diffusers 0.33.1__py3-none-any.whl → 0.35.0__py3-none-any.whl

diffusers/models/attention_processor.py

@@ -1,4 +1,4 @@
-# Copyright 2024 The HuggingFace Team. All rights reserved.
+# Copyright 2025 The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -203,8 +203,8 @@ class Attention(nn.Module):
             self.norm_q = nn.LayerNorm(dim_head * heads, eps=eps)
             self.norm_k = nn.LayerNorm(dim_head * kv_heads, eps=eps)
         elif qk_norm == "rms_norm":
-            self.norm_q = RMSNorm(dim_head, eps=eps)
-            self.norm_k = RMSNorm(dim_head, eps=eps)
+            self.norm_q = RMSNorm(dim_head, eps=eps, elementwise_affine=elementwise_affine)
+            self.norm_k = RMSNorm(dim_head, eps=eps, elementwise_affine=elementwise_affine)
         elif qk_norm == "rms_norm_across_heads":
             # LTX applies qk norm across all heads
             self.norm_q = RMSNorm(dim_head * heads, eps=eps)
@@ -2272,554 +2272,6 @@ class FusedAuraFlowAttnProcessor2_0:
             return hidden_states
 
 
-class FluxAttnProcessor2_0:
-    """Attention processor used typically in processing the SD3-like self-attention projections."""
-
-    def __init__(self):
-        if not hasattr(F, "scaled_dot_product_attention"):
-            raise ImportError("FluxAttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
-
-    def __call__(
-        self,
-        attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: torch.FloatTensor = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        image_rotary_emb: Optional[torch.Tensor] = None,
-    ) -> torch.FloatTensor:
-        batch_size, _, _ = hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
-
-        # `sample` projections.
-        query = attn.to_q(hidden_states)
-        key = attn.to_k(hidden_states)
-        value = attn.to_v(hidden_states)
-
-        inner_dim = key.shape[-1]
-        head_dim = inner_dim // attn.heads
-
-        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-        if attn.norm_q is not None:
-            query = attn.norm_q(query)
-        if attn.norm_k is not None:
-            key = attn.norm_k(key)
-
-        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
-        if encoder_hidden_states is not None:
-            # `context` projections.
-            encoder_hidden_states_query_proj = attn.add_q_proj(encoder_hidden_states)
-            encoder_hidden_states_key_proj = attn.add_k_proj(encoder_hidden_states)
-            encoder_hidden_states_value_proj = attn.add_v_proj(encoder_hidden_states)
-
-            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-
-            if attn.norm_added_q is not None:
-                encoder_hidden_states_query_proj = attn.norm_added_q(encoder_hidden_states_query_proj)
-            if attn.norm_added_k is not None:
-                encoder_hidden_states_key_proj = attn.norm_added_k(encoder_hidden_states_key_proj)
-
-            # attention
-            query = torch.cat([encoder_hidden_states_query_proj, query], dim=2)
-            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
-            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)
-
-        if image_rotary_emb is not None:
-            from .embeddings import apply_rotary_emb
-
-            query = apply_rotary_emb(query, image_rotary_emb)
-            key = apply_rotary_emb(key, image_rotary_emb)
-
-        hidden_states = F.scaled_dot_product_attention(
-            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
-        )
-
-        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
-        hidden_states = hidden_states.to(query.dtype)
-
-        if encoder_hidden_states is not None:
-            encoder_hidden_states, hidden_states = (
-                hidden_states[:, : encoder_hidden_states.shape[1]],
-                hidden_states[:, encoder_hidden_states.shape[1] :],
-            )
-
-            # linear proj
-            hidden_states = attn.to_out[0](hidden_states)
-            # dropout
-            hidden_states = attn.to_out[1](hidden_states)
-
-            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)
-
-            return hidden_states, encoder_hidden_states
-        else:
-            return hidden_states
-
-
-class FluxAttnProcessor2_0_NPU:
-    """Attention processor used typically in processing the SD3-like self-attention projections."""
-
-    def __init__(self):
-        if not hasattr(F, "scaled_dot_product_attention"):
-            raise ImportError(
-                "FluxAttnProcessor2_0_NPU requires PyTorch 2.0 and torch NPU, to use it, please upgrade PyTorch to 2.0 and install torch NPU"
-            )
-
-    def __call__(
-        self,
-        attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: torch.FloatTensor = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        image_rotary_emb: Optional[torch.Tensor] = None,
-    ) -> torch.FloatTensor:
-        batch_size, _, _ = hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
-
-        # `sample` projections.
-        query = attn.to_q(hidden_states)
-        key = attn.to_k(hidden_states)
-        value = attn.to_v(hidden_states)
-
-        inner_dim = key.shape[-1]
-        head_dim = inner_dim // attn.heads
-
-        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-        if attn.norm_q is not None:
-            query = attn.norm_q(query)
-        if attn.norm_k is not None:
-            key = attn.norm_k(key)
-
-        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
-        if encoder_hidden_states is not None:
-            # `context` projections.
-            encoder_hidden_states_query_proj = attn.add_q_proj(encoder_hidden_states)
-            encoder_hidden_states_key_proj = attn.add_k_proj(encoder_hidden_states)
-            encoder_hidden_states_value_proj = attn.add_v_proj(encoder_hidden_states)
-
-            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-
-            if attn.norm_added_q is not None:
-                encoder_hidden_states_query_proj = attn.norm_added_q(encoder_hidden_states_query_proj)
-            if attn.norm_added_k is not None:
-                encoder_hidden_states_key_proj = attn.norm_added_k(encoder_hidden_states_key_proj)
-
-            # attention
-            query = torch.cat([encoder_hidden_states_query_proj, query], dim=2)
-            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
-            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)
-
-        if image_rotary_emb is not None:
-            from .embeddings import apply_rotary_emb
-
-            query = apply_rotary_emb(query, image_rotary_emb)
-            key = apply_rotary_emb(key, image_rotary_emb)
-
-        if query.dtype in (torch.float16, torch.bfloat16):
-            hidden_states = torch_npu.npu_fusion_attention(
-                query,
-                key,
-                value,
-                attn.heads,
-                input_layout="BNSD",
-                pse=None,
-                scale=1.0 / math.sqrt(query.shape[-1]),
-                pre_tockens=65536,
-                next_tockens=65536,
-                keep_prob=1.0,
-                sync=False,
-                inner_precise=0,
-            )[0]
-        else:
-            hidden_states = F.scaled_dot_product_attention(query, key, value, dropout_p=0.0, is_causal=False)
-        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
-        hidden_states = hidden_states.to(query.dtype)
-
-        if encoder_hidden_states is not None:
-            encoder_hidden_states, hidden_states = (
-                hidden_states[:, : encoder_hidden_states.shape[1]],
-                hidden_states[:, encoder_hidden_states.shape[1] :],
-            )
-
-            # linear proj
-            hidden_states = attn.to_out[0](hidden_states)
-            # dropout
-            hidden_states = attn.to_out[1](hidden_states)
-            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)
-
-            return hidden_states, encoder_hidden_states
-        else:
-            return hidden_states
-
-
-class FusedFluxAttnProcessor2_0:
-    """Attention processor used typically in processing the SD3-like self-attention projections."""
-
-    def __init__(self):
-        if not hasattr(F, "scaled_dot_product_attention"):
-            raise ImportError(
-                "FusedFluxAttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
-            )
-
-    def __call__(
-        self,
-        attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: torch.FloatTensor = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        image_rotary_emb: Optional[torch.Tensor] = None,
-    ) -> torch.FloatTensor:
-        batch_size, _, _ = hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
-
-        # `sample` projections.
-        qkv = attn.to_qkv(hidden_states)
-        split_size = qkv.shape[-1] // 3
-        query, key, value = torch.split(qkv, split_size, dim=-1)
-
-        inner_dim = key.shape[-1]
-        head_dim = inner_dim // attn.heads
-
-        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-        if attn.norm_q is not None:
-            query = attn.norm_q(query)
-        if attn.norm_k is not None:
-            key = attn.norm_k(key)
-
-        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
-        # `context` projections.
-        if encoder_hidden_states is not None:
-            encoder_qkv = attn.to_added_qkv(encoder_hidden_states)
-            split_size = encoder_qkv.shape[-1] // 3
-            (
-                encoder_hidden_states_query_proj,
-                encoder_hidden_states_key_proj,
-                encoder_hidden_states_value_proj,
-            ) = torch.split(encoder_qkv, split_size, dim=-1)
-
-            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-
-            if attn.norm_added_q is not None:
-                encoder_hidden_states_query_proj = attn.norm_added_q(encoder_hidden_states_query_proj)
-            if attn.norm_added_k is not None:
-                encoder_hidden_states_key_proj = attn.norm_added_k(encoder_hidden_states_key_proj)
-
-            # attention
-            query = torch.cat([encoder_hidden_states_query_proj, query], dim=2)
-            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
-            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)
-
-        if image_rotary_emb is not None:
-            from .embeddings import apply_rotary_emb
-
-            query = apply_rotary_emb(query, image_rotary_emb)
-            key = apply_rotary_emb(key, image_rotary_emb)
-
-        hidden_states = F.scaled_dot_product_attention(query, key, value, dropout_p=0.0, is_causal=False)
-
-        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
-        hidden_states = hidden_states.to(query.dtype)
-
-        if encoder_hidden_states is not None:
-            encoder_hidden_states, hidden_states = (
-                hidden_states[:, : encoder_hidden_states.shape[1]],
-                hidden_states[:, encoder_hidden_states.shape[1] :],
-            )
-
-            # linear proj
-            hidden_states = attn.to_out[0](hidden_states)
-            # dropout
-            hidden_states = attn.to_out[1](hidden_states)
-            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)
-
-            return hidden_states, encoder_hidden_states
-        else:
-            return hidden_states
-
-
-class FusedFluxAttnProcessor2_0_NPU:
-    """Attention processor used typically in processing the SD3-like self-attention projections."""
-
-    def __init__(self):
-        if not hasattr(F, "scaled_dot_product_attention"):
-            raise ImportError(
-                "FluxAttnProcessor2_0_NPU requires PyTorch 2.0 and torch NPU, to use it, please upgrade PyTorch to 2.0, and install torch NPU"
-            )
-
-    def __call__(
-        self,
-        attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: torch.FloatTensor = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        image_rotary_emb: Optional[torch.Tensor] = None,
-    ) -> torch.FloatTensor:
-        batch_size, _, _ = hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
-
-        # `sample` projections.
-        qkv = attn.to_qkv(hidden_states)
-        split_size = qkv.shape[-1] // 3
-        query, key, value = torch.split(qkv, split_size, dim=-1)
-
-        inner_dim = key.shape[-1]
-        head_dim = inner_dim // attn.heads
-
-        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-        if attn.norm_q is not None:
-            query = attn.norm_q(query)
-        if attn.norm_k is not None:
-            key = attn.norm_k(key)
-
-        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
-        # `context` projections.
-        if encoder_hidden_states is not None:
-            encoder_qkv = attn.to_added_qkv(encoder_hidden_states)
-            split_size = encoder_qkv.shape[-1] // 3
-            (
-                encoder_hidden_states_query_proj,
-                encoder_hidden_states_key_proj,
-                encoder_hidden_states_value_proj,
-            ) = torch.split(encoder_qkv, split_size, dim=-1)
-
-            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-
-            if attn.norm_added_q is not None:
-                encoder_hidden_states_query_proj = attn.norm_added_q(encoder_hidden_states_query_proj)
-            if attn.norm_added_k is not None:
-                encoder_hidden_states_key_proj = attn.norm_added_k(encoder_hidden_states_key_proj)
-
-            # attention
-            query = torch.cat([encoder_hidden_states_query_proj, query], dim=2)
-            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
-            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)
-
-        if image_rotary_emb is not None:
-            from .embeddings import apply_rotary_emb
-
-            query = apply_rotary_emb(query, image_rotary_emb)
-            key = apply_rotary_emb(key, image_rotary_emb)
-
-        if query.dtype in (torch.float16, torch.bfloat16):
-            hidden_states = torch_npu.npu_fusion_attention(
-                query,
-                key,
-                value,
-                attn.heads,
-                input_layout="BNSD",
-                pse=None,
-                scale=1.0 / math.sqrt(query.shape[-1]),
-                pre_tockens=65536,
-                next_tockens=65536,
-                keep_prob=1.0,
-                sync=False,
-                inner_precise=0,
-            )[0]
-        else:
-            hidden_states = F.scaled_dot_product_attention(query, key, value, dropout_p=0.0, is_causal=False)
-
-        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
-        hidden_states = hidden_states.to(query.dtype)
-
-        if encoder_hidden_states is not None:
-            encoder_hidden_states, hidden_states = (
-                hidden_states[:, : encoder_hidden_states.shape[1]],
-                hidden_states[:, encoder_hidden_states.shape[1] :],
-            )
-
-            # linear proj
-            hidden_states = attn.to_out[0](hidden_states)
-            # dropout
-            hidden_states = attn.to_out[1](hidden_states)
-            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)
-
-            return hidden_states, encoder_hidden_states
-        else:
-            return hidden_states
-
-
-class FluxIPAdapterJointAttnProcessor2_0(torch.nn.Module):
-    """Flux Attention processor for IP-Adapter."""
-
-    def __init__(
-        self, hidden_size: int, cross_attention_dim: int, num_tokens=(4,), scale=1.0, device=None, dtype=None
-    ):
-        super().__init__()
-
-        if not hasattr(F, "scaled_dot_product_attention"):
-            raise ImportError(
-                f"{self.__class__.__name__} requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
-            )
-
-        self.hidden_size = hidden_size
-        self.cross_attention_dim = cross_attention_dim
-
-        if not isinstance(num_tokens, (tuple, list)):
-            num_tokens = [num_tokens]
-
-        if not isinstance(scale, list):
-            scale = [scale] * len(num_tokens)
-        if len(scale) != len(num_tokens):
-            raise ValueError("`scale` should be a list of integers with the same length as `num_tokens`.")
-        self.scale = scale
-
-        self.to_k_ip = nn.ModuleList(
-            [
-                nn.Linear(cross_attention_dim, hidden_size, bias=True, device=device, dtype=dtype)
-                for _ in range(len(num_tokens))
-            ]
-        )
-        self.to_v_ip = nn.ModuleList(
-            [
-                nn.Linear(cross_attention_dim, hidden_size, bias=True, device=device, dtype=dtype)
-                for _ in range(len(num_tokens))
-            ]
-        )
-
-    def __call__(
-        self,
-        attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: torch.FloatTensor = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        image_rotary_emb: Optional[torch.Tensor] = None,
-        ip_hidden_states: Optional[List[torch.Tensor]] = None,
-        ip_adapter_masks: Optional[torch.Tensor] = None,
-    ) -> torch.FloatTensor:
-        batch_size, _, _ = hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
-
-        # `sample` projections.
-        hidden_states_query_proj = attn.to_q(hidden_states)
-        key = attn.to_k(hidden_states)
-        value = attn.to_v(hidden_states)
-
-        inner_dim = key.shape[-1]
-        head_dim = inner_dim // attn.heads
-
-        hidden_states_query_proj = hidden_states_query_proj.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-        if attn.norm_q is not None:
-            hidden_states_query_proj = attn.norm_q(hidden_states_query_proj)
-        if attn.norm_k is not None:
-            key = attn.norm_k(key)
-
-        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
-        if encoder_hidden_states is not None:
-            # `context` projections.
-            encoder_hidden_states_query_proj = attn.add_q_proj(encoder_hidden_states)
-            encoder_hidden_states_key_proj = attn.add_k_proj(encoder_hidden_states)
-            encoder_hidden_states_value_proj = attn.add_v_proj(encoder_hidden_states)
-
-            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-
-            if attn.norm_added_q is not None:
-                encoder_hidden_states_query_proj = attn.norm_added_q(encoder_hidden_states_query_proj)
-            if attn.norm_added_k is not None:
-                encoder_hidden_states_key_proj = attn.norm_added_k(encoder_hidden_states_key_proj)
-
-            # attention
-            query = torch.cat([encoder_hidden_states_query_proj, hidden_states_query_proj], dim=2)
-            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
-            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)
-
-        if image_rotary_emb is not None:
-            from .embeddings import apply_rotary_emb
-
-            query = apply_rotary_emb(query, image_rotary_emb)
-            key = apply_rotary_emb(key, image_rotary_emb)
-
-        hidden_states = F.scaled_dot_product_attention(query, key, value, dropout_p=0.0, is_causal=False)
-        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
-        hidden_states = hidden_states.to(query.dtype)
-
-        if encoder_hidden_states is not None:
-            encoder_hidden_states, hidden_states = (
-                hidden_states[:, : encoder_hidden_states.shape[1]],
-                hidden_states[:, encoder_hidden_states.shape[1] :],
-            )
-
-            # linear proj
-            hidden_states = attn.to_out[0](hidden_states)
-            # dropout
-            hidden_states = attn.to_out[1](hidden_states)
-            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)
-
-            # IP-adapter
-            ip_query = hidden_states_query_proj
-            ip_attn_output = torch.zeros_like(hidden_states)
-
-            for current_ip_hidden_states, scale, to_k_ip, to_v_ip in zip(
-                ip_hidden_states, self.scale, self.to_k_ip, self.to_v_ip
-            ):
-                ip_key = to_k_ip(current_ip_hidden_states)
-                ip_value = to_v_ip(current_ip_hidden_states)
-
-                ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-                ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-                # the output of sdp = (batch, num_heads, seq_len, head_dim)
-                # TODO: add support for attn.scale when we move to Torch 2.1
-                current_ip_hidden_states = F.scaled_dot_product_attention(
-                    ip_query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
-                )
-                current_ip_hidden_states = current_ip_hidden_states.transpose(1, 2).reshape(
-                    batch_size, -1, attn.heads * head_dim
-                )
-                current_ip_hidden_states = current_ip_hidden_states.to(ip_query.dtype)
-                ip_attn_output += scale * current_ip_hidden_states
-
-            return hidden_states, encoder_hidden_states, ip_attn_output
-        else:
-            return hidden_states
-
-
 class CogVideoXAttnProcessor2_0:
     r"""
     Processor for implementing scaled dot-product attention for the CogVideoX model. It applies a rotary embedding on
@@ -3449,106 +2901,6 @@ class XLAFlashAttnProcessor2_0:
         return hidden_states
 
 
-class XLAFluxFlashAttnProcessor2_0:
-    r"""
-    Processor for implementing scaled dot-product attention with pallas flash attention kernel if using `torch_xla`.
-    """
-
-    def __init__(self, partition_spec: Optional[Tuple[Optional[str], ...]] = None):
-        if not hasattr(F, "scaled_dot_product_attention"):
-            raise ImportError(
-                "XLAFlashAttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
-            )
-        if is_torch_xla_version("<", "2.3"):
-            raise ImportError("XLA flash attention requires torch_xla version >= 2.3.")
-        if is_spmd() and is_torch_xla_version("<", "2.4"):
-            raise ImportError("SPMD support for XLA flash attention needs torch_xla version >= 2.4.")
-        self.partition_spec = partition_spec
-
-    def __call__(
-        self,
-        attn: Attention,
-        hidden_states: torch.FloatTensor,
-        encoder_hidden_states: torch.FloatTensor = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        image_rotary_emb: Optional[torch.Tensor] = None,
-    ) -> torch.FloatTensor:
-        batch_size, _, _ = hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
-
-        # `sample` projections.
-        query = attn.to_q(hidden_states)
-        key = attn.to_k(hidden_states)
-        value = attn.to_v(hidden_states)
-
-        inner_dim = key.shape[-1]
-        head_dim = inner_dim // attn.heads
-
-        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-        if attn.norm_q is not None:
-            query = attn.norm_q(query)
-        if attn.norm_k is not None:
-            key = attn.norm_k(key)
-
-        # the attention in FluxSingleTransformerBlock does not use `encoder_hidden_states`
-        if encoder_hidden_states is not None:
-            # `context` projections.
-            encoder_hidden_states_query_proj = attn.add_q_proj(encoder_hidden_states)
-            encoder_hidden_states_key_proj = attn.add_k_proj(encoder_hidden_states)
-            encoder_hidden_states_value_proj = attn.add_v_proj(encoder_hidden_states)
-
-            encoder_hidden_states_query_proj = encoder_hidden_states_query_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_key_proj = encoder_hidden_states_key_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-            encoder_hidden_states_value_proj = encoder_hidden_states_value_proj.view(
-                batch_size, -1, attn.heads, head_dim
-            ).transpose(1, 2)
-
-            if attn.norm_added_q is not None:
-                encoder_hidden_states_query_proj = attn.norm_added_q(encoder_hidden_states_query_proj)
-            if attn.norm_added_k is not None:
-                encoder_hidden_states_key_proj = attn.norm_added_k(encoder_hidden_states_key_proj)
-
-            # attention
-            query = torch.cat([encoder_hidden_states_query_proj, query], dim=2)
-            key = torch.cat([encoder_hidden_states_key_proj, key], dim=2)
-            value = torch.cat([encoder_hidden_states_value_proj, value], dim=2)
-
-        if image_rotary_emb is not None:
-            from .embeddings import apply_rotary_emb
-
-            query = apply_rotary_emb(query, image_rotary_emb)
-            key = apply_rotary_emb(key, image_rotary_emb)
-
-        query /= math.sqrt(head_dim)
-        hidden_states = flash_attention(query, key, value, causal=False)
-
-        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
-        hidden_states = hidden_states.to(query.dtype)
-
-        if encoder_hidden_states is not None:
-            encoder_hidden_states, hidden_states = (
-                hidden_states[:, : encoder_hidden_states.shape[1]],
-                hidden_states[:, encoder_hidden_states.shape[1] :],
-            )
-
-            # linear proj
-            hidden_states = attn.to_out[0](hidden_states)
-            # dropout
-            hidden_states = attn.to_out[1](hidden_states)
-
-            encoder_hidden_states = attn.to_add_out(encoder_hidden_states)
-
-            return hidden_states, encoder_hidden_states
-        else:
-            return hidden_states
-
-
 class MochiVaeAttnProcessor2_0:
     r"""
     Attention processor used in Mochi VAE.
@@ -3972,7 +3324,7 @@ class PAGHunyuanAttnProcessor2_0:
     r"""
     Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0). This is
     used in the HunyuanDiT model. It applies a normalization layer and rotary embedding on query and key vector. This
-    variant of the processor employs [Pertubed Attention Guidance](https://arxiv.org/abs/2403.17377).
+    variant of the processor employs [Pertubed Attention Guidance](https://huggingface.co/papers/2403.17377).
     """
 
     def __init__(self):
@@ -4095,7 +3447,7 @@ class PAGCFGHunyuanAttnProcessor2_0:
     r"""
     Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0). This is
     used in the HunyuanDiT model. It applies a normalization layer and rotary embedding on query and key vector. This
-    variant of the processor employs [Pertubed Attention Guidance](https://arxiv.org/abs/2403.17377).
+    variant of the processor employs [Pertubed Attention Guidance](https://huggingface.co/papers/2403.17377).
     """
 
     def __init__(self):
@@ -4828,7 +4180,7 @@ class SlicedAttnAddedKVProcessor:
 
 class SpatialNorm(nn.Module):
     """
-    Spatially conditioned normalization as defined in https://arxiv.org/abs/2209.09002.
+    Spatially conditioned normalization as defined in https://huggingface.co/papers/2209.09002.
 
     Args:
         f_channels (`int`):
@@ -5693,7 +5045,7 @@ class SD3IPAdapterJointAttnProcessor2_0(torch.nn.Module):
 class PAGIdentitySelfAttnProcessor2_0:
     r"""
     Processor for implementing PAG using scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
-    PAG reference: https://arxiv.org/abs/2403.17377
+    PAG reference: https://huggingface.co/papers/2403.17377
     """
 
     def __init__(self):
@@ -5792,7 +5144,7 @@ class PAGIdentitySelfAttnProcessor2_0:
 class PAGCFGIdentitySelfAttnProcessor2_0:
     r"""
     Processor for implementing PAG using scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
-    PAG reference: https://arxiv.org/abs/2403.17377
+    PAG reference: https://huggingface.co/papers/2403.17377
     """
 
     def __init__(self):
@@ -5988,17 +5340,6 @@ class LoRAAttnAddedKVProcessor:
         pass
 
 
-class FluxSingleAttnProcessor2_0(FluxAttnProcessor2_0):
-    r"""
-    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
-    """
-
-    def __init__(self):
-        deprecation_message = "`FluxSingleAttnProcessor2_0` is deprecated and will be removed in a future version. Please use `FluxAttnProcessor2_0` instead."
-        deprecate("FluxSingleAttnProcessor2_0", "0.32.0", deprecation_message)
-        super().__init__()
-
-
 class SanaLinearAttnProcessor2_0:
     r"""
     Processor for implementing scaled dot-product linear attention.
@@ -6163,6 +5504,111 @@ class PAGIdentitySanaLinearAttnProcessor2_0:
         return hidden_states
 
 
+class FluxAttnProcessor2_0:
+    def __new__(cls, *args, **kwargs):
+        deprecation_message = "`FluxAttnProcessor2_0` is deprecated and this will be removed in a future version. Please use `FluxAttnProcessor`"
+        deprecate("FluxAttnProcessor2_0", "1.0.0", deprecation_message)
+
+        from .transformers.transformer_flux import FluxAttnProcessor
+
+        return FluxAttnProcessor(*args, **kwargs)
+
+
+class FluxSingleAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
+    """
+
+    def __new__(cls, *args, **kwargs):
+        deprecation_message = "`FluxSingleAttnProcessor` is deprecated and will be removed in a future version. Please use `FluxAttnProcessorSDPA` instead."
+        deprecate("FluxSingleAttnProcessor2_0", "1.0.0", deprecation_message)
+
+        from .transformers.transformer_flux import FluxAttnProcessor
+
+        return FluxAttnProcessor(*args, **kwargs)
+
+
+class FusedFluxAttnProcessor2_0:
+    def __new__(cls, *args, **kwargs):
+        deprecation_message = "`FusedFluxAttnProcessor2_0` is deprecated and this will be removed in a future version. Please use `FluxAttnProcessor`"
+        deprecate("FusedFluxAttnProcessor2_0", "1.0.0", deprecation_message)
+
+        from .transformers.transformer_flux import FluxAttnProcessor
+
+        return FluxAttnProcessor(*args, **kwargs)
+
+
+class FluxIPAdapterJointAttnProcessor2_0:
+    def __new__(cls, *args, **kwargs):
+        deprecation_message = "`FluxIPAdapterJointAttnProcessor2_0` is deprecated and this will be removed in a future version. Please use `FluxIPAdapterAttnProcessor`"
+        deprecate("FluxIPAdapterJointAttnProcessor2_0", "1.0.0", deprecation_message)
+
+        from .transformers.transformer_flux import FluxIPAdapterAttnProcessor
+
+        return FluxIPAdapterAttnProcessor(*args, **kwargs)
+
+
+class FluxAttnProcessor2_0_NPU:
+    def __new__(cls, *args, **kwargs):
+        deprecation_message = (
+            "FluxAttnProcessor2_0_NPU is deprecated and will be removed in a future version. An "
+            "alternative solution to use NPU Flash Attention will be provided in the future."
+        )
+        deprecate("FluxAttnProcessor2_0_NPU", "1.0.0", deprecation_message, standard_warn=False)
+
+        from .transformers.transformer_flux import FluxAttnProcessor
+
+        processor = FluxAttnProcessor()
+        processor._attention_backend = "_native_npu"
+        return processor
+
+
+class FusedFluxAttnProcessor2_0_NPU:
+    def __new__(self):
+        deprecation_message = (
+            "FusedFluxAttnProcessor2_0_NPU is deprecated and will be removed in a future version. An "
+            "alternative solution to use NPU Flash Attention will be provided in the future."
+        )
+        deprecate("FusedFluxAttnProcessor2_0_NPU", "1.0.0", deprecation_message, standard_warn=False)
+
+        from .transformers.transformer_flux import FluxAttnProcessor
+
+        processor = FluxAttnProcessor()
+        processor._attention_backend = "_fused_npu"
+        return processor
+
+
+class XLAFluxFlashAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention with pallas flash attention kernel if using `torch_xla`.
+    """
+
+    def __new__(cls, *args, **kwargs):
+        deprecation_message = (
+            "XLAFluxFlashAttnProcessor2_0 is deprecated and will be removed in diffusers 1.0.0. An "
+            "alternative solution to using XLA Flash Attention will be provided in the future."
+        )
+        deprecate("XLAFluxFlashAttnProcessor2_0", "1.0.0", deprecation_message, standard_warn=False)
+
+        if is_torch_xla_version("<", "2.3"):
+            raise ImportError("XLA flash attention requires torch_xla version >= 2.3.")
+        if is_spmd() and is_torch_xla_version("<", "2.4"):
+            raise ImportError("SPMD support for XLA flash attention needs torch_xla version >= 2.4.")
+
+        from .transformers.transformer_flux import FluxAttnProcessor
+
+        if len(args) > 0 or kwargs.get("partition_spec", None) is not None:
+            deprecation_message = (
+                "partition_spec was not used in the processor implementation when it was added. Passing it "
+                "is a no-op and support for it will be removed."
+            )
+            deprecate("partition_spec", "1.0.0", deprecation_message)
+
+        processor = FluxAttnProcessor(*args, **kwargs)
+        processor._attention_backend = "_native_xla"
+        return processor
+
+
 ADDED_KV_ATTENTION_PROCESSORS = (
     AttnAddedKVProcessor,
     SlicedAttnAddedKVProcessor,