custom_forward(*inputs):
+    #                 return func(*inputs)
+    #             return custom_forward
+    #         return torch.utils.checkpoint.checkpoint(create_custom_forward(self._forward), x, use_reentrant=USE_REENTRANT)
+    #     else:
+    #         return self._forward(x)
+
+
+class RMSNorm(torch.nn.Module):
+    def __init__(self, dim: int):
+        super().__init__()
+        self.scale = nn.Parameter(torch.ones(dim))
+
+    def forward(self, x: Tensor):
+        x_dtype = x.dtype
+        x = x.float()
+        rrms = torch.rsqrt(torch.mean(x**2, dim=-1, keepdim=True) + 1e-6)
+        # return (x * rrms).to(dtype=x_dtype) * self.scale
+        return ((x * rrms) * self.scale.float()).to(dtype=x_dtype)
+
+
+class QKNorm(torch.nn.Module):
+    def __init__(self, dim: int):
+        super().__init__()
+        self.query_norm = RMSNorm(dim)
+        self.key_norm = RMSNorm(dim)
+
+    def forward(self, q: Tensor, k: Tensor, v: Tensor) -> tuple[Tensor, Tensor]:
+        q = self.query_norm(q)
+        k = self.key_norm(k)
+        return q.to(v), k.to(v)
+
+
+class SelfAttention(nn.Module):
+    def __init__(self, dim: int, num_heads: int = 8, qkv_bias: bool = False):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.norm = QKNorm(head_dim)
+        self.proj = nn.Linear(dim, dim)
+
+    #     self.gradient_checkpointing = False
+
+    # def enable_gradient_checkpointing(self):
+    #     self.gradient_checkpointing = True
+
+    def forward(self, x: Tensor, pe: Tensor) -> Tensor:
+        qkv = self.qkv(x)
+        q, k, v = rearrange(qkv, "B L (K H D) -> K B H L D", K=3, H=self.num_heads)
+        q, k = self.norm(q, k, v)
+        x = attention(q, k, v, pe=pe)
+        x = self.proj(x)
+        return x
+
+    # def forward(self, *args, **kwargs):
+    #     if self.training and self.gradient_checkpointing:
+    #         return checkpoint(self._forward, *args, use_reentrant=False, **kwargs)
+    #     else:
+    #         return self._forward(*args, **kwargs)
+
+
+@dataclass
+class ModulationOut:
+    shift: Tensor
+    scale: Tensor
+    gate: Tensor
+
+
+class Modulation(nn.Module):
+    def __init__(self, dim: int, double: bool):
+        super().__init__()
+        self.is_double = double
+        self.multiplier = 6 if double else 3
+        self.lin = nn.Linear(dim, self.multiplier * dim, bias=True)
+
+    def forward(self, vec: Tensor) -> tuple[ModulationOut, ModulationOut | None]:
+        out = self.lin(nn.functional.silu(vec))[:, None, :].chunk(self.multiplier, dim=-1)
+
+        return (
+            ModulationOut(*out[:3]),
+            ModulationOut(*out[3:]) if self.is_double else None,
+        )
+
+
+class DoubleStreamBlock(nn.Module):
+    def __init__(self, hidden_size: int, num_heads: int, mlp_ratio: float, qkv_bias: bool = False):
+        super().__init__()
+
+        mlp_hidden_dim = int(hidden_size * mlp_ratio)
+        self.num_heads = num_heads
+        self.hidden_size = hidden_size
+        self.img_mod = Modulation(hidden_size, double=True)
+        self.img_norm1 = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
+        self.img_attn = SelfAttention(dim=hidden_size, num_heads=num_heads, qkv_bias=qkv_bias)
+
+        self.img_norm2 = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
+        self.img_mlp = nn.Sequential(
+            nn.Linear(hidden_size, mlp_hidden_dim, bias=True),
+            nn.GELU(approximate="tanh"),
+            nn.Linear(mlp_hidden_dim, hidden_size, bias=True),
+        )
+
+        self.txt_mod = Modulation(hidden_size, double=True)
+        self.txt_norm1 = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
+        self.txt_attn = SelfAttention(dim=hidden_size, num_heads=num_heads, qkv_bias=qkv_bias)
+
+        self.txt_norm2 = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
+        self.txt_mlp = nn.Sequential(
+            nn.Linear(hidden_size, mlp_hidden_dim, bias=True),
+            nn.GELU(approximate="tanh"),
+            nn.Linear(mlp_hidden_dim, hidden_size, bias=True),
+        )
+
+        self.gradient_checkpointing = False
+
+    def enable_gradient_checkpointing(self):
+        self.gradient_checkpointing = True
+        # self.img_attn.enable_gradient_checkpointing()
+        # self.txt_attn.enable_gradient_checkpointing()
+
+    def disable_gradient_checkpointing(self):
+        self.gradient_checkpointing = False
+        # self.img_attn.disable_gradient_checkpointing()
+        # self.txt_attn.disable_gradient_checkpointing()
+
+    def _forward(self, img: Tensor, txt: Tensor, vec: Tensor, pe: Tensor) -> tuple[Tensor, Tensor]:
+        img_mod1, img_mod2 = self.img_mod(vec)
+        txt_mod1, txt_mod2 = self.txt_mod(vec)
+
+        # prepare image for attention
+        img_modulated = self.img_norm1(img)
+        img_modulated = (1 + img_mod1.scale) * img_modulated + img_mod1.shift
+        img_qkv = self.img_attn.qkv(img_modulated)
+        img_q, img_k, img_v = rearrange(img_qkv, "B L (K H D) -> K B H L D", K=3, H=self.num_heads)
+        img_q, img_k = self.img_attn.norm(img_q, img_k, img_v)
+
+        # prepare txt for attention
+        txt_modulated = self.txt_norm1(txt)
+        txt_modulated = (1 + txt_mod1.scale) * txt_modulated + txt_mod1.shift
+        txt_qkv = self.txt_attn.qkv(txt_modulated)
+        txt_q, txt_k, txt_v = rearrange(txt_qkv, "B L (K H D) -> K B H L D", K=3, H=self.num_heads)
+        txt_q, txt_k = self.txt_attn.norm(txt_q, txt_k, txt_v)
+
+        # run actual attention
+        q = torch.cat((txt_q, img_q), dim=2)
+        k = torch.cat((txt_k, img_k), dim=2)
+        v = torch.cat((txt_v, img_v), dim=2)
+
+        attn = attention(q, k, v, pe=pe)
+        txt_attn, img_attn = attn[:, : txt.shape[1]], attn[:, txt.shape[1] :]
+
+        # calculate the img bloks
+        img = img + img_mod1.gate * self.img_attn.proj(img_attn)
+        img = img + img_mod2.gate * self.img_mlp((1 + img_mod2.scale) * self.img_norm2(img) + img_mod2.shift)
+
+        # calculate the txt bloks
+        txt = txt + txt_mod1.gate * self.txt_attn.proj(txt_attn)
+        txt = txt + txt_mod2.gate * self.txt_mlp((1 + txt_mod2.scale) * self.txt_norm2(txt) + txt_mod2.shift)
+        return img, txt
+
+    def forward(self, *args, **kwargs):
+        if self.training and self.gradient_checkpointing:
+            return checkpoint(self._forward, *args, use_reentrant=False, **kwargs)
+        else:
+            return self._forward(*args, **kwargs)
+
+    # def forward(self, img: Tensor, txt: Tensor, vec: Tensor, pe: Tensor):
+    #     if self.training and self.gradient_checkpointing:
+    #         def create_custom_forward(func):
+    #             def custom_forward(*inputs):
+    #                 return func(*inputs)
+    #             return custom_forward
+    #         return torch.utils.checkpoint.checkpoint(
+    #             create_custom_forward(self._forward), img, txt, vec, pe, use_reentrant=USE_REENTRANT
+    #         )
+    #     else:
+    #         return self._forward(img, txt, vec, pe)
+
+
+class SingleStreamBlock(nn.Module):
+    """
+    A DiT block with parallel linear layers as described in
+    https://arxiv.org/abs/2302.05442 and adapted modulation interface.
+    """
+
+    def __init__(
+        self,
+        hidden_size: int,
+        num_heads: int,
+        mlp_ratio: float = 4.0,
+        qk_scale: float | None = None,
+    ):
+        super().__init__()
+        self.hidden_dim = hidden_size
+        self.num_heads = num_heads
+        head_dim = hidden_size // num_heads
+        self.scale = qk_scale or head_dim**-0.5
+
+        self.mlp_hidden_dim = int(hidden_size * mlp_ratio)
+        # qkv and mlp_in
+        self.linear1 = nn.Linear(hidden_size, hidden_size * 3 + self.mlp_hidden_dim)
+        # proj and mlp_out
+        self.linear2 = nn.Linear(hidden_size + self.mlp_hidden_dim, hidden_size)
+
+        self.norm = QKNorm(head_dim)
+
+        self.hidden_size = hidden_size
+        self.pre_norm = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
+
+        self.mlp_act = nn.GELU(approximate="tanh")
+        self.modulation = Modulation(hidden_size, double=False)
+
+        self.gradient_checkpointing = False
+
+    def enable_gradient_checkpointing(self):
+        self.gradient_checkpointing = True
+
+    def disable_gradient_checkpointing(self):
+        self.gradient_checkpointing = False
+
+    def _forward(self, x: Tensor, vec: Tensor, pe: Tensor) -> Tensor:
+        mod, _ = self.modulation(vec)
+        x_mod = (1 + mod.scale) * self.pre_norm(x) + mod.shift
+        qkv, mlp = torch.split(self.linear1(x_mod), [3 * self.hidden_size, self.mlp_hidden_dim], dim=-1)
+
+        q, k, v = rearrange(qkv, "B L (K H D) -> K B H L D", K=3, H=self.num_heads)
+        q, k = self.norm(q, k, v)
+
+        # compute attention
+        attn = attention(q, k, v, pe=pe)
+        # compute activation in mlp stream, cat again and run second linear layer
+        output = self.linear2(torch.cat((attn, self.mlp_act(mlp)), 2))
+        return x + mod.gate * output
+
+    def forward(self, *args, **kwargs):
+        if self.training and self.gradient_checkpointing:
+            return checkpoint(self._forward, *args, use_reentrant=False, **kwargs)
+        else:
+            return self._forward(*args, **kwargs)
+
+    # def forward(self, x: Tensor, vec: Tensor, pe: Tensor):
+    #     if self.training and self.gradient_checkpointing:
+    #         def create_custom_forward(func):
+    #             def custom_forward(*inputs):
+    #                 return func(*inputs)
+    #             return custom_forward
+    #         return torch.utils.checkpoint.checkpoint(create_custom_forward(self._forward), x, vec, pe, use_reentrant=USE_REENTRANT)
+    #     else:
+    #         return self._forward(x, vec, pe)
+
+
+class LastLayer(nn.Module):
+    def __init__(self, hidden_size: int, patch_size: int, out_channels: int):
+        super().__init__()
+        self.norm_final = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
+        self.linear = nn.Linear(hidden_size, patch_size * patch_size * out_channels, bias=True)
+        self.adaLN_modulation = nn.Sequential(nn.SiLU(), nn.Linear(hidden_size, 2 * hidden_size, bias=True))
+
+    def forward(self, x: Tensor, vec: Tensor) -> Tensor:
+        shift, scale = self.adaLN_modulation(vec).chunk(2, dim=1)
+        x = (1 + scale[:, None, :]) * self.norm_final(x) + shift[:, None, :]
+        x = self.linear(x)
+        return x
+
+
+# endregion
+
+
+class Flux(nn.Module):
+    """
+    Transformer model for flow matching on sequences.
+    """
+
+    def __init__(self, params: FluxParams):
+        super().__init__()
+
+        self.params = params
+        self.in_channels = params.in_channels
+        self.out_channels = self.in_channels
+        if params.hidden_size % params.num_heads != 0:
+            raise ValueError(f"Hidden size {params.hidden_size} must be divisible by num_heads {params.num_heads}")
+        pe_dim = params.hidden_size // params.num_heads
+        if sum(params.axes_dim) != pe_dim:
+            raise ValueError(f"Got {params.axes_dim} but expected positional dim {pe_dim}")
+        self.hidden_size = params.hidden_size
+        self.num_heads = params.num_heads
+        self.pe_embedder = EmbedND(dim=pe_dim, theta=params.theta, axes_dim=params.axes_dim)
+        self.img_in = nn.Linear(self.in_channels, self.hidden_size, bias=True)
+        self.time_in = MLPEmbedder(in_dim=256, hidden_dim=self.hidden_size)
+        self.vector_in = MLPEmbedder(params.vec_in_dim, self.hidden_size)
+        self.guidance_in = MLPEmbedder(in_dim=256, hidden_dim=self.hidden_size) if params.guidance_embed else nn.Identity()
+        self.txt_in = nn.Linear(params.context_in_dim, self.hidden_size)
+
+        self.double_blocks = nn.ModuleList(
+            [
+                DoubleStreamBlock(
+                    self.hidden_size,
+                    self.num_heads,
+                    mlp_ratio=params.mlp_ratio,
+                    qkv_bias=params.qkv_bias,
+                )
+                for _ in range(params.depth)
+            ]
+        )
+
+        self.single_blocks = nn.ModuleList(
+            [
+                SingleStreamBlock(self.hidden_size, self.num_heads, mlp_ratio=params.mlp_ratio)
+                for _ in range(params.depth_single_blocks)
+            ]
+        )
+
+        self.final_layer = LastLayer(self.hidden_size, 1, self.out_channels)
+
+        self.gradient_checkpointing = False
+
+    @property
+    def device(self):
+        return next(self.parameters()).device
+
+    @property
+    def dtype(self):
+        return next(self.parameters()).dtype
+
+    def enable_gradient_checkpointing(self):
+        self.gradient_checkpointing = True
+
+        self.time_in.enable_gradient_checkpointing()
+        self.vector_in.enable_gradient_checkpointing()
+        self.guidance_in.enable_gradient_checkpointing()
+
+        for block in self.double_blocks + self.single_blocks:
+            block.enable_gradient_checkpointing()
+
+        print("FLUX: Gradient checkpointing enabled.")
+
+    def disable_gradient_checkpointing(self):
+        self.gradient_checkpointing = False
+
+        self.time_in.disable_gradient_checkpointing()
+        self.vector_in.disable_gradient_checkpointing()
+        self.guidance_in.disable_gradient_checkpointing()
+
+        for block in self.double_blocks + self.single_blocks:
+            block.disable_gradient_checkpointing()
+
+        print("FLUX: Gradient checkpointing disabled.")
+
+    def forward(
+        self,
+        img: Tensor,
+        img_ids: Tensor,
+        txt: Tensor,
+        txt_ids: Tensor,
+        timesteps: Tensor,
+        y: Tensor,
+        guidance: Tensor | None = None,
+    ) -> Tensor:
+        if img.ndim != 3 or txt.ndim != 3:
+            raise ValueError("Input img and txt tensors must have 3 dimensions.")
+
+        # running on sequences img
+        img = self.img_in(img)
+        vec = self.time_in(timestep_embedding(timesteps, 256))
+        if self.params.guidance_embed:
+            if guidance is None:
+                raise ValueError("Didn't get guidance strength for guidance distilled model.")
+            vec = vec + self.guidance_in(timestep_embedding(guidance, 256))
+        vec = vec + self.vector_in(y)
+        txt = self.txt_in(txt)
+
+        ids = torch.cat((txt_ids, img_ids), dim=1)
+        pe = self.pe_embedder(ids)
+
+        for block in self.double_blocks:
+            img, txt = block(img=img, txt=txt, vec=vec, pe=pe)
+
+        img = torch.cat((txt, img), 1)
+        for block in self.single_blocks:
+            img = block(img, vec=vec, pe=pe)
+        img = img[:, txt.shape[1] :, ...]
+
+        img = self.final_layer(img, vec)  # (N, T, patch_size ** 2 * out_channels)
+        return img

library/flux_utils.py

@@ -0,0 +1,215 @@
+import json
+from typing import Union
+import einops
+import torch
+
+from safetensors.torch import load_file
+from accelerate import init_empty_weights
+from transformers import CLIPTextModel, CLIPConfig, T5EncoderModel, T5Config
+
+from library import flux_models
+
+from library.utils import setup_logging
+
+setup_logging()
+import logging
+
+logger = logging.getLogger(__name__)
+
+MODEL_VERSION_FLUX_V1 = "flux1"
+
+
+def load_flow_model(name: str, ckpt_path: str, dtype: torch.dtype, device: Union[str, torch.device]) -> flux_models.Flux:
+    logger.info(f"Bulding Flux model {name}")
+    with torch.device("meta"):
+        model = flux_models.Flux(flux_models.configs[name].params).to(dtype)
+
+    # load_sft doesn't support torch.device
+    logger.info(f"Loading state dict from {ckpt_path}")
+    sd = load_file(ckpt_path, device=str(device))
+    info = model.load_state_dict(sd, strict=False, assign=True)
+    logger.info(f"Loaded Flux: {info}")
+    return model
+
+
+def load_ae(name: str, ckpt_path: str, dtype: torch.dtype, device: Union[str, torch.device]) -> flux_models.AutoEncoder:
+    logger.info("Building AutoEncoder")
+    with torch.device("meta"):
+        ae = flux_models.AutoEncoder(flux_models.configs[name].ae_params).to(dtype)
+
+    logger.info(f"Loading state dict from {ckpt_path}")
+    sd = load_file(ckpt_path, device=str(device))
+    info = ae.load_state_dict(sd, strict=False, assign=True)
+    logger.info(f"Loaded AE: {info}")
+    return ae
+
+
+def load_clip_l(ckpt_path: str, dtype: torch.dtype, device: Union[str, torch.device]) -> CLIPTextModel:
+    logger.info("Building CLIP")
+    CLIPL_CONFIG = {
+        "_name_or_path": "clip-vit-large-patch14/",
+        "architectures": ["CLIPModel"],
+        "initializer_factor": 1.0,
+        "logit_scale_init_value": 2.6592,
+        "model_type": "clip",
+        "projection_dim": 768,
+        # "text_config": {
+        "_name_or_path": "",
+        "add_cross_attention": False,
+        "architectures": None,
+        "attention_dropout": 0.0,
+        "bad_words_ids": None,
+        "bos_token_id": 0,
+        "chunk_size_feed_forward": 0,
+        "cross_attention_hidden_size": None,
+        "decoder_start_token_id": None,
+        "diversity_penalty": 0.0,
+        "do_sample": False,
+        "dropout": 0.0,
+        "early_stopping": False,
+        "encoder_no_repeat_ngram_size": 0,
+        "eos_token_id": 2,
+        "finetuning_task": None,
+        "forced_bos_token_id": None,
+        "forced_eos_token_id": None,
+        "hidden_act": "quick_gelu",
+        "hidden_size": 768,
+        "id2label": {"0": "LABEL_0", "1": "LABEL_1"},
+        "initializer_factor": 1.0,
+        "initializer_range": 0.02,
+        "intermediate_size": 3072,
+        "is_decoder": False,
+        "is_encoder_decoder": False,
+        "label2id": {"LABEL_0": 0, "LABEL_1": 1},
+        "layer_norm_eps": 1e-05,
+        "length_penalty": 1.0,
+        "max_length": 20,
+        "max_position_embeddings": 77,
+        "min_length": 0,
+        "model_type": "clip_text_model",
+        "no_repeat_ngram_size": 0,
+        "num_attention_heads": 12,
+        "num_beam_groups": 1,
+        "num_beams": 1,
+        "num_hidden_layers": 12,
+        "num_return_sequences": 1,
+        "output_attentions": False,
+        "output_hidden_states": False,
+        "output_scores": False,
+        "pad_token_id": 1,
+        "prefix": None,
+        "problem_type": None,
+        "projection_dim": 768,
+        "pruned_heads": {},
+        "remove_invalid_values": False,
+        "repetition_penalty": 1.0,
+        "return_dict": True,
+        "return_dict_in_generate": False,
+        "sep_token_id": None,
+        "task_specific_params": None,
+        "temperature": 1.0,
+        "tie_encoder_decoder": False,
+        "tie_word_embeddings": True,
+        "tokenizer_class": None,
+        "top_k": 50,
+        "top_p": 1.0,
+        "torch_dtype": None,
+        "torchscript": False,
+        "transformers_version": "4.16.0.dev0",
+        "use_bfloat16": False,
+        "vocab_size": 49408,
+        "hidden_act": "gelu",
+        "hidden_size": 1280,
+        "intermediate_size": 5120,
+        "num_attention_heads": 20,
+        "num_hidden_layers": 32,
+        # },
+        # "text_config_dict": {
+        "hidden_size": 768,
+        "intermediate_size": 3072,
+        "num_attention_heads": 12,
+        "num_hidden_layers": 12,
+        "projection_dim": 768,
+        # },
+        # "torch_dtype": "float32",
+        # "transformers_version": None,
+    }
+    config = CLIPConfig(**CLIPL_CONFIG)
+    with init_empty_weights():
+        clip = CLIPTextModel._from_config(config)
+
+    logger.info(f"Loading state dict from {ckpt_path}")
+    sd = load_file(ckpt_path, device=str(device))
+    info = clip.load_state_dict(sd, strict=False, assign=True)
+    logger.info(f"Loaded CLIP: {info}")
+    return clip
+
+
+def load_t5xxl(ckpt_path: str, dtype: torch.dtype, device: Union[str, torch.device]) -> T5EncoderModel:
+    T5_CONFIG_JSON = """
+{
+  "architectures": [
+    "T5EncoderModel"
+  ],
+  "classifier_dropout": 0.0,
+  "d_ff": 10240,
+  "d_kv": 64,
+  "d_model": 4096,
+  "decoder_start_token_id": 0,
+  "dense_act_fn": "gelu_new",
+  "dropout_rate": 0.1,
+  "eos_token_id": 1,
+  "feed_forward_proj": "gated-gelu",
+  "initializer_factor": 1.0,
+  "is_encoder_decoder": true,
+  "is_gated_act": true,
+  "layer_norm_epsilon": 1e-06,
+  "model_type": "t5",
+  "num_decoder_layers": 24,
+  "num_heads": 64,
+  "num_layers": 24,
+  "output_past": true,
+  "pad_token_id": 0,
+  "relative_attention_max_distance": 128,
+  "relative_attention_num_buckets": 32,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.41.2",
+  "use_cache": true,
+  "vocab_size": 32128
+}
+"""
+    config = json.loads(T5_CONFIG_JSON)
+    config = T5Config(**config)
+    with init_empty_weights():
+        t5xxl = T5EncoderModel._from_config(config)
+
+    logger.info(f"Loading state dict from {ckpt_path}")
+    sd = load_file(ckpt_path, device=str(device))
+    info = t5xxl.load_state_dict(sd, strict=False, assign=True)
+    logger.info(f"Loaded T5xxl: {info}")
+    return t5xxl
+
+
+def prepare_img_ids(batch_size: int, packed_latent_height: int, packed_latent_width: int):
+    img_ids = torch.zeros(packed_latent_height, packed_latent_width, 3)
+    img_ids[..., 1] = img_ids[..., 1] + torch.arange(packed_latent_height)[:, None]
+    img_ids[..., 2] = img_ids[..., 2] + torch.arange(packed_latent_width)[None, :]
+    img_ids = einops.repeat(img_ids, "h w c -> b (h w) c", b=batch_size)
+    return img_ids
+
+
+def unpack_latents(x: torch.Tensor, packed_latent_height: int, packed_latent_width: int) -> torch.Tensor:
+    """
+    x: [b (h w) (c ph pw)] -> [b c (h ph) (w pw)], ph=2, pw=2
+    """
+    x = einops.rearrange(x, "b (h w) (c ph pw) -> b c (h ph) (w pw)", h=packed_latent_height, w=packed_latent_width, ph=2, pw=2)
+    return x
+
+
+def pack_latents(x: torch.Tensor) -> torch.Tensor:
+    """
+    x: [b c (h ph) (w pw)] -> [b (h w) (c ph pw)], ph=2, pw=2
+    """
+    x = einops.rearrange(x, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=2, pw=2)
+    return x

library/sd3_models.py

@@ -15,6 +15,12 @@ import torch.nn as nn
 import torch.nn.functional as F
 from torch.utils.checkpoint import checkpoint
 from transformers import CLIPTokenizer, T5TokenizerFast
+from .utils import setup_logging
+
+setup_logging()
+import logging
+
+logger = logging.getLogger(__name__)
 
 
 memory_efficient_attention = None
@@ -95,7 +101,9 @@ class SDTokenizer:
             batch.extend([(pad_token, 1.0)] * (self.min_length - len(batch)))
 
         # truncate to max_length
-        print(f"batch: {batch}, max_length: {self.max_length}, truncate: {truncate_to_max_length}, truncate_length: {truncate_length}")
+        print(
+            f"batch: {batch}, max_length: {self.max_length}, truncate: {truncate_to_max_length}, truncate_length: {truncate_length}"
+        )
         if truncate_to_max_length and len(batch) > self.max_length:
             batch = batch[: self.max_length]
         if truncate_length is not None and len(batch) > truncate_length:
@@ -1554,6 +1562,17 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
             self.set_clip_options({"layer": layer_idx})
         self.options_default = (self.layer, self.layer_idx, self.return_projected_pooled)
 
+    @property
+    def device(self):
+        return next(self.parameters()).device
+
+    @property
+    def dtype(self):
+        return next(self.parameters()).dtype
+
+    def gradient_checkpointing_enable(self):
+        logger.warning("Gradient checkpointing is not supported for this model")
+
     def set_attn_mode(self, mode):
         raise NotImplementedError("This model does not support setting the attention mode")
 
@@ -1925,6 +1944,7 @@ def create_clip_l(device="cpu", dtype=torch.float32, state_dict: Optional[Dict[s
             return_projected_pooled=False,
             textmodel_json_config=CLIPL_CONFIG,
         )
+        clip_l.gradient_checkpointing_enable()
     if state_dict is not None:
         # update state_dict if provided to include logit_scale and text_projection.weight avoid errors
         if "logit_scale" not in state_dict:

library/strategy_flux.py

@@ -0,0 +1,244 @@
+import os
+import glob
+from typing import Any, List, Optional, Tuple, Union
+import torch
+import numpy as np
+from transformers import CLIPTokenizer, T5TokenizerFast
+
+from library import sd3_utils, train_util
+from library import sd3_models
+from library.strategy_base import LatentsCachingStrategy, TextEncodingStrategy, TokenizeStrategy, TextEncoderOutputsCachingStrategy
+
+from library.utils import setup_logging
+
+setup_logging()
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+CLIP_L_TOKENIZER_ID = "openai/clip-vit-large-patch14"
+T5_XXL_TOKENIZER_ID = "google/t5-v1_1-xxl"
+
+
+class FluxTokenizeStrategy(TokenizeStrategy):
+    def __init__(self, t5xxl_max_length: int = 256, tokenizer_cache_dir: Optional[str] = None) -> None:
+        self.t5xxl_max_length = t5xxl_max_length
+        self.clip_l = self._load_tokenizer(CLIPTokenizer, CLIP_L_TOKENIZER_ID, tokenizer_cache_dir=tokenizer_cache_dir)
+        self.t5xxl = self._load_tokenizer(T5TokenizerFast, T5_XXL_TOKENIZER_ID, tokenizer_cache_dir=tokenizer_cache_dir)
+
+    def tokenize(self, text: Union[str, List[str]]) -> List[torch.Tensor]:
+        text = [text] if isinstance(text, str) else text
+
+        l_tokens = self.clip_l(text, max_length=77, padding="max_length", truncation=True, return_tensors="pt")
+        t5_tokens = self.t5xxl(text, max_length=self.t5xxl_max_length, padding="max_length", truncation=True, return_tensors="pt")
+
+        t5_attn_mask = t5_tokens["attention_mask"]
+        l_tokens = l_tokens["input_ids"]
+        t5_tokens = t5_tokens["input_ids"]
+
+        return [l_tokens, t5_tokens, t5_attn_mask]
+
+
+class FluxTextEncodingStrategy(TextEncodingStrategy):
+    def __init__(self) -> None:
+        pass
+
+    def encode_tokens(
+        self,
+        tokenize_strategy: TokenizeStrategy,
+        models: List[Any],
+        tokens: List[torch.Tensor],
+        apply_t5_attn_mask: bool = False,
+    ) -> List[torch.Tensor]:
+        # supports single model inference only
+
+        clip_l, t5xxl = models
+        l_tokens, t5_tokens = tokens[:2]
+        t5_attn_mask = tokens[2] if len(tokens) > 2 else None
+
+        if clip_l is not None and l_tokens is not None:
+            l_pooled = clip_l(l_tokens.to(clip_l.device))["pooler_output"]
+        else:
+            l_pooled = None
+
+        if t5xxl is not None and t5_tokens is not None:
+            # t5_out is [1, max length, 4096]
+            t5_out, _ = t5xxl(t5_tokens.to(t5xxl.device), return_dict=False, output_hidden_states=True)
+            if apply_t5_attn_mask:
+                t5_out = t5_out * t5_attn_mask.to(t5_out.device).unsqueeze(-1)
+            txt_ids = torch.zeros(1, t5_out.shape[1], 3, device=t5_out.device)
+        else:
+            t5_out = None
+            txt_ids = None
+
+        return [l_pooled, t5_out, txt_ids]
+
+
+class FluxTextEncoderOutputsCachingStrategy(TextEncoderOutputsCachingStrategy):
+    FLUX_TEXT_ENCODER_OUTPUTS_NPZ_SUFFIX = "_flux_te.npz"
+
+    def __init__(
+        self,
+        cache_to_disk: bool,
+        batch_size: int,
+        skip_disk_cache_validity_check: bool,
+        is_partial: bool = False,
+        apply_t5_attn_mask: bool = False,
+    ) -> None:
+        super().__init__(cache_to_disk, batch_size, skip_disk_cache_validity_check, is_partial)
+        self.apply_t5_attn_mask = apply_t5_attn_mask
+
+    def get_outputs_npz_path(self, image_abs_path: str) -> str:
+        return os.path.splitext(image_abs_path)[0] + FluxTextEncoderOutputsCachingStrategy.FLUX_TEXT_ENCODER_OUTPUTS_NPZ_SUFFIX
+
+    def is_disk_cached_outputs_expected(self, npz_path: str):
+        if not self.cache_to_disk:
+            return False
+        if not os.path.exists(npz_path):
+            return False
+        if self.skip_disk_cache_validity_check:
+            return True
+
+        try:
+            npz = np.load(npz_path)
+            if "l_pooled" not in npz:
+                return False
+            if "t5_out" not in npz:
+                return False
+            if "txt_ids" not in npz:
+                return False
+        except Exception as e:
+            logger.error(f"Error loading file: {npz_path}")
+            raise e
+
+        return True
+
+    def mask_t5_attn(self, t5_out: np.ndarray, t5_attn_mask: np.ndarray) -> np.ndarray:
+        return t5_out * np.expand_dims(t5_attn_mask, -1)
+
+    def load_outputs_npz(self, npz_path: str) -> List[np.ndarray]:
+        data = np.load(npz_path)
+        l_pooled = data["l_pooled"]
+        t5_out = data["t5_out"]
+        txt_ids = data["txt_ids"]
+
+        if self.apply_t5_attn_mask:
+            t5_attn_mask = data["t5_attn_mask"]
+            t5_out = self.mask_t5_attn(t5_out, t5_attn_mask)
+
+        return [l_pooled, t5_out, txt_ids]
+
+    def cache_batch_outputs(
+        self, tokenize_strategy: TokenizeStrategy, models: List[Any], text_encoding_strategy: TextEncodingStrategy, infos: List
+    ):
+        flux_text_encoding_strategy: FluxTextEncodingStrategy = text_encoding_strategy
+        captions = [info.caption for info in infos]
+
+        tokens_and_masks = tokenize_strategy.tokenize(captions)
+        with torch.no_grad():
+            l_pooled, t5_out, txt_ids = flux_text_encoding_strategy.encode_tokens(
+                tokenize_strategy, models, tokens_and_masks, self.apply_t5_attn_mask
+            )
+
+        if l_pooled.dtype == torch.bfloat16:
+            l_pooled = l_pooled.float()
+        if t5_out.dtype == torch.bfloat16:
+            t5_out = t5_out.float()
+        if txt_ids.dtype == torch.bfloat16:
+            txt_ids = txt_ids.float()
+
+        l_pooled = l_pooled.cpu().numpy()
+        t5_out = t5_out.cpu().numpy()
+        txt_ids = txt_ids.cpu().numpy()
+
+        for i, info in enumerate(infos):
+            l_pooled_i = l_pooled[i]
+            t5_out_i = t5_out[i]
+            txt_ids_i = txt_ids[i]
+
+            if self.cache_to_disk:
+                t5_attn_mask = tokens_and_masks[2]
+                t5_attn_mask_i = t5_attn_mask[i].cpu().numpy()
+                np.savez(
+                    info.text_encoder_outputs_npz,
+                    l_pooled=l_pooled_i,
+                    t5_out=t5_out_i,
+                    txt_ids=txt_ids_i,
+                    t5_attn_mask=t5_attn_mask_i,
+                )
+            else:
+                info.text_encoder_outputs = (l_pooled_i, t5_out_i, txt_ids_i)
+
+
+class FluxLatentsCachingStrategy(LatentsCachingStrategy):
+    FLUX_LATENTS_NPZ_SUFFIX = "_flux.npz"
+
+    def __init__(self, cache_to_disk: bool, batch_size: int, skip_disk_cache_validity_check: bool) -> None:
+        super().__init__(cache_to_disk, batch_size, skip_disk_cache_validity_check)
+
+    def get_image_size_from_disk_cache_path(self, absolute_path: str) -> Tuple[Optional[int], Optional[int]]:
+        npz_file = glob.glob(os.path.splitext(absolute_path)[0] + "_*" + FluxLatentsCachingStrategy.FLUX_LATENTS_NPZ_SUFFIX)
+        if len(npz_file) == 0:
+            return None, None
+        w, h = os.path.splitext(npz_file[0])[0].split("_")[-2].split("x")
+        return int(w), int(h)
+
+    def get_latents_npz_path(self, absolute_path: str, image_size: Tuple[int, int]) -> str:
+        return (
+            os.path.splitext(absolute_path)[0]
+            + f"_{image_size[0]:04d}x{image_size[1]:04d}"
+            + FluxLatentsCachingStrategy.FLUX_LATENTS_NPZ_SUFFIX
+        )
+
+    def is_disk_cached_latents_expected(self, bucket_reso: Tuple[int, int], npz_path: str, flip_aug: bool, alpha_mask: bool):
+        return self._default_is_disk_cached_latents_expected(8, bucket_reso, npz_path, flip_aug, alpha_mask)
+
+    # TODO remove circular dependency for ImageInfo
+    def cache_batch_latents(self, vae, image_infos: List, flip_aug: bool, alpha_mask: bool, random_crop: bool):
+        encode_by_vae = lambda img_tensor: vae.encode(img_tensor).to("cpu")
+        vae_device = vae.device
+        vae_dtype = vae.dtype
+
+        self._default_cache_batch_latents(encode_by_vae, vae_device, vae_dtype, image_infos, flip_aug, alpha_mask, random_crop)
+
+        if not train_util.HIGH_VRAM:
+            train_util.clean_memory_on_device(vae.device)
+
+
+if __name__ == "__main__":
+    # test code for FluxTokenizeStrategy
+    # tokenizer = sd3_models.SD3Tokenizer()
+    strategy = FluxTokenizeStrategy(256)
+    text = "hello world"
+
+    l_tokens, g_tokens, t5_tokens = strategy.tokenize(text)
+    # print(l_tokens.shape)
+    print(l_tokens)
+    print(g_tokens)
+    print(t5_tokens)
+
+    texts = ["hello world", "the quick brown fox jumps over the lazy dog"]
+    l_tokens_2 = strategy.clip_l(texts, max_length=77, padding="max_length", truncation=True, return_tensors="pt")
+    g_tokens_2 = strategy.clip_g(texts, max_length=77, padding="max_length", truncation=True, return_tensors="pt")
+    t5_tokens_2 = strategy.t5xxl(
+        texts, max_length=strategy.t5xxl_max_length, padding="max_length", truncation=True, return_tensors="pt"
+    )
+    print(l_tokens_2)
+    print(g_tokens_2)
+    print(t5_tokens_2)
+
+    # compare
+    print(torch.allclose(l_tokens, l_tokens_2["input_ids"][0]))
+    print(torch.allclose(g_tokens, g_tokens_2["input_ids"][0]))
+    print(torch.allclose(t5_tokens, t5_tokens_2["input_ids"][0]))
+
+    text = ",".join(["hello world! this is long text"] * 50)
+    l_tokens, g_tokens, t5_tokens = strategy.tokenize(text)
+    print(l_tokens)
+    print(g_tokens)
+    print(t5_tokens)
+
+    print(f"model max length l: {strategy.clip_l.model_max_length}")
+    print(f"model max length g: {strategy.clip_g.model_max_length}")
+    print(f"model max length t5: {strategy.t5xxl.model_max_length}")