`bridge.diffusion.models.wan.wan_provider`#

Module Contents#

Classes#

WanModelProvider

Data#

logger

API#

bridge.diffusion.models.wan.wan_provider.logger#: ‘getLogger(…)’

class bridge.diffusion.models.wan.wan_provider.WanModelProvider#

Bases: megatron.bridge.models.transformer_config.TransformerConfig, megatron.bridge.models.model_provider.ModelProviderMixin[megatron.core.models.common.vision_module.vision_module.VisionModule]

crossattn_emb_size: int#: 1536

add_bias_linear: bool#: True

gated_linear_unit: bool#: False

num_layers: int#: 30

hidden_size: int#: 1536

ffn_hidden_size: int#: 8960

num_attention_heads: int#: 12

layernorm_epsilon: float#: 1e-06

normalization: str#: ‘RMSNorm’

layernorm_zero_centered_gamma: bool#: False

layernorm_across_heads: bool#: True

add_qkv_bias: bool#: True

rotary_interleaved: bool#: True

activation_func: Callable#: None

hidden_dropout: float#: 0

attention_dropout: float#: 0

fp16_lm_cross_entropy: bool#: False

parallel_output: bool#: True

bf16: bool#: False

params_dtype: torch.dtype#: None

qkv_format: str#: ‘thd’

apply_rope_fusion: bool#: False

bias_activation_fusion: bool#: True

seq_length: int#: 1024

share_embeddings_and_output_weights: bool#: False

vocab_size: int#: None

make_vocab_size_divisible_by: int#: 128

in_channels: int#: 16

out_channels: int#: 16

patch_spatial: int#: 2

patch_temporal: int#: 1

freq_dim: int#: 256

text_len: int#: 512

text_dim: int#: 4096

provide( pre_process=None, post_process=None, vp_stage=None, ) → megatron.bridge.diffusion.models.wan.wan_model.WanModel#

bridge.diffusion.models.wan.wan_provider#

Module Contents#

Classes#

Data#

API#

`bridge.diffusion.models.wan.wan_provider`#