Buckets:
| # Attention Processor | |
| Attention processor adalah kelas untuk menerapkan berbagai jenis mekanisme perhatian. | |
| ## AttnProcessor[[diffusers.models.attention_processor.AttnProcessor]] | |
| #### diffusers.models.attention_processor.AttnProcessor[[diffusers.models.attention_processor.AttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.AttnProcessor() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1103) | |
| Pemroses default untuk melakukan komputasi terkait perhatian. | |
| #### diffusers.models.attention_processor.AttnProcessor2_0[[diffusers.models.attention_processor.AttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.AttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2684) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). | |
| #### diffusers.models.attention_processor.AttnAddedKVProcessor[[diffusers.models.attention_processor.AttnAddedKVProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.AttnAddedKVProcessor() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1279) | |
| Processor untuk melakukan komputasi terkait perhatian dengan matriks kunci dan nilai yang dapat dipelajari tambahan untuk encoder teks. | |
| #### diffusers.models.attention_processor.AttnAddedKVProcessor2_0[[diffusers.models.attention_processor.AttnAddedKVProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.AttnAddedKVProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1346) | |
| Processor untuk melakukan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0), dengan matriks key dan value yang dapat dipelajari tambahan untuk text encoder. | |
| #### diffusers.models.attention_processor.AttnProcessorNPU[[diffusers.models.attention_processor.AttnProcessorNPU]] | |
| ```python | |
| diffusers.models.attention_processor.AttnProcessorNPU() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2570) | |
| Processor untuk menerapkan flash attention menggunakan torch_npu. Torch_npu hanya mendukung tipe data fp16 dan bf16. Jika fp32 digunakan, F.scaled_dot_product_attention akan digunakan untuk komputasi, tetapi efek akselerasi pada NPU tidak signifikan. | |
| #### diffusers.models.attention_processor.FusedAttnProcessor2_0[[diffusers.models.attention_processor.FusedAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FusedAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3656) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Ini menggunakan lapisan proyusi yang digabungkan. Untuk modul self-attention, semua matriks proyusi (yaitu, query, key, value) digabungkan. Untuk modul cross-attention, matriks proyusi key dan value digabungkan. | |
| > [!WARNING] > API ini saat ini bersifat 🧪 eksperimental dan dapat berubah di masa depan. | |
| ## Allegro[[diffusers.models.attention_processor.AllegroAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.AllegroAttnProcessor2_0[[diffusers.models.attention_processor.AllegroAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.AllegroAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1993) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Ini digunakan dalam model Allegro. Ia menerapkan lapisan normalisasi dan rotary embedding pada vektor query dan key. | |
| ## AuraFlow[[diffusers.models.attention_processor.AuraFlowAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.AuraFlowAttnProcessor2_0[[diffusers.models.attention_processor.AuraFlowAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.AuraFlowAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2087) | |
| Processor perhatian yang biasanya digunakan dalam pemrosesan Aura Flow. | |
| #### diffusers.models.attention_processor.FusedAuraFlowAttnProcessor2_0[[diffusers.models.attention_processor.FusedAuraFlowAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FusedAuraFlowAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2174) | |
| Processor perhatian yang biasanya digunakan dalam pemrosesan Aura Flow dengan proyeksi yang digabungkan. | |
| ## CogVideoX[[diffusers.models.attention_processor.CogVideoXAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.CogVideoXAttnProcessor2_0[[diffusers.models.attention_processor.CogVideoXAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.CogVideoXAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2265) | |
| Processor untuk menerapkan scaled dot-product attention pada model CogVideoX. Ini menerapkan rotary embedding pada vektor query dan key, tetapi tidak termasuk normalisasi spasial. | |
| #### diffusers.models.attention_processor.FusedCogVideoXAttnProcessor2_0[[diffusers.models.attention_processor.FusedCogVideoXAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FusedCogVideoXAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2334) | |
| Processor untuk menerapkan scaled dot-product attention pada model CogVideoX. Ini menerapkan rotary embedding pada vektor query dan key, tetapi tidak termasuk normalisasi spasial. | |
| ## DreamLite[[diffusers.models.unets.unet_dreamlite.DreamLiteAttnProcessor2_0]] | |
| #### diffusers.models.unets.unet_dreamlite.DreamLiteAttnProcessor2_0[[diffusers.models.unets.unet_dreamlite.DreamLiteAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.unets.unet_dreamlite.DreamLiteAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/unets/unet_dreamlite.py#L285) | |
| Processor untuk menerapkan scaled dot-product attention dengan dukungan Grouped Query Attention (GQA / MQA). | |
| Identik dengan `AttnProcessor2_0` kecuali cabang reshape key/value yang menangani `attn.kv_heads != attn.heads` dengan benar, yaitu dengan melakukan reshape K/V menjadi `kv_heads` dan kemudian memperluasnya hingga `attn.heads` menggunakan `repeat_interleave`. Hal ini diperlukan oleh DreamLite UNet, yang menggabungkan GQA dengan `qk_norm` — sebuah kombinasi yang tidak ditangani oleh `AttnProcessor2_0` bawaan. SDPA didelegasikan ke `dispatch_attention_fn` sehingga backend attention diffusers mana pun (native PyTorch SDPA, FlashAttention, dll.) dapat digunakan. | |
| ## Custom Diffusion[[diffusers.models.attention_processor.CustomDiffusionAttnProcessor]] | |
| #### diffusers.models.attention_processor.CustomDiffusionAttnProcessor[[diffusers.models.attention_processor.CustomDiffusionAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.CustomDiffusionAttnProcessor(train_kv: bool = True, train_q_out: bool = True, hidden_size: int | None = None, cross_attention_dim: int | None = None, out_bias: bool = True, dropout: float = 0.0) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1175) | |
| **Parameter:** | |
| train_kv (`bool`, defaultnya `True`) : Apakah akan melatih ulang matriks kunci dan nilai yang sesuai dengan fitur teks. | |
| train_q_out (`bool`, defaultnya `True`) : Apakah akan melatih ulang matriks query yang sesuai dengan fitur gambar laten. | |
| hidden_size (`int`, *opsional*, defaultnya `None`) : Ukuran tersembunyi dari lapisan perhatian. | |
| cross_attention_dim (`int`, *opsional*, defaultnya `None`) : Jumlah channel dalam `encoder_hidden_states`. | |
| out_bias (`bool`, defaultnya `True`) : Apakah akan menyertakan parameter bias di `train_q_out`. | |
| dropout (`float`, *opsional*, default 0.0) : Probabilitas dropout yang akan digunakan. | |
| Processor untuk menerapkan perhatian (attention) untuk metode Custom Diffusion. | |
| #### diffusers.models.attention_processor.CustomDiffusionAttnProcessor2_0[[diffusers.models.attention_processor.CustomDiffusionAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.CustomDiffusionAttnProcessor2_0(train_kv: bool = True, train_q_out: bool = True, hidden_size: int | None = None, cross_attention_dim: int | None = None, out_bias: bool = True, dropout: float = 0.0) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3874) | |
| **Parameter:** | |
| train_kv (`bool`, defaultnya `True`) : Apakah akan melatih ulang matriks kunci dan nilai yang sesuai dengan fitur teks. | |
| train_q_out (`bool`, defaultnya `True`) : Apakah akan melatih ulang matriks query yang sesuai dengan fitur gambar laten. | |
| hidden_size (`int`, *opsional*, defaultnya `None`) : Ukuran tersembunyi dari lapisan perhatian. | |
| cross_attention_dim (`int`, *opsional*, defaultnya `None`) : Jumlah channel dalam `encoder_hidden_states`. | |
| out_bias (`bool`, defaultnya `True`) : Apakah akan menyertakan parameter bias di `train_q_out`. | |
| dropout (`float`, *opsional*, default 0.0) : Probabilitas dropout yang akan digunakan. | |
| Processor untuk menerapkan perhatian (attention) untuk metode Custom Diffusion menggunakan scaled dot-product attention yang hemat memori dari PyTorch 2.0. | |
| #### diffusers.models.attention_processor.CustomDiffusionXFormersAttnProcessor[[diffusers.models.attention_processor.CustomDiffusionXFormersAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.CustomDiffusionXFormersAttnProcessor(train_kv: bool = True, train_q_out: bool = False, hidden_size: int | None = None, cross_attention_dim: int | None = None, out_bias: bool = True, dropout: float = 0.0, attention_op: Callable | None = None) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3758) | |
| **Parameter:** | |
| train_kv (`bool`, defaultnya `True`) : Apakah akan melatih ulang matriks kunci dan nilai yang sesuai dengan fitur teks. | |
| train_q_out (`bool`, defaultnya `True`) : Apakah akan melatih ulang matriks query yang sesuai dengan fitur gambar laten. | |
| hidden_size (`int`, *opsional*, defaultnya `None`) : Ukuran tersembunyi dari lapisan perhatian. | |
| cross_attention_dim (`int`, *opsional*, defaultnya `None`) : Jumlah channel dalam `encoder_hidden_states`. | |
| out_bias (`bool`, defaultnya `True`) : Apakah akan menyertakan parameter bias di `train_q_out`. | |
| dropout (`float`, *opsional*, default 0.0) : Probabilitas dropout yang akan digunakan. | |
| attention_op (`Callable`, *opsional*, defaultnya `None`) : [operator](https://facebookresearch.github.io/xformers/components/ops.html#xformers.ops.AttentionOpBase) dasar yang digunakan sebagai operator perhatian. Disarankan untuk diset ke `None`, dan membiarkan xFormers memilih operator terbaik. | |
| Processor untuk menerapkan perhatian yang efisien memori menggunakan xFormers untuk metode Custom Diffusion. | |
| ## Flux[[diffusers.models.attention_processor.FluxAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.FluxAttnProcessor2_0[[diffusers.models.attention_processor.FluxAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FluxAttnProcessor2_0(*args, **kwargs) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5493) | |
| #### diffusers.models.attention_processor.FusedFluxAttnProcessor2_0[[diffusers.models.attention_processor.FusedFluxAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FusedFluxAttnProcessor2_0(*args, **kwargs) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5517) | |
| #### diffusers.models.attention_processor.FluxSingleAttnProcessor2_0[[diffusers.models.attention_processor.FluxSingleAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FluxSingleAttnProcessor2_0(*args, **kwargs) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5503) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). | |
| ## Hunyuan[[diffusers.models.attention_processor.HunyuanAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.HunyuanAttnProcessor2_0[[diffusers.models.attention_processor.HunyuanAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.HunyuanAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3112) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Ini digunakan dalam model HunyuanDiT. Menerapkan lapisan normalisasi s dan rotary embedding pada vektor query dan key. | |
| #### diffusers.models.attention_processor.FusedHunyuanAttnProcessor2_0[[diffusers.models.attention_processor.FusedHunyuanAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FusedHunyuanAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3210) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0) dengan lapisan proyeksi yang digabungkan. Ini digunakan dalam model HunyuanDiT. Ini menerapkan lapisan normalisasi s dan rotary embedding pada vektor query dan key. | |
| #### diffusers.models.attention_processor.PAGHunyuanAttnProcessor2_0[[diffusers.models.attention_processor.PAGHunyuanAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGHunyuanAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3313) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Ini digunakan dalam model HunyuanDiT. Ia menerapkan lapisan normalisasi dan rotary embedding pada vektor query dan key. Varian processor ini menggunakan [Pertubed Attention Guidance](https://huggingface.co/papers/2403.17377). | |
| #### diffusers.models.attention_processor.PAGCFGHunyuanAttnProcessor2_0[[diffusers.models.attention_processor.PAGCFGHunyuanAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGCFGHunyuanAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3436) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Ini digunakan dalam model HunyuanDiT. Ia menerapkan lapisan normalisasi dan rotary embedding pada vektor query dan key. Varian processor ini menggunakan [Pertubed Attention Guidance](https://huggingface.co/papers/2403.17377). | |
| ## IdentitySelfAttnProcessor2_0[[diffusers.models.attention_processor.PAGIdentitySelfAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.PAGIdentitySelfAttnProcessor2_0[[diffusers.models.attention_processor.PAGIdentitySelfAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGIdentitySelfAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5031) | |
| Processor untuk menerapkan PAG menggunakan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Referensi PAG: https://huggingface.co/papers/2403.17377 | |
| #### diffusers.models.attention_processor.PAGCFGIdentitySelfAttnProcessor2_0[[diffusers.models.attention_processor.PAGCFGIdentitySelfAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGCFGIdentitySelfAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5130) | |
| Processor untuk menerapkan PAG menggunakan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Referensi PAG: https://huggingface.co/papers/2403.17377 | |
| ## IP-Adapter[[diffusers.models.attention_processor.IPAdapterAttnProcessor]] | |
| #### diffusers.models.attention_processor.IPAdapterAttnProcessor[[diffusers.models.attention_processor.IPAdapterAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.IPAdapterAttnProcessor(hidden_size, cross_attention_dim = None, num_tokens = (4,), scale = 1.0) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L4196) | |
| **Parameter:** | |
| hidden_size (`int`) : Ukuran tersembunyi dari lapisan perhatian. | |
| cross_attention_dim (`int`) : Jumlah channel dalam `encoder_hidden_states`. | |
| num_tokens (`int`, `tuple[int]` atau `list[int]`, defaultnya `(4,)`) : Panjang konteks dari fitur gambar. | |
| scale (`float` atau list`float`, default 1.0) : skala bobot untuk prompt gambar. | |
| Processor perhatian untuk beberapa IP-Adapter. | |
| #### diffusers.models.attention_processor.IPAdapterAttnProcessor2_0[[diffusers.models.attention_processor.IPAdapterAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.IPAdapterAttnProcessor2_0(hidden_size, cross_attention_dim = None, num_tokens = (4,), scale = 1.0) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L4396) | |
| **Parameter:** | |
| hidden_size (`int`) : Ukuran tersembunyi dari lapisan perhatian. | |
| cross_attention_dim (`int`) : Jumlah channel dalam `encoder_hidden_states`. | |
| num_tokens (`int`, `tuple[int]` atau `list[int]`, defaultnya `(4,)`) : Panjang konteks dari fitur gambar. | |
| scale (`float` atau `list[float]`, default 1.0) : skala bobot untuk prompt gambar. | |
| Processor perhatian untuk IP-Adapter untuk PyTorch 2.0. | |
| #### diffusers.models.attention_processor.SD3IPAdapterJointAttnProcessor2_0[[diffusers.models.attention_processor.SD3IPAdapterJointAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.SD3IPAdapterJointAttnProcessor2_0(hidden_size: int, ip_hidden_states_dim: int, head_dim: int, timesteps_emb_dim: int = 1280, scale: float = 0.5) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L4860) | |
| **Parameter:** | |
| hidden_size (`int`) : Jumlah kanal tersembunyi. | |
| ip_hidden_states_dim (`int`) : Dimensi fitur gambar. | |
| head_dim (`int`) : Jumlah kanal head. | |
| timesteps_emb_dim (`int`, defaultnya 1280) : Jumlah saluran input untuk embedding timestep. | |
| scale (`float`, default 0.5) : skala IP-Adapter. | |
| Attention processor untuk IP-Adapter yang biasanya digunakan dalam memproses proyeksi self-attention mirip SD3, dengan informasi berbasis gambar tambahan dan timestep embeddings. | |
| ## JointAttnProcessor2_0[[diffusers.models.attention_processor.JointAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.JointAttnProcessor2_0[[diffusers.models.attention_processor.JointAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.JointAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1422) | |
| Attention processor yang biasanya digunakan dalam memproses proyeksi self-attention yang mirip SD3. | |
| #### diffusers.models.attention_processor.PAGJointAttnProcessor2_0[[diffusers.models.attention_processor.PAGJointAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGJointAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1508) | |
| Attention processor yang biasanya digunakan dalam memproses proyeksi self-attention yang mirip SD3. | |
| #### diffusers.models.attention_processor.PAGCFGJointAttnProcessor2_0[[diffusers.models.attention_processor.PAGCFGJointAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGCFGJointAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1664) | |
| Attention processor yang biasanya digunakan dalam memproses proyeksi self-attention yang mirip SD3. | |
| #### diffusers.models.attention_processor.FusedJointAttnProcessor2_0[[diffusers.models.attention_processor.FusedJointAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FusedJointAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1829) | |
| Attention processor yang biasanya digunakan dalam memproses proyeksi self-attention yang mirip SD3. | |
| ## LoRA[[diffusers.models.attention_processor.LoRAAttnProcessor]] | |
| #### diffusers.models.attention_processor.LoRAAttnProcessor[[diffusers.models.attention_processor.LoRAAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.LoRAAttnProcessor() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5293) | |
| Processor untuk menerapkan perhatian dengan LoRA. | |
| #### diffusers.models.attention_processor.LoRAAttnProcessor2_0[[diffusers.models.attention_processor.LoRAAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.LoRAAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5302) | |
| Processor untuk menerapkan perhatian dengan LoRA (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). | |
| #### diffusers.models.attention_processor.LoRAAttnAddedKVProcessor[[diffusers.models.attention_processor.LoRAAttnAddedKVProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.LoRAAttnAddedKVProcessor() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5320) | |
| Processor untuk menerapkan perhatian dengan LoRA dengan matriks kunci dan nilai yang dapat dipelajari tambahan untuk pengkode teks. | |
| #### diffusers.models.attention_processor.LoRAXFormersAttnProcessor[[diffusers.models.attention_processor.LoRAXFormersAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.LoRAXFormersAttnProcessor() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5311) | |
| Processor untuk menerapkan perhatian dengan LoRA menggunakan xFormers. | |
| ## Lumina-T2X[[diffusers.models.attention_processor.LuminaAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.LuminaAttnProcessor2_0[[diffusers.models.attention_processor.LuminaAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.LuminaAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3560) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Ini digunakan dalam model LuminaNextDiT. Menerapkan lapisan normalisasi s dan rotary embedding pada vektor query dan key. | |
| ## Mochi[[diffusers.models.attention_processor.MochiAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.MochiAttnProcessor2_0[[diffusers.models.attention_processor.MochiAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.MochiAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L998) | |
| Attention processor yang digunakan dalam Mochi. | |
| #### diffusers.models.attention_processor.MochiVaeAttnProcessor2_0[[diffusers.models.attention_processor.MochiVaeAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.MochiVaeAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2894) | |
| Attention processor yang digunakan dalam Mochi VAE. | |
| ## Sana[[diffusers.models.attention_processor.SanaLinearAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.SanaLinearAttnProcessor2_0[[diffusers.models.attention_processor.SanaLinearAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.SanaLinearAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5329) | |
| Processor untuk menerapkan perhatian linear dot-product berskala. | |
| #### diffusers.models.attention_processor.SanaMultiscaleAttnProcessor2_0[[diffusers.models.attention_processor.SanaMultiscaleAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.SanaMultiscaleAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5233) | |
| Processor untuk menerapkan perhatian kuadratis multiscale. | |
| #### diffusers.models.attention_processor.PAGCFGSanaLinearAttnProcessor2_0[[diffusers.models.attention_processor.PAGCFGSanaLinearAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGCFGSanaLinearAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5381) | |
| Processor untuk menerapkan perhatian linear dot-product berskala. | |
| #### diffusers.models.attention_processor.PAGIdentitySanaLinearAttnProcessor2_0[[diffusers.models.attention_processor.PAGIdentitySanaLinearAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.PAGIdentitySanaLinearAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5436) | |
| Processor untuk menerapkan perhatian linear dot-product berskala. | |
| ## Stable Audio[[diffusers.models.attention_processor.StableAudioAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.StableAudioAttnProcessor2_0[[diffusers.models.attention_processor.StableAudioAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.StableAudioAttnProcessor2_0() | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2979) | |
| Processor untuk menerapkan scaled dot-product attention (diaktifkan secara default jika Anda menggunakan PyTorch 2.0). Ini digunakan dalam model Stable Audio. Menerapkan rotary embedding pada vektor query dan key, serta memungkinkan MHA, GQA, atau MQA. | |
| ## SlicedAttnProcessor[[diffusers.models.attention_processor.SlicedAttnProcessor]] | |
| #### diffusers.models.attention_processor.SlicedAttnProcessor[[diffusers.models.attention_processor.SlicedAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.SlicedAttnProcessor(slice_size: int) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L3988) | |
| **Parameter:** | |
| slice_size (`int`, *opsional*) : Jumlah langkah untuk menghitung perhatian. Menggunakan sebanyak `attention_head_dim // slice_size` irisan, dan `attention_head_dim` harus merupakan kelipatan dari `slice_size`. | |
| Processor untuk menerapkan sliced attention. | |
| #### diffusers.models.attention_processor.SlicedAttnAddedKVProcessor[[diffusers.models.attention_processor.SlicedAttnAddedKVProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.SlicedAttnAddedKVProcessor(slice_size) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L4075) | |
| **Parameter:** | |
| slice_size (`int`, *opsional*) : Jumlah langkah untuk menghitung perhatian. Menggunakan sebanyak `attention_head_dim // slice_size` irisan, dan `attention_head_dim` harus merupakan kelipatan dari `slice_size`. | |
| Processor untuk menerapkan sliced attention dengan matriks key dan value yang dapat dipelajari tambahan untuk text encoder. | |
| ## XFormersAttnProcessor[[diffusers.models.attention_processor.XFormersAttnProcessor]] | |
| #### diffusers.models.attention_processor.XFormersAttnProcessor[[diffusers.models.attention_processor.XFormersAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.XFormersAttnProcessor(attention_op: Callable | None = None) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2476) | |
| **Parameter:** | |
| attention_op (`Callable`, *opsional*, defaultnya `None`) : [operator](https://facebookresearch.github.io/xformers/components/ops.html#xformers.ops.AttentionOpBase) dasar yang digunakan sebagai operator perhatian. Disarankan untuk diset ke `None`, dan membiarkan xFormers memilih operator terbaik. | |
| Processor untuk menerapkan perhatian yang hemat memori menggunakan xFormers. | |
| #### diffusers.models.attention_processor.XFormersAttnAddedKVProcessor[[diffusers.models.attention_processor.XFormersAttnAddedKVProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.XFormersAttnAddedKVProcessor(attention_op: Callable | None = None) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2405) | |
| **Parameter:** | |
| attention_op (`Callable`, *opsional*, defaultnya `None`) : [operator](https://facebookresearch.github.io/xformers/components/ops.html#xformers.ops.AttentionOpBase) dasar yang digunakan sebagai operator perhatian. Disarankan untuk diset ke `None`, dan membiarkan xFormers memilih operator terbaik. | |
| Processor untuk menerapkan perhatian yang hemat memori menggunakan xFormers. | |
| ## XLAFlashAttnProcessor2_0[[diffusers.models.attention_processor.XLAFlashAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.XLAFlashAttnProcessor2_0[[diffusers.models.attention_processor.XLAFlashAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.XLAFlashAttnProcessor2_0(partition_spec: tuple[str | None, ...] | None = None) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L2778) | |
| Processor untuk menerapkan scaled dot-product attention dengan kernel pallas flash attention jika menggunakan `torch_xla`. | |
| ## XFormersJointAttnProcessor[[diffusers.models.attention_processor.XFormersJointAttnProcessor]] | |
| #### diffusers.models.attention_processor.XFormersJointAttnProcessor[[diffusers.models.attention_processor.XFormersJointAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.XFormersJointAttnProcessor(attention_op: Callable | None = None) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L1908) | |
| **Parameter:** | |
| attention_op (`Callable`, *opsional*, defaultnya `None`) : [operator](https://facebookresearch.github.io/xformers/components/ops.html#xformers.ops.AttentionOpBase) dasar yang digunakan sebagai operator perhatian. Disarankan untuk diset ke `None`, dan membiarkan xFormers memilih operator terbaik. | |
| Processor untuk menerapkan perhatian yang hemat memori menggunakan xFormers. | |
| ## IPAdapterXFormersAttnProcessor[[diffusers.models.attention_processor.IPAdapterXFormersAttnProcessor]] | |
| #### diffusers.models.attention_processor.IPAdapterXFormersAttnProcessor[[diffusers.models.attention_processor.IPAdapterXFormersAttnProcessor]] | |
| ```python | |
| diffusers.models.attention_processor.IPAdapterXFormersAttnProcessor(hidden_size, cross_attention_dim = None, num_tokens = (4,), scale = 1.0, attention_op: Callable | None = None) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L4628) | |
| **Parameter:** | |
| hidden_size (`int`) : Ukuran tersembunyi dari lapisan perhatian. | |
| cross_attention_dim (`int`) : Jumlah channel dalam `encoder_hidden_states`. | |
| num_tokens (`int`, `tuple[int]` atau `list[int]`, defaultnya `(4,)`) : Panjang konteks dari fitur gambar. | |
| scale (`float` atau `list[float]`, default 1.0) : skala bobot untuk prompt gambar. | |
| attention_op (`Callable`, *opsional*, defaultnya `None`) : [operator](https://facebookresearch.github.io/xformers/components/ops.html#xformers.ops.AttentionOpBase) dasar yang digunakan sebagai operator perhatian. Disarankan untuk diset ke `None`, dan membiarkan xFormers memilih operator terbaik. | |
| Processor perhatian untuk IP-Adapter menggunakan xFormers. | |
| ## FluxIPAdapterJointAttnProcessor2_0[[diffusers.models.attention_processor.FluxIPAdapterJointAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.FluxIPAdapterJointAttnProcessor2_0[[diffusers.models.attention_processor.FluxIPAdapterJointAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.FluxIPAdapterJointAttnProcessor2_0(*args, **kwargs) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5527) | |
| ## XLAFluxFlashAttnProcessor2_0[[diffusers.models.attention_processor.XLAFluxFlashAttnProcessor2_0]] | |
| #### diffusers.models.attention_processor.XLAFluxFlashAttnProcessor2_0[[diffusers.models.attention_processor.XLAFluxFlashAttnProcessor2_0]] | |
| ```python | |
| diffusers.models.attention_processor.XLAFluxFlashAttnProcessor2_0(*args, **kwargs) | |
| ``` | |
| [Sumber](https://github.com/huggingface/diffusers/blob/v0.43.0/src/diffusers/models/attention_processor.py#L5567) | |
| Processor untuk menerapkan scaled dot-product attention dengan kernel pallas flash attention jika menggunakan `torch_xla`. | |
Xet Storage Details
- Size:
- 33.1 kB
- Xet hash:
- ab297a22a7852e7b47afa24e1d6123781f1e03923cf6b331bd4ffe5cc6f225b1
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.