MCPcopy Create free account
hub / github.com/MeiGen-AI/MultiTalk / __init__

Method __init__

wan/modules/multitalk_model.py:442–545  ·  view source on GitHub ↗
(self,
                 model_type='i2v',
                 patch_size=(1, 2, 2),
                 text_len=512,
                 in_dim=16,
                 dim=2048,
                 ffn_dim=8192,
                 freq_dim=256,
                 text_dim=4096,
                 out_dim=16,
                 num_heads=16,
                 num_layers=32,
                 window_size=(-1, -1),
                 qk_norm=True,
                 cross_attn_norm=True,
                 eps=1e-6,
                 # audio params
                 audio_window=5,
                 intermediate_dim=512,
                 output_dim=768,
                 context_tokens=32,
                 vae_scale=4, # vae timedownsample scale

                 norm_input_visual=True,
                 norm_output_audio=True,
                 weight_init=True)

Source from the content-addressed store, hash-verified

440
441 @register_to_config
442 def __init__(self,
443 model_type='i2v',
444 patch_size=(1, 2, 2),
445 text_len=512,
446 in_dim=16,
447 dim=2048,
448 ffn_dim=8192,
449 freq_dim=256,
450 text_dim=4096,
451 out_dim=16,
452 num_heads=16,
453 num_layers=32,
454 window_size=(-1, -1),
455 qk_norm=True,
456 cross_attn_norm=True,
457 eps=1e-6,
458 # audio params
459 audio_window=5,
460 intermediate_dim=512,
461 output_dim=768,
462 context_tokens=32,
463 vae_scale=4, # vae timedownsample scale
464
465 norm_input_visual=True,
466 norm_output_audio=True,
467 weight_init=True):
468 super().__init__()
469
470 assert model_type == 'i2v', 'MultiTalk model requires your model_type is i2v.'
471 self.model_type = model_type
472
473 self.patch_size = patch_size
474 self.text_len = text_len
475 self.in_dim = in_dim
476 self.dim = dim
477 self.ffn_dim = ffn_dim
478 self.freq_dim = freq_dim
479 self.text_dim = text_dim
480 self.out_dim = out_dim
481 self.num_heads = num_heads
482 self.num_layers = num_layers
483 self.window_size = window_size
484 self.qk_norm = qk_norm
485 self.cross_attn_norm = cross_attn_norm
486 self.eps = eps
487
488
489 self.norm_output_audio = norm_output_audio
490 self.audio_window = audio_window
491 self.intermediate_dim = intermediate_dim
492 self.vae_scale = vae_scale
493
494
495 # embeddings
496 self.patch_embedding = nn.Conv3d(
497 in_dim, dim, kernel_size=patch_size, stride=patch_size)
498 self.text_embedding = nn.Sequential(
499 nn.Linear(text_dim, dim), nn.GELU(approximate='tanh'),

Callers

nothing calls this directly

Calls 7

init_weightsMethod · 0.95
AudioProjModelClass · 0.85
WanAttentionBlockClass · 0.70
HeadClass · 0.70
rope_paramsFunction · 0.70
MLPProjClass · 0.70
__init__Method · 0.45

Tested by

no test coverage detected