mirror of
https://github.com/modelscope/DiffSynth-Studio.git
synced 2026-03-22 08:40:47 +00:00
svd
This commit is contained in:
Artiprocher
@@ -16,6 +16,11 @@ from .sd_controlnet import SDControlNet
|
||||
|
||||
from .sd_motion import SDMotionModel
|
||||
|
||||
from .svd_image_encoder import SVDImageEncoder
|
||||
from .svd_unet import SVDUNet
|
||||
from .svd_vae_decoder import SVDVAEDecoder
|
||||
from .svd_vae_encoder import SVDVAEEncoder
|
||||
|
||||
|
||||
class ModelManager:
|
||||
def __init__(self, torch_dtype=torch.float16, device="cuda"):
|
||||
@@ -25,6 +30,10 @@ class ModelManager:
|
||||
self.model_path = {}
|
||||
self.textual_inversion_dict = {}
|
||||
|
||||
def is_stable_video_diffusion(self, state_dict):
|
||||
param_name = "model.diffusion_model.output_blocks.9.1.time_stack.0.norm_in.weight"
|
||||
return param_name in state_dict
|
||||
|
||||
def is_RIFE(self, state_dict):
|
||||
param_name = "block_tea.convblock3.0.1.weight"
|
||||
return param_name in state_dict or ("module." + param_name) in state_dict
|
||||
@@ -60,6 +69,21 @@ class ModelManager:
|
||||
param_name = "model.encoder.layers.5.self_attn_layer_norm.weight"
|
||||
return param_name in state_dict and len(state_dict) == 254
|
||||
|
||||
def load_stable_video_diffusion(self, state_dict, components=None, file_path=""):
|
||||
component_dict = {
|
||||
"image_encoder": SVDImageEncoder,
|
||||
"unet": SVDUNet,
|
||||
"vae_decoder": SVDVAEDecoder,
|
||||
"vae_encoder": SVDVAEEncoder,
|
||||
}
|
||||
if components is None:
|
||||
components = ["image_encoder", "unet", "vae_decoder", "vae_encoder"]
|
||||
for component in components:
|
||||
self.model[component] = component_dict[component]()
|
||||
self.model[component].load_state_dict(self.model[component].state_dict_converter().from_civitai(state_dict))
|
||||
self.model[component].to(self.torch_dtype).to(self.device)
|
||||
self.model_path[component] = file_path
|
||||
|
||||
def load_stable_diffusion(self, state_dict, components=None, file_path=""):
|
||||
component_dict = {
|
||||
"text_encoder": SDTextEncoder,
|
||||
@@ -190,7 +214,9 @@ class ModelManager:
|
||||
|
||||
def load_model(self, file_path, components=None, lora_alphas=[]):
|
||||
state_dict = load_state_dict(file_path, torch_dtype=self.torch_dtype)
|
||||
if self.is_animatediff(state_dict):
|
||||
if self.is_stable_video_diffusion(state_dict):
|
||||
self.load_stable_video_diffusion(state_dict, file_path=file_path)
|
||||
elif self.is_animatediff(state_dict):
|
||||
self.load_animatediff(state_dict, file_path=file_path)
|
||||
elif self.is_controlnet(state_dict):
|
||||
self.load_controlnet(state_dict, file_path=file_path)
|
||||
|
||||
@@ -165,6 +165,8 @@ class AttentionBlock(torch.nn.Module):
|
||||
encoder_hidden_states = text_emb.mean(dim=0, keepdim=True)
|
||||
else:
|
||||
encoder_hidden_states = text_emb
|
||||
if encoder_hidden_states.shape[0] != hidden_states.shape[0]:
|
||||
encoder_hidden_states = encoder_hidden_states.repeat(hidden_states.shape[0], 1, 1)
|
||||
|
||||
if tiled:
|
||||
tile_size = min(tile_size, min(height, width))
|
||||
|
||||
@@ -87,3 +87,416 @@ class SVDImageEncoderStateDictConverter:
|
||||
name_ = ".".join(["encoders", layer_id, attn_rename_dict[layer_type], tail])
|
||||
state_dict_[name_] = param
|
||||
return state_dict_
|
||||
|
||||
def from_civitai(self, state_dict):
|
||||
rename_dict = {
|
||||
"conditioner.embedders.0.open_clip.model.visual.class_embedding": "embeddings.class_embedding",
|
||||
"conditioner.embedders.0.open_clip.model.visual.conv1.weight": "embeddings.patch_embedding.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.ln_post.bias": "post_layernorm.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.ln_post.weight": "post_layernorm.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.ln_pre.bias": "pre_layernorm.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.ln_pre.weight": "pre_layernorm.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.positional_embedding": "embeddings.position_embeds",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.attn.in_proj_bias": ['encoders.0.attn.to_q.bias', 'encoders.0.attn.to_k.bias', 'encoders.0.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.attn.in_proj_weight": ['encoders.0.attn.to_q.weight', 'encoders.0.attn.to_k.weight', 'encoders.0.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.attn.out_proj.bias": "encoders.0.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.attn.out_proj.weight": "encoders.0.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_1.bias": "encoders.0.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_1.weight": "encoders.0.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_2.bias": "encoders.0.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_2.weight": "encoders.0.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_fc.bias": "encoders.0.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_fc.weight": "encoders.0.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_proj.bias": "encoders.0.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_proj.weight": "encoders.0.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.attn.in_proj_bias": ['encoders.1.attn.to_q.bias', 'encoders.1.attn.to_k.bias', 'encoders.1.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.attn.in_proj_weight": ['encoders.1.attn.to_q.weight', 'encoders.1.attn.to_k.weight', 'encoders.1.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.attn.out_proj.bias": "encoders.1.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.attn.out_proj.weight": "encoders.1.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_1.bias": "encoders.1.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_1.weight": "encoders.1.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_2.bias": "encoders.1.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_2.weight": "encoders.1.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_fc.bias": "encoders.1.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_fc.weight": "encoders.1.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_proj.bias": "encoders.1.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_proj.weight": "encoders.1.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.attn.in_proj_bias": ['encoders.10.attn.to_q.bias', 'encoders.10.attn.to_k.bias', 'encoders.10.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.attn.in_proj_weight": ['encoders.10.attn.to_q.weight', 'encoders.10.attn.to_k.weight', 'encoders.10.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.attn.out_proj.bias": "encoders.10.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.attn.out_proj.weight": "encoders.10.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_1.bias": "encoders.10.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_1.weight": "encoders.10.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_2.bias": "encoders.10.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_2.weight": "encoders.10.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_fc.bias": "encoders.10.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_fc.weight": "encoders.10.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_proj.bias": "encoders.10.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_proj.weight": "encoders.10.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.attn.in_proj_bias": ['encoders.11.attn.to_q.bias', 'encoders.11.attn.to_k.bias', 'encoders.11.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.attn.in_proj_weight": ['encoders.11.attn.to_q.weight', 'encoders.11.attn.to_k.weight', 'encoders.11.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.attn.out_proj.bias": "encoders.11.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.attn.out_proj.weight": "encoders.11.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_1.bias": "encoders.11.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_1.weight": "encoders.11.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_2.bias": "encoders.11.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_2.weight": "encoders.11.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_fc.bias": "encoders.11.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_fc.weight": "encoders.11.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_proj.bias": "encoders.11.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_proj.weight": "encoders.11.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.attn.in_proj_bias": ['encoders.12.attn.to_q.bias', 'encoders.12.attn.to_k.bias', 'encoders.12.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.attn.in_proj_weight": ['encoders.12.attn.to_q.weight', 'encoders.12.attn.to_k.weight', 'encoders.12.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.attn.out_proj.bias": "encoders.12.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.attn.out_proj.weight": "encoders.12.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_1.bias": "encoders.12.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_1.weight": "encoders.12.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_2.bias": "encoders.12.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_2.weight": "encoders.12.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_fc.bias": "encoders.12.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_fc.weight": "encoders.12.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_proj.bias": "encoders.12.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_proj.weight": "encoders.12.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.attn.in_proj_bias": ['encoders.13.attn.to_q.bias', 'encoders.13.attn.to_k.bias', 'encoders.13.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.attn.in_proj_weight": ['encoders.13.attn.to_q.weight', 'encoders.13.attn.to_k.weight', 'encoders.13.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.attn.out_proj.bias": "encoders.13.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.attn.out_proj.weight": "encoders.13.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_1.bias": "encoders.13.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_1.weight": "encoders.13.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_2.bias": "encoders.13.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_2.weight": "encoders.13.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_fc.bias": "encoders.13.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_fc.weight": "encoders.13.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_proj.bias": "encoders.13.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_proj.weight": "encoders.13.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.attn.in_proj_bias": ['encoders.14.attn.to_q.bias', 'encoders.14.attn.to_k.bias', 'encoders.14.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.attn.in_proj_weight": ['encoders.14.attn.to_q.weight', 'encoders.14.attn.to_k.weight', 'encoders.14.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.attn.out_proj.bias": "encoders.14.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.attn.out_proj.weight": "encoders.14.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_1.bias": "encoders.14.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_1.weight": "encoders.14.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_2.bias": "encoders.14.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_2.weight": "encoders.14.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_fc.bias": "encoders.14.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_fc.weight": "encoders.14.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_proj.bias": "encoders.14.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_proj.weight": "encoders.14.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.attn.in_proj_bias": ['encoders.15.attn.to_q.bias', 'encoders.15.attn.to_k.bias', 'encoders.15.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.attn.in_proj_weight": ['encoders.15.attn.to_q.weight', 'encoders.15.attn.to_k.weight', 'encoders.15.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.attn.out_proj.bias": "encoders.15.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.attn.out_proj.weight": "encoders.15.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_1.bias": "encoders.15.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_1.weight": "encoders.15.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_2.bias": "encoders.15.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_2.weight": "encoders.15.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_fc.bias": "encoders.15.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_fc.weight": "encoders.15.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_proj.bias": "encoders.15.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_proj.weight": "encoders.15.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.attn.in_proj_bias": ['encoders.16.attn.to_q.bias', 'encoders.16.attn.to_k.bias', 'encoders.16.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.attn.in_proj_weight": ['encoders.16.attn.to_q.weight', 'encoders.16.attn.to_k.weight', 'encoders.16.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.attn.out_proj.bias": "encoders.16.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.attn.out_proj.weight": "encoders.16.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_1.bias": "encoders.16.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_1.weight": "encoders.16.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_2.bias": "encoders.16.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_2.weight": "encoders.16.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_fc.bias": "encoders.16.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_fc.weight": "encoders.16.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_proj.bias": "encoders.16.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_proj.weight": "encoders.16.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.attn.in_proj_bias": ['encoders.17.attn.to_q.bias', 'encoders.17.attn.to_k.bias', 'encoders.17.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.attn.in_proj_weight": ['encoders.17.attn.to_q.weight', 'encoders.17.attn.to_k.weight', 'encoders.17.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.attn.out_proj.bias": "encoders.17.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.attn.out_proj.weight": "encoders.17.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_1.bias": "encoders.17.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_1.weight": "encoders.17.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_2.bias": "encoders.17.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_2.weight": "encoders.17.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_fc.bias": "encoders.17.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_fc.weight": "encoders.17.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_proj.bias": "encoders.17.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_proj.weight": "encoders.17.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.attn.in_proj_bias": ['encoders.18.attn.to_q.bias', 'encoders.18.attn.to_k.bias', 'encoders.18.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.attn.in_proj_weight": ['encoders.18.attn.to_q.weight', 'encoders.18.attn.to_k.weight', 'encoders.18.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.attn.out_proj.bias": "encoders.18.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.attn.out_proj.weight": "encoders.18.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_1.bias": "encoders.18.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_1.weight": "encoders.18.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_2.bias": "encoders.18.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_2.weight": "encoders.18.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_fc.bias": "encoders.18.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_fc.weight": "encoders.18.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_proj.bias": "encoders.18.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_proj.weight": "encoders.18.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.attn.in_proj_bias": ['encoders.19.attn.to_q.bias', 'encoders.19.attn.to_k.bias', 'encoders.19.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.attn.in_proj_weight": ['encoders.19.attn.to_q.weight', 'encoders.19.attn.to_k.weight', 'encoders.19.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.attn.out_proj.bias": "encoders.19.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.attn.out_proj.weight": "encoders.19.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_1.bias": "encoders.19.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_1.weight": "encoders.19.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_2.bias": "encoders.19.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_2.weight": "encoders.19.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_fc.bias": "encoders.19.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_fc.weight": "encoders.19.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_proj.bias": "encoders.19.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_proj.weight": "encoders.19.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.attn.in_proj_bias": ['encoders.2.attn.to_q.bias', 'encoders.2.attn.to_k.bias', 'encoders.2.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.attn.in_proj_weight": ['encoders.2.attn.to_q.weight', 'encoders.2.attn.to_k.weight', 'encoders.2.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.attn.out_proj.bias": "encoders.2.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.attn.out_proj.weight": "encoders.2.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_1.bias": "encoders.2.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_1.weight": "encoders.2.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_2.bias": "encoders.2.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_2.weight": "encoders.2.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_fc.bias": "encoders.2.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_fc.weight": "encoders.2.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_proj.bias": "encoders.2.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_proj.weight": "encoders.2.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.attn.in_proj_bias": ['encoders.20.attn.to_q.bias', 'encoders.20.attn.to_k.bias', 'encoders.20.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.attn.in_proj_weight": ['encoders.20.attn.to_q.weight', 'encoders.20.attn.to_k.weight', 'encoders.20.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.attn.out_proj.bias": "encoders.20.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.attn.out_proj.weight": "encoders.20.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_1.bias": "encoders.20.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_1.weight": "encoders.20.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_2.bias": "encoders.20.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_2.weight": "encoders.20.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_fc.bias": "encoders.20.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_fc.weight": "encoders.20.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_proj.bias": "encoders.20.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_proj.weight": "encoders.20.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.attn.in_proj_bias": ['encoders.21.attn.to_q.bias', 'encoders.21.attn.to_k.bias', 'encoders.21.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.attn.in_proj_weight": ['encoders.21.attn.to_q.weight', 'encoders.21.attn.to_k.weight', 'encoders.21.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.attn.out_proj.bias": "encoders.21.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.attn.out_proj.weight": "encoders.21.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_1.bias": "encoders.21.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_1.weight": "encoders.21.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_2.bias": "encoders.21.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_2.weight": "encoders.21.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_fc.bias": "encoders.21.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_fc.weight": "encoders.21.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_proj.bias": "encoders.21.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_proj.weight": "encoders.21.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.attn.in_proj_bias": ['encoders.22.attn.to_q.bias', 'encoders.22.attn.to_k.bias', 'encoders.22.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.attn.in_proj_weight": ['encoders.22.attn.to_q.weight', 'encoders.22.attn.to_k.weight', 'encoders.22.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.attn.out_proj.bias": "encoders.22.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.attn.out_proj.weight": "encoders.22.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_1.bias": "encoders.22.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_1.weight": "encoders.22.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_2.bias": "encoders.22.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_2.weight": "encoders.22.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_fc.bias": "encoders.22.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_fc.weight": "encoders.22.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_proj.bias": "encoders.22.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_proj.weight": "encoders.22.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.attn.in_proj_bias": ['encoders.23.attn.to_q.bias', 'encoders.23.attn.to_k.bias', 'encoders.23.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.attn.in_proj_weight": ['encoders.23.attn.to_q.weight', 'encoders.23.attn.to_k.weight', 'encoders.23.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.attn.out_proj.bias": "encoders.23.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.attn.out_proj.weight": "encoders.23.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_1.bias": "encoders.23.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_1.weight": "encoders.23.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_2.bias": "encoders.23.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_2.weight": "encoders.23.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_fc.bias": "encoders.23.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_fc.weight": "encoders.23.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_proj.bias": "encoders.23.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_proj.weight": "encoders.23.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.attn.in_proj_bias": ['encoders.24.attn.to_q.bias', 'encoders.24.attn.to_k.bias', 'encoders.24.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.attn.in_proj_weight": ['encoders.24.attn.to_q.weight', 'encoders.24.attn.to_k.weight', 'encoders.24.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.attn.out_proj.bias": "encoders.24.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.attn.out_proj.weight": "encoders.24.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_1.bias": "encoders.24.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_1.weight": "encoders.24.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_2.bias": "encoders.24.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_2.weight": "encoders.24.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_fc.bias": "encoders.24.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_fc.weight": "encoders.24.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_proj.bias": "encoders.24.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_proj.weight": "encoders.24.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.attn.in_proj_bias": ['encoders.25.attn.to_q.bias', 'encoders.25.attn.to_k.bias', 'encoders.25.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.attn.in_proj_weight": ['encoders.25.attn.to_q.weight', 'encoders.25.attn.to_k.weight', 'encoders.25.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.attn.out_proj.bias": "encoders.25.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.attn.out_proj.weight": "encoders.25.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_1.bias": "encoders.25.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_1.weight": "encoders.25.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_2.bias": "encoders.25.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_2.weight": "encoders.25.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_fc.bias": "encoders.25.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_fc.weight": "encoders.25.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_proj.bias": "encoders.25.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_proj.weight": "encoders.25.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.attn.in_proj_bias": ['encoders.26.attn.to_q.bias', 'encoders.26.attn.to_k.bias', 'encoders.26.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.attn.in_proj_weight": ['encoders.26.attn.to_q.weight', 'encoders.26.attn.to_k.weight', 'encoders.26.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.attn.out_proj.bias": "encoders.26.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.attn.out_proj.weight": "encoders.26.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_1.bias": "encoders.26.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_1.weight": "encoders.26.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_2.bias": "encoders.26.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_2.weight": "encoders.26.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_fc.bias": "encoders.26.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_fc.weight": "encoders.26.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_proj.bias": "encoders.26.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_proj.weight": "encoders.26.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.attn.in_proj_bias": ['encoders.27.attn.to_q.bias', 'encoders.27.attn.to_k.bias', 'encoders.27.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.attn.in_proj_weight": ['encoders.27.attn.to_q.weight', 'encoders.27.attn.to_k.weight', 'encoders.27.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.attn.out_proj.bias": "encoders.27.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.attn.out_proj.weight": "encoders.27.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_1.bias": "encoders.27.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_1.weight": "encoders.27.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_2.bias": "encoders.27.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_2.weight": "encoders.27.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_fc.bias": "encoders.27.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_fc.weight": "encoders.27.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_proj.bias": "encoders.27.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_proj.weight": "encoders.27.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.attn.in_proj_bias": ['encoders.28.attn.to_q.bias', 'encoders.28.attn.to_k.bias', 'encoders.28.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.attn.in_proj_weight": ['encoders.28.attn.to_q.weight', 'encoders.28.attn.to_k.weight', 'encoders.28.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.attn.out_proj.bias": "encoders.28.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.attn.out_proj.weight": "encoders.28.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_1.bias": "encoders.28.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_1.weight": "encoders.28.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_2.bias": "encoders.28.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_2.weight": "encoders.28.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_fc.bias": "encoders.28.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_fc.weight": "encoders.28.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_proj.bias": "encoders.28.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_proj.weight": "encoders.28.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.attn.in_proj_bias": ['encoders.29.attn.to_q.bias', 'encoders.29.attn.to_k.bias', 'encoders.29.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.attn.in_proj_weight": ['encoders.29.attn.to_q.weight', 'encoders.29.attn.to_k.weight', 'encoders.29.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.attn.out_proj.bias": "encoders.29.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.attn.out_proj.weight": "encoders.29.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_1.bias": "encoders.29.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_1.weight": "encoders.29.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_2.bias": "encoders.29.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_2.weight": "encoders.29.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_fc.bias": "encoders.29.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_fc.weight": "encoders.29.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_proj.bias": "encoders.29.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_proj.weight": "encoders.29.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.attn.in_proj_bias": ['encoders.3.attn.to_q.bias', 'encoders.3.attn.to_k.bias', 'encoders.3.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.attn.in_proj_weight": ['encoders.3.attn.to_q.weight', 'encoders.3.attn.to_k.weight', 'encoders.3.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.attn.out_proj.bias": "encoders.3.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.attn.out_proj.weight": "encoders.3.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_1.bias": "encoders.3.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_1.weight": "encoders.3.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_2.bias": "encoders.3.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_2.weight": "encoders.3.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_fc.bias": "encoders.3.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_fc.weight": "encoders.3.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_proj.bias": "encoders.3.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_proj.weight": "encoders.3.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.attn.in_proj_bias": ['encoders.30.attn.to_q.bias', 'encoders.30.attn.to_k.bias', 'encoders.30.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.attn.in_proj_weight": ['encoders.30.attn.to_q.weight', 'encoders.30.attn.to_k.weight', 'encoders.30.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.attn.out_proj.bias": "encoders.30.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.attn.out_proj.weight": "encoders.30.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_1.bias": "encoders.30.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_1.weight": "encoders.30.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_2.bias": "encoders.30.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_2.weight": "encoders.30.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_fc.bias": "encoders.30.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_fc.weight": "encoders.30.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_proj.bias": "encoders.30.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_proj.weight": "encoders.30.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.attn.in_proj_bias": ['encoders.31.attn.to_q.bias', 'encoders.31.attn.to_k.bias', 'encoders.31.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.attn.in_proj_weight": ['encoders.31.attn.to_q.weight', 'encoders.31.attn.to_k.weight', 'encoders.31.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.attn.out_proj.bias": "encoders.31.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.attn.out_proj.weight": "encoders.31.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_1.bias": "encoders.31.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_1.weight": "encoders.31.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_2.bias": "encoders.31.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_2.weight": "encoders.31.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_fc.bias": "encoders.31.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_fc.weight": "encoders.31.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_proj.bias": "encoders.31.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_proj.weight": "encoders.31.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.attn.in_proj_bias": ['encoders.4.attn.to_q.bias', 'encoders.4.attn.to_k.bias', 'encoders.4.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.attn.in_proj_weight": ['encoders.4.attn.to_q.weight', 'encoders.4.attn.to_k.weight', 'encoders.4.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.attn.out_proj.bias": "encoders.4.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.attn.out_proj.weight": "encoders.4.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_1.bias": "encoders.4.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_1.weight": "encoders.4.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_2.bias": "encoders.4.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_2.weight": "encoders.4.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_fc.bias": "encoders.4.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_fc.weight": "encoders.4.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_proj.bias": "encoders.4.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_proj.weight": "encoders.4.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.attn.in_proj_bias": ['encoders.5.attn.to_q.bias', 'encoders.5.attn.to_k.bias', 'encoders.5.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.attn.in_proj_weight": ['encoders.5.attn.to_q.weight', 'encoders.5.attn.to_k.weight', 'encoders.5.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.attn.out_proj.bias": "encoders.5.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.attn.out_proj.weight": "encoders.5.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_1.bias": "encoders.5.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_1.weight": "encoders.5.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_2.bias": "encoders.5.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_2.weight": "encoders.5.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_fc.bias": "encoders.5.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_fc.weight": "encoders.5.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_proj.bias": "encoders.5.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_proj.weight": "encoders.5.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.attn.in_proj_bias": ['encoders.6.attn.to_q.bias', 'encoders.6.attn.to_k.bias', 'encoders.6.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.attn.in_proj_weight": ['encoders.6.attn.to_q.weight', 'encoders.6.attn.to_k.weight', 'encoders.6.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.attn.out_proj.bias": "encoders.6.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.attn.out_proj.weight": "encoders.6.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_1.bias": "encoders.6.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_1.weight": "encoders.6.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_2.bias": "encoders.6.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_2.weight": "encoders.6.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_fc.bias": "encoders.6.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_fc.weight": "encoders.6.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_proj.bias": "encoders.6.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_proj.weight": "encoders.6.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.attn.in_proj_bias": ['encoders.7.attn.to_q.bias', 'encoders.7.attn.to_k.bias', 'encoders.7.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.attn.in_proj_weight": ['encoders.7.attn.to_q.weight', 'encoders.7.attn.to_k.weight', 'encoders.7.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.attn.out_proj.bias": "encoders.7.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.attn.out_proj.weight": "encoders.7.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_1.bias": "encoders.7.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_1.weight": "encoders.7.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_2.bias": "encoders.7.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_2.weight": "encoders.7.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_fc.bias": "encoders.7.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_fc.weight": "encoders.7.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_proj.bias": "encoders.7.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_proj.weight": "encoders.7.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.attn.in_proj_bias": ['encoders.8.attn.to_q.bias', 'encoders.8.attn.to_k.bias', 'encoders.8.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.attn.in_proj_weight": ['encoders.8.attn.to_q.weight', 'encoders.8.attn.to_k.weight', 'encoders.8.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.attn.out_proj.bias": "encoders.8.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.attn.out_proj.weight": "encoders.8.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_1.bias": "encoders.8.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_1.weight": "encoders.8.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_2.bias": "encoders.8.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_2.weight": "encoders.8.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_fc.bias": "encoders.8.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_fc.weight": "encoders.8.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_proj.bias": "encoders.8.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_proj.weight": "encoders.8.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.attn.in_proj_bias": ['encoders.9.attn.to_q.bias', 'encoders.9.attn.to_k.bias', 'encoders.9.attn.to_v.bias'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.attn.in_proj_weight": ['encoders.9.attn.to_q.weight', 'encoders.9.attn.to_k.weight', 'encoders.9.attn.to_v.weight'],
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.attn.out_proj.bias": "encoders.9.attn.to_out.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.attn.out_proj.weight": "encoders.9.attn.to_out.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_1.bias": "encoders.9.layer_norm1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_1.weight": "encoders.9.layer_norm1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_2.bias": "encoders.9.layer_norm2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_2.weight": "encoders.9.layer_norm2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_fc.bias": "encoders.9.fc1.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_fc.weight": "encoders.9.fc1.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_proj.bias": "encoders.9.fc2.bias",
|
||||
"conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_proj.weight": "encoders.9.fc2.weight",
|
||||
"conditioner.embedders.0.open_clip.model.visual.proj": "visual_projection.weight",
|
||||
}
|
||||
state_dict_ = {}
|
||||
for name in state_dict:
|
||||
if name in rename_dict:
|
||||
param = state_dict[name]
|
||||
if name == "conditioner.embedders.0.open_clip.model.visual.class_embedding":
|
||||
param = param.reshape((1, 1, param.shape[0]))
|
||||
elif name == "conditioner.embedders.0.open_clip.model.visual.positional_embedding":
|
||||
param = param.reshape((1, param.shape[0], param.shape[1]))
|
||||
elif name == "conditioner.embedders.0.open_clip.model.visual.proj":
|
||||
param = param.T
|
||||
if isinstance(rename_dict[name], str):
|
||||
state_dict_[rename_dict[name]] = param
|
||||
else:
|
||||
length = param.shape[0] // 3
|
||||
for i, rename in enumerate(rename_dict[name]):
|
||||
state_dict_[rename] = param[i*length: i*length+length]
|
||||
return state_dict_
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@@ -2,7 +2,7 @@ import torch
|
||||
from .attention import Attention
|
||||
from .sd_unet import ResnetBlock, UpSampler
|
||||
from .tiler import TileWorker
|
||||
from einops import rearrange
|
||||
from einops import rearrange, repeat
|
||||
|
||||
|
||||
class VAEAttentionBlock(torch.nn.Module):
|
||||
@@ -119,14 +119,13 @@ class SVDVAEDecoder(torch.nn.Module):
|
||||
self.conv_out = torch.nn.Conv2d(128, 3, kernel_size=3, padding=1)
|
||||
self.time_conv_out = torch.nn.Conv3d(3, 3, kernel_size=(3, 1, 1), padding=(1, 0, 0))
|
||||
|
||||
|
||||
def forward(self, sample):
|
||||
# 1. pre-process
|
||||
hidden_states = sample.flatten(0, 1)
|
||||
hidden_states = rearrange(sample, "C T H W -> T C H W")
|
||||
hidden_states = hidden_states / self.scaling_factor
|
||||
hidden_states = self.conv_in(hidden_states)
|
||||
time_emb = None
|
||||
text_emb = None
|
||||
res_stack = None
|
||||
time_emb, text_emb, res_stack = None, None, None
|
||||
|
||||
# 2. blocks
|
||||
for i, block in enumerate(self.blocks):
|
||||
@@ -136,11 +135,70 @@ class SVDVAEDecoder(torch.nn.Module):
|
||||
hidden_states = self.conv_norm_out(hidden_states)
|
||||
hidden_states = self.conv_act(hidden_states)
|
||||
hidden_states = self.conv_out(hidden_states)
|
||||
hidden_states = rearrange(hidden_states, "T C H W -> 1 C T H W")
|
||||
hidden_states = rearrange(hidden_states, "T C H W -> C T H W")
|
||||
hidden_states = self.time_conv_out(hidden_states)
|
||||
|
||||
return hidden_states
|
||||
|
||||
|
||||
def build_mask(self, data, is_bound):
|
||||
_, T, H, W = data.shape
|
||||
t = repeat(torch.arange(T), "T -> T H W", T=T, H=H, W=W)
|
||||
h = repeat(torch.arange(H), "H -> T H W", T=T, H=H, W=W)
|
||||
w = repeat(torch.arange(W), "W -> T H W", T=T, H=H, W=W)
|
||||
border_width = (T + H + W) // 6
|
||||
pad = torch.ones_like(t) * border_width
|
||||
mask = torch.stack([
|
||||
pad if is_bound[0] else t + 1,
|
||||
pad if is_bound[1] else T - t,
|
||||
pad if is_bound[2] else h + 1,
|
||||
pad if is_bound[3] else H - h,
|
||||
pad if is_bound[4] else w + 1,
|
||||
pad if is_bound[5] else W - w
|
||||
]).min(dim=0).values
|
||||
mask = mask.clip(1, border_width)
|
||||
mask = (mask / border_width).to(dtype=data.dtype, device=data.device)
|
||||
mask = rearrange(mask, "T H W -> 1 T H W")
|
||||
return mask
|
||||
|
||||
|
||||
def decode_video(
|
||||
self, sample,
|
||||
batch_time=8, batch_height=128, batch_width=128,
|
||||
stride_time=4, stride_height=32, stride_width=32,
|
||||
progress_bar=lambda x:x
|
||||
):
|
||||
sample = sample.permute(1, 0, 2, 3)
|
||||
data_device = sample.device
|
||||
computation_device = self.conv_in.weight.device
|
||||
torch_dtype = sample.dtype
|
||||
_, T, H, W = sample.shape
|
||||
|
||||
weight = torch.zeros((1, T, H*8, W*8), dtype=torch_dtype, device=data_device)
|
||||
values = torch.zeros((3, T, H*8, W*8), dtype=torch_dtype, device=data_device)
|
||||
|
||||
# Split tasks
|
||||
tasks = []
|
||||
for t in range(0, T, stride_time):
|
||||
for h in range(0, H, stride_height):
|
||||
for w in range(0, W, stride_width):
|
||||
if (t-stride_time >= 0 and t-stride_time+batch_time >= T)\
|
||||
or (h-stride_height >= 0 and h-stride_height+batch_height >= H)\
|
||||
or (w-stride_width >= 0 and w-stride_width+batch_width >= W):
|
||||
continue
|
||||
tasks.append((t, t+batch_time, h, h+batch_height, w, w+batch_width))
|
||||
|
||||
# Run
|
||||
for tl, tr, hl, hr, wl, wr in progress_bar(tasks):
|
||||
sample_batch = sample[:, tl:tr, hl:hr, wl:wr].to(computation_device)
|
||||
sample_batch = self.forward(sample_batch).to(data_device)
|
||||
mask = self.build_mask(sample_batch, is_bound=(tl==0, tr>=T, hl==0, hr>=H, wl==0, wr>=W))
|
||||
values[:, tl:tr, hl*8:hr*8, wl*8:wr*8] += sample_batch * mask
|
||||
weight[:, tl:tr, hl*8:hr*8, wl*8:wr*8] += mask
|
||||
values /= weight
|
||||
return values
|
||||
|
||||
|
||||
def state_dict_converter(self):
|
||||
return SVDVAEDecoderStateDictConverter()
|
||||
|
||||
@@ -238,3 +296,282 @@ class SVDVAEDecoderStateDictConverter:
|
||||
state_dict_[name_] = state_dict[name]
|
||||
|
||||
return state_dict_
|
||||
|
||||
|
||||
def from_civitai(self, state_dict):
|
||||
rename_dict = {
|
||||
"first_stage_model.decoder.conv_in.bias": "conv_in.bias",
|
||||
"first_stage_model.decoder.conv_in.weight": "conv_in.weight",
|
||||
"first_stage_model.decoder.conv_out.bias": "conv_out.bias",
|
||||
"first_stage_model.decoder.conv_out.time_mix_conv.bias": "time_conv_out.bias",
|
||||
"first_stage_model.decoder.conv_out.time_mix_conv.weight": "time_conv_out.weight",
|
||||
"first_stage_model.decoder.conv_out.weight": "conv_out.weight",
|
||||
"first_stage_model.decoder.mid.attn_1.k.bias": "blocks.2.transformer_blocks.0.to_k.bias",
|
||||
"first_stage_model.decoder.mid.attn_1.k.weight": "blocks.2.transformer_blocks.0.to_k.weight",
|
||||
"first_stage_model.decoder.mid.attn_1.norm.bias": "blocks.2.norm.bias",
|
||||
"first_stage_model.decoder.mid.attn_1.norm.weight": "blocks.2.norm.weight",
|
||||
"first_stage_model.decoder.mid.attn_1.proj_out.bias": "blocks.2.transformer_blocks.0.to_out.bias",
|
||||
"first_stage_model.decoder.mid.attn_1.proj_out.weight": "blocks.2.transformer_blocks.0.to_out.weight",
|
||||
"first_stage_model.decoder.mid.attn_1.q.bias": "blocks.2.transformer_blocks.0.to_q.bias",
|
||||
"first_stage_model.decoder.mid.attn_1.q.weight": "blocks.2.transformer_blocks.0.to_q.weight",
|
||||
"first_stage_model.decoder.mid.attn_1.v.bias": "blocks.2.transformer_blocks.0.to_v.bias",
|
||||
"first_stage_model.decoder.mid.attn_1.v.weight": "blocks.2.transformer_blocks.0.to_v.weight",
|
||||
"first_stage_model.decoder.mid.block_1.conv1.bias": "blocks.0.conv1.bias",
|
||||
"first_stage_model.decoder.mid.block_1.conv1.weight": "blocks.0.conv1.weight",
|
||||
"first_stage_model.decoder.mid.block_1.conv2.bias": "blocks.0.conv2.bias",
|
||||
"first_stage_model.decoder.mid.block_1.conv2.weight": "blocks.0.conv2.weight",
|
||||
"first_stage_model.decoder.mid.block_1.mix_factor": "blocks.1.mix_factor",
|
||||
"first_stage_model.decoder.mid.block_1.norm1.bias": "blocks.0.norm1.bias",
|
||||
"first_stage_model.decoder.mid.block_1.norm1.weight": "blocks.0.norm1.weight",
|
||||
"first_stage_model.decoder.mid.block_1.norm2.bias": "blocks.0.norm2.bias",
|
||||
"first_stage_model.decoder.mid.block_1.norm2.weight": "blocks.0.norm2.weight",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.in_layers.0.bias": "blocks.1.norm1.bias",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.in_layers.0.weight": "blocks.1.norm1.weight",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.in_layers.2.bias": "blocks.1.conv1.bias",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.in_layers.2.weight": "blocks.1.conv1.weight",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.out_layers.0.bias": "blocks.1.norm2.bias",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.out_layers.0.weight": "blocks.1.norm2.weight",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.out_layers.3.bias": "blocks.1.conv2.bias",
|
||||
"first_stage_model.decoder.mid.block_1.time_stack.out_layers.3.weight": "blocks.1.conv2.weight",
|
||||
"first_stage_model.decoder.mid.block_2.conv1.bias": "blocks.3.conv1.bias",
|
||||
"first_stage_model.decoder.mid.block_2.conv1.weight": "blocks.3.conv1.weight",
|
||||
"first_stage_model.decoder.mid.block_2.conv2.bias": "blocks.3.conv2.bias",
|
||||
"first_stage_model.decoder.mid.block_2.conv2.weight": "blocks.3.conv2.weight",
|
||||
"first_stage_model.decoder.mid.block_2.mix_factor": "blocks.4.mix_factor",
|
||||
"first_stage_model.decoder.mid.block_2.norm1.bias": "blocks.3.norm1.bias",
|
||||
"first_stage_model.decoder.mid.block_2.norm1.weight": "blocks.3.norm1.weight",
|
||||
"first_stage_model.decoder.mid.block_2.norm2.bias": "blocks.3.norm2.bias",
|
||||
"first_stage_model.decoder.mid.block_2.norm2.weight": "blocks.3.norm2.weight",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.in_layers.0.bias": "blocks.4.norm1.bias",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.in_layers.0.weight": "blocks.4.norm1.weight",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.in_layers.2.bias": "blocks.4.conv1.bias",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.in_layers.2.weight": "blocks.4.conv1.weight",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.out_layers.0.bias": "blocks.4.norm2.bias",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.out_layers.0.weight": "blocks.4.norm2.weight",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.out_layers.3.bias": "blocks.4.conv2.bias",
|
||||
"first_stage_model.decoder.mid.block_2.time_stack.out_layers.3.weight": "blocks.4.conv2.weight",
|
||||
"first_stage_model.decoder.norm_out.bias": "conv_norm_out.bias",
|
||||
"first_stage_model.decoder.norm_out.weight": "conv_norm_out.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.conv1.bias": "blocks.26.conv1.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.conv1.weight": "blocks.26.conv1.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.conv2.bias": "blocks.26.conv2.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.conv2.weight": "blocks.26.conv2.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.mix_factor": "blocks.27.mix_factor",
|
||||
"first_stage_model.decoder.up.0.block.0.nin_shortcut.bias": "blocks.26.conv_shortcut.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.nin_shortcut.weight": "blocks.26.conv_shortcut.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.norm1.bias": "blocks.26.norm1.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.norm1.weight": "blocks.26.norm1.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.norm2.bias": "blocks.26.norm2.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.norm2.weight": "blocks.26.norm2.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.in_layers.0.bias": "blocks.27.norm1.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.in_layers.0.weight": "blocks.27.norm1.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.in_layers.2.bias": "blocks.27.conv1.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.in_layers.2.weight": "blocks.27.conv1.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.out_layers.0.bias": "blocks.27.norm2.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.out_layers.0.weight": "blocks.27.norm2.weight",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.out_layers.3.bias": "blocks.27.conv2.bias",
|
||||
"first_stage_model.decoder.up.0.block.0.time_stack.out_layers.3.weight": "blocks.27.conv2.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.conv1.bias": "blocks.28.conv1.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.conv1.weight": "blocks.28.conv1.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.conv2.bias": "blocks.28.conv2.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.conv2.weight": "blocks.28.conv2.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.mix_factor": "blocks.29.mix_factor",
|
||||
"first_stage_model.decoder.up.0.block.1.norm1.bias": "blocks.28.norm1.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.norm1.weight": "blocks.28.norm1.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.norm2.bias": "blocks.28.norm2.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.norm2.weight": "blocks.28.norm2.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.in_layers.0.bias": "blocks.29.norm1.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.in_layers.0.weight": "blocks.29.norm1.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.in_layers.2.bias": "blocks.29.conv1.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.in_layers.2.weight": "blocks.29.conv1.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.out_layers.0.bias": "blocks.29.norm2.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.out_layers.0.weight": "blocks.29.norm2.weight",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.out_layers.3.bias": "blocks.29.conv2.bias",
|
||||
"first_stage_model.decoder.up.0.block.1.time_stack.out_layers.3.weight": "blocks.29.conv2.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.conv1.bias": "blocks.30.conv1.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.conv1.weight": "blocks.30.conv1.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.conv2.bias": "blocks.30.conv2.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.conv2.weight": "blocks.30.conv2.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.mix_factor": "blocks.31.mix_factor",
|
||||
"first_stage_model.decoder.up.0.block.2.norm1.bias": "blocks.30.norm1.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.norm1.weight": "blocks.30.norm1.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.norm2.bias": "blocks.30.norm2.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.norm2.weight": "blocks.30.norm2.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.in_layers.0.bias": "blocks.31.norm1.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.in_layers.0.weight": "blocks.31.norm1.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.in_layers.2.bias": "blocks.31.conv1.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.in_layers.2.weight": "blocks.31.conv1.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.out_layers.0.bias": "blocks.31.norm2.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.out_layers.0.weight": "blocks.31.norm2.weight",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.out_layers.3.bias": "blocks.31.conv2.bias",
|
||||
"first_stage_model.decoder.up.0.block.2.time_stack.out_layers.3.weight": "blocks.31.conv2.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.conv1.bias": "blocks.19.conv1.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.conv1.weight": "blocks.19.conv1.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.conv2.bias": "blocks.19.conv2.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.conv2.weight": "blocks.19.conv2.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.mix_factor": "blocks.20.mix_factor",
|
||||
"first_stage_model.decoder.up.1.block.0.nin_shortcut.bias": "blocks.19.conv_shortcut.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.nin_shortcut.weight": "blocks.19.conv_shortcut.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.norm1.bias": "blocks.19.norm1.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.norm1.weight": "blocks.19.norm1.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.norm2.bias": "blocks.19.norm2.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.norm2.weight": "blocks.19.norm2.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.in_layers.0.bias": "blocks.20.norm1.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.in_layers.0.weight": "blocks.20.norm1.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.in_layers.2.bias": "blocks.20.conv1.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.in_layers.2.weight": "blocks.20.conv1.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.out_layers.0.bias": "blocks.20.norm2.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.out_layers.0.weight": "blocks.20.norm2.weight",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.out_layers.3.bias": "blocks.20.conv2.bias",
|
||||
"first_stage_model.decoder.up.1.block.0.time_stack.out_layers.3.weight": "blocks.20.conv2.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.conv1.bias": "blocks.21.conv1.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.conv1.weight": "blocks.21.conv1.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.conv2.bias": "blocks.21.conv2.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.conv2.weight": "blocks.21.conv2.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.mix_factor": "blocks.22.mix_factor",
|
||||
"first_stage_model.decoder.up.1.block.1.norm1.bias": "blocks.21.norm1.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.norm1.weight": "blocks.21.norm1.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.norm2.bias": "blocks.21.norm2.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.norm2.weight": "blocks.21.norm2.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.in_layers.0.bias": "blocks.22.norm1.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.in_layers.0.weight": "blocks.22.norm1.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.in_layers.2.bias": "blocks.22.conv1.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.in_layers.2.weight": "blocks.22.conv1.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.out_layers.0.bias": "blocks.22.norm2.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.out_layers.0.weight": "blocks.22.norm2.weight",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.out_layers.3.bias": "blocks.22.conv2.bias",
|
||||
"first_stage_model.decoder.up.1.block.1.time_stack.out_layers.3.weight": "blocks.22.conv2.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.conv1.bias": "blocks.23.conv1.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.conv1.weight": "blocks.23.conv1.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.conv2.bias": "blocks.23.conv2.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.conv2.weight": "blocks.23.conv2.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.mix_factor": "blocks.24.mix_factor",
|
||||
"first_stage_model.decoder.up.1.block.2.norm1.bias": "blocks.23.norm1.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.norm1.weight": "blocks.23.norm1.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.norm2.bias": "blocks.23.norm2.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.norm2.weight": "blocks.23.norm2.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.in_layers.0.bias": "blocks.24.norm1.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.in_layers.0.weight": "blocks.24.norm1.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.in_layers.2.bias": "blocks.24.conv1.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.in_layers.2.weight": "blocks.24.conv1.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.out_layers.0.bias": "blocks.24.norm2.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.out_layers.0.weight": "blocks.24.norm2.weight",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.out_layers.3.bias": "blocks.24.conv2.bias",
|
||||
"first_stage_model.decoder.up.1.block.2.time_stack.out_layers.3.weight": "blocks.24.conv2.weight",
|
||||
"first_stage_model.decoder.up.1.upsample.conv.bias": "blocks.25.conv.bias",
|
||||
"first_stage_model.decoder.up.1.upsample.conv.weight": "blocks.25.conv.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.conv1.bias": "blocks.12.conv1.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.conv1.weight": "blocks.12.conv1.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.conv2.bias": "blocks.12.conv2.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.conv2.weight": "blocks.12.conv2.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.mix_factor": "blocks.13.mix_factor",
|
||||
"first_stage_model.decoder.up.2.block.0.norm1.bias": "blocks.12.norm1.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.norm1.weight": "blocks.12.norm1.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.norm2.bias": "blocks.12.norm2.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.norm2.weight": "blocks.12.norm2.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.in_layers.0.bias": "blocks.13.norm1.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.in_layers.0.weight": "blocks.13.norm1.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.in_layers.2.bias": "blocks.13.conv1.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.in_layers.2.weight": "blocks.13.conv1.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.out_layers.0.bias": "blocks.13.norm2.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.out_layers.0.weight": "blocks.13.norm2.weight",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.out_layers.3.bias": "blocks.13.conv2.bias",
|
||||
"first_stage_model.decoder.up.2.block.0.time_stack.out_layers.3.weight": "blocks.13.conv2.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.conv1.bias": "blocks.14.conv1.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.conv1.weight": "blocks.14.conv1.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.conv2.bias": "blocks.14.conv2.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.conv2.weight": "blocks.14.conv2.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.mix_factor": "blocks.15.mix_factor",
|
||||
"first_stage_model.decoder.up.2.block.1.norm1.bias": "blocks.14.norm1.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.norm1.weight": "blocks.14.norm1.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.norm2.bias": "blocks.14.norm2.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.norm2.weight": "blocks.14.norm2.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.in_layers.0.bias": "blocks.15.norm1.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.in_layers.0.weight": "blocks.15.norm1.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.in_layers.2.bias": "blocks.15.conv1.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.in_layers.2.weight": "blocks.15.conv1.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.out_layers.0.bias": "blocks.15.norm2.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.out_layers.0.weight": "blocks.15.norm2.weight",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.out_layers.3.bias": "blocks.15.conv2.bias",
|
||||
"first_stage_model.decoder.up.2.block.1.time_stack.out_layers.3.weight": "blocks.15.conv2.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.conv1.bias": "blocks.16.conv1.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.conv1.weight": "blocks.16.conv1.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.conv2.bias": "blocks.16.conv2.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.conv2.weight": "blocks.16.conv2.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.mix_factor": "blocks.17.mix_factor",
|
||||
"first_stage_model.decoder.up.2.block.2.norm1.bias": "blocks.16.norm1.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.norm1.weight": "blocks.16.norm1.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.norm2.bias": "blocks.16.norm2.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.norm2.weight": "blocks.16.norm2.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.in_layers.0.bias": "blocks.17.norm1.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.in_layers.0.weight": "blocks.17.norm1.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.in_layers.2.bias": "blocks.17.conv1.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.in_layers.2.weight": "blocks.17.conv1.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.out_layers.0.bias": "blocks.17.norm2.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.out_layers.0.weight": "blocks.17.norm2.weight",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.out_layers.3.bias": "blocks.17.conv2.bias",
|
||||
"first_stage_model.decoder.up.2.block.2.time_stack.out_layers.3.weight": "blocks.17.conv2.weight",
|
||||
"first_stage_model.decoder.up.2.upsample.conv.bias": "blocks.18.conv.bias",
|
||||
"first_stage_model.decoder.up.2.upsample.conv.weight": "blocks.18.conv.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.conv1.bias": "blocks.5.conv1.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.conv1.weight": "blocks.5.conv1.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.conv2.bias": "blocks.5.conv2.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.conv2.weight": "blocks.5.conv2.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.mix_factor": "blocks.6.mix_factor",
|
||||
"first_stage_model.decoder.up.3.block.0.norm1.bias": "blocks.5.norm1.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.norm1.weight": "blocks.5.norm1.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.norm2.bias": "blocks.5.norm2.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.norm2.weight": "blocks.5.norm2.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.in_layers.0.bias": "blocks.6.norm1.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.in_layers.0.weight": "blocks.6.norm1.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.in_layers.2.bias": "blocks.6.conv1.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.in_layers.2.weight": "blocks.6.conv1.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.out_layers.0.bias": "blocks.6.norm2.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.out_layers.0.weight": "blocks.6.norm2.weight",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.out_layers.3.bias": "blocks.6.conv2.bias",
|
||||
"first_stage_model.decoder.up.3.block.0.time_stack.out_layers.3.weight": "blocks.6.conv2.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.conv1.bias": "blocks.7.conv1.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.conv1.weight": "blocks.7.conv1.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.conv2.bias": "blocks.7.conv2.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.conv2.weight": "blocks.7.conv2.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.mix_factor": "blocks.8.mix_factor",
|
||||
"first_stage_model.decoder.up.3.block.1.norm1.bias": "blocks.7.norm1.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.norm1.weight": "blocks.7.norm1.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.norm2.bias": "blocks.7.norm2.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.norm2.weight": "blocks.7.norm2.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.in_layers.0.bias": "blocks.8.norm1.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.in_layers.0.weight": "blocks.8.norm1.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.in_layers.2.bias": "blocks.8.conv1.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.in_layers.2.weight": "blocks.8.conv1.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.out_layers.0.bias": "blocks.8.norm2.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.out_layers.0.weight": "blocks.8.norm2.weight",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.out_layers.3.bias": "blocks.8.conv2.bias",
|
||||
"first_stage_model.decoder.up.3.block.1.time_stack.out_layers.3.weight": "blocks.8.conv2.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.conv1.bias": "blocks.9.conv1.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.conv1.weight": "blocks.9.conv1.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.conv2.bias": "blocks.9.conv2.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.conv2.weight": "blocks.9.conv2.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.mix_factor": "blocks.10.mix_factor",
|
||||
"first_stage_model.decoder.up.3.block.2.norm1.bias": "blocks.9.norm1.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.norm1.weight": "blocks.9.norm1.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.norm2.bias": "blocks.9.norm2.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.norm2.weight": "blocks.9.norm2.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.in_layers.0.bias": "blocks.10.norm1.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.in_layers.0.weight": "blocks.10.norm1.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.in_layers.2.bias": "blocks.10.conv1.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.in_layers.2.weight": "blocks.10.conv1.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.out_layers.0.bias": "blocks.10.norm2.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.out_layers.0.weight": "blocks.10.norm2.weight",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.out_layers.3.bias": "blocks.10.conv2.bias",
|
||||
"first_stage_model.decoder.up.3.block.2.time_stack.out_layers.3.weight": "blocks.10.conv2.weight",
|
||||
"first_stage_model.decoder.up.3.upsample.conv.bias": "blocks.11.conv.bias",
|
||||
"first_stage_model.decoder.up.3.upsample.conv.weight": "blocks.11.conv.weight",
|
||||
}
|
||||
state_dict_ = {}
|
||||
for name in state_dict:
|
||||
if name in rename_dict:
|
||||
param = state_dict[name]
|
||||
if "blocks.2.transformer_blocks.0" in rename_dict[name]:
|
||||
param = param.squeeze()
|
||||
state_dict_[rename_dict[name]] = param
|
||||
return state_dict_
|
||||
|
||||
138
diffsynth/models/svd_vae_encoder.py
Normal file
138
diffsynth/models/svd_vae_encoder.py
Normal file
@@ -0,0 +1,138 @@
|
||||
from .sd_vae_encoder import SDVAEEncoderStateDictConverter, SDVAEEncoder
|
||||
|
||||
|
||||
class SVDVAEEncoder(SDVAEEncoder):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
self.scaling_factor = 0.13025
|
||||
|
||||
def state_dict_converter(self):
|
||||
return SVDVAEEncoderStateDictConverter()
|
||||
|
||||
|
||||
class SVDVAEEncoderStateDictConverter(SDVAEEncoderStateDictConverter):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
|
||||
def from_diffusers(self, state_dict):
|
||||
return super().from_diffusers(state_dict)
|
||||
|
||||
def from_civitai(self, state_dict):
|
||||
rename_dict = {
|
||||
"conditioner.embedders.3.encoder.encoder.conv_in.bias": "conv_in.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.conv_in.weight": "conv_in.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.conv_out.bias": "conv_out.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.conv_out.weight": "conv_out.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.conv1.bias": "blocks.0.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.conv1.weight": "blocks.0.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.conv2.bias": "blocks.0.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.conv2.weight": "blocks.0.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.norm1.bias": "blocks.0.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.norm1.weight": "blocks.0.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.norm2.bias": "blocks.0.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.0.norm2.weight": "blocks.0.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.conv1.bias": "blocks.1.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.conv1.weight": "blocks.1.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.conv2.bias": "blocks.1.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.conv2.weight": "blocks.1.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.norm1.bias": "blocks.1.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.norm1.weight": "blocks.1.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.norm2.bias": "blocks.1.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.block.1.norm2.weight": "blocks.1.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.downsample.conv.bias": "blocks.2.conv.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.0.downsample.conv.weight": "blocks.2.conv.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.conv1.bias": "blocks.3.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.conv1.weight": "blocks.3.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.conv2.bias": "blocks.3.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.conv2.weight": "blocks.3.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.nin_shortcut.bias": "blocks.3.conv_shortcut.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.nin_shortcut.weight": "blocks.3.conv_shortcut.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.norm1.bias": "blocks.3.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.norm1.weight": "blocks.3.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.norm2.bias": "blocks.3.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.0.norm2.weight": "blocks.3.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.conv1.bias": "blocks.4.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.conv1.weight": "blocks.4.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.conv2.bias": "blocks.4.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.conv2.weight": "blocks.4.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.norm1.bias": "blocks.4.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.norm1.weight": "blocks.4.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.norm2.bias": "blocks.4.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.block.1.norm2.weight": "blocks.4.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.downsample.conv.bias": "blocks.5.conv.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.1.downsample.conv.weight": "blocks.5.conv.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.conv1.bias": "blocks.6.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.conv1.weight": "blocks.6.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.conv2.bias": "blocks.6.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.conv2.weight": "blocks.6.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.nin_shortcut.bias": "blocks.6.conv_shortcut.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.nin_shortcut.weight": "blocks.6.conv_shortcut.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.norm1.bias": "blocks.6.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.norm1.weight": "blocks.6.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.norm2.bias": "blocks.6.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.0.norm2.weight": "blocks.6.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.conv1.bias": "blocks.7.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.conv1.weight": "blocks.7.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.conv2.bias": "blocks.7.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.conv2.weight": "blocks.7.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.norm1.bias": "blocks.7.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.norm1.weight": "blocks.7.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.norm2.bias": "blocks.7.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.block.1.norm2.weight": "blocks.7.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.downsample.conv.bias": "blocks.8.conv.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.2.downsample.conv.weight": "blocks.8.conv.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.conv1.bias": "blocks.9.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.conv1.weight": "blocks.9.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.conv2.bias": "blocks.9.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.conv2.weight": "blocks.9.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.norm1.bias": "blocks.9.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.norm1.weight": "blocks.9.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.norm2.bias": "blocks.9.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.0.norm2.weight": "blocks.9.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.conv1.bias": "blocks.10.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.conv1.weight": "blocks.10.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.conv2.bias": "blocks.10.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.conv2.weight": "blocks.10.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.norm1.bias": "blocks.10.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.norm1.weight": "blocks.10.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.norm2.bias": "blocks.10.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.down.3.block.1.norm2.weight": "blocks.10.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.k.bias": "blocks.12.transformer_blocks.0.to_k.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.k.weight": "blocks.12.transformer_blocks.0.to_k.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.norm.bias": "blocks.12.norm.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.norm.weight": "blocks.12.norm.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.proj_out.bias": "blocks.12.transformer_blocks.0.to_out.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.proj_out.weight": "blocks.12.transformer_blocks.0.to_out.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.q.bias": "blocks.12.transformer_blocks.0.to_q.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.q.weight": "blocks.12.transformer_blocks.0.to_q.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.v.bias": "blocks.12.transformer_blocks.0.to_v.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.attn_1.v.weight": "blocks.12.transformer_blocks.0.to_v.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.conv1.bias": "blocks.11.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.conv1.weight": "blocks.11.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.conv2.bias": "blocks.11.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.conv2.weight": "blocks.11.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.norm1.bias": "blocks.11.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.norm1.weight": "blocks.11.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.norm2.bias": "blocks.11.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_1.norm2.weight": "blocks.11.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.conv1.bias": "blocks.13.conv1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.conv1.weight": "blocks.13.conv1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.conv2.bias": "blocks.13.conv2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.conv2.weight": "blocks.13.conv2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.norm1.bias": "blocks.13.norm1.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.norm1.weight": "blocks.13.norm1.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.norm2.bias": "blocks.13.norm2.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.mid.block_2.norm2.weight": "blocks.13.norm2.weight",
|
||||
"conditioner.embedders.3.encoder.encoder.norm_out.bias": "conv_norm_out.bias",
|
||||
"conditioner.embedders.3.encoder.encoder.norm_out.weight": "conv_norm_out.weight",
|
||||
"conditioner.embedders.3.encoder.quant_conv.bias": "quant_conv.bias",
|
||||
"conditioner.embedders.3.encoder.quant_conv.weight": "quant_conv.weight",
|
||||
}
|
||||
state_dict_ = {}
|
||||
for name in state_dict:
|
||||
if name in rename_dict:
|
||||
param = state_dict[name]
|
||||
if "transformer_blocks" in rename_dict[name]:
|
||||
param = param.squeeze()
|
||||
state_dict_[rename_dict[name]] = param
|
||||
return state_dict_
|
||||
Reference in New Issue
Block a user