modelscope/DiffSynth-Painter
14
1import torch2from .sd_text_encoder import CLIPEncoderLayer3 4 5class CLIPVisionEmbeddings(torch.nn.Module):6 def __init__(self, embed_dim=1280, image_size=224, patch_size=14, num_channels=3):7 super().__init__()8 9 # class_embeds (This is a fixed tensor)10 self.class_embedding = torch.nn.Parameter(torch.randn(1, 1, embed_dim))11 12 # position_embeds13 self.patch_embedding = torch.nn.Conv2d(in_channels=num_channels, out_channels=embed_dim, kernel_size=patch_size, stride=patch_size, bias=False)14 15 # position_embeds (This is a fixed tensor)16 self.position_embeds = torch.nn.Parameter(torch.zeros(1, (image_size // patch_size) ** 2 + 1, embed_dim))17 18 def forward(self, pixel_values):19 batch_size = pixel_values.shape[0]20 patch_embeds = self.patch_embedding(pixel_values)21 patch_embeds = patch_embeds.flatten(2).transpose(1, 2)22 class_embeds = self.class_embedding.repeat(batch_size, 1, 1)23 embeddings = torch.cat([class_embeds, patch_embeds], dim=1) + self.position_embeds24 return embeddings25 26 27class SVDImageEncoder(torch.nn.Module):28 def __init__(self, embed_dim=1280, layer_norm_eps=1e-5, num_encoder_layers=32, encoder_intermediate_size=5120, projection_dim=1024, num_heads=16, head_dim=80):29 super().__init__()30 self.embeddings = CLIPVisionEmbeddings(embed_dim=embed_dim)31 self.pre_layernorm = torch.nn.LayerNorm(embed_dim, eps=layer_norm_eps)32 self.encoders = torch.nn.ModuleList([33 CLIPEncoderLayer(embed_dim, encoder_intermediate_size, num_heads=num_heads, head_dim=head_dim, use_quick_gelu=False)34 for _ in range(num_encoder_layers)])35 self.post_layernorm = torch.nn.LayerNorm(embed_dim, eps=layer_norm_eps)36 self.visual_projection = torch.nn.Linear(embed_dim, projection_dim, bias=False)37 38 def forward(self, pixel_values):39 embeds = self.embeddings(pixel_values)40 embeds = self.pre_layernorm(embeds)41 for encoder_id, encoder in enumerate(self.encoders):42 embeds = encoder(embeds)43 embeds = self.post_layernorm(embeds[:, 0, :])44 embeds = self.visual_projection(embeds)45 return embeds46 47 @staticmethod48 def state_dict_converter():49 return SVDImageEncoderStateDictConverter()50 51 52class SVDImageEncoderStateDictConverter:53 def __init__(self):54 pass55 56 def from_diffusers(self, state_dict):57 rename_dict = {58 "vision_model.embeddings.patch_embedding.weight": "embeddings.patch_embedding.weight",59 "vision_model.embeddings.class_embedding": "embeddings.class_embedding",60 "vision_model.embeddings.position_embedding.weight": "embeddings.position_embeds",61 "vision_model.pre_layrnorm.weight": "pre_layernorm.weight",62 "vision_model.pre_layrnorm.bias": "pre_layernorm.bias",63 "vision_model.post_layernorm.weight": "post_layernorm.weight",64 "vision_model.post_layernorm.bias": "post_layernorm.bias",65 "visual_projection.weight": "visual_projection.weight"66 }67 attn_rename_dict = {68 "self_attn.q_proj": "attn.to_q",69 "self_attn.k_proj": "attn.to_k",70 "self_attn.v_proj": "attn.to_v",71 "self_attn.out_proj": "attn.to_out",72 "layer_norm1": "layer_norm1",73 "layer_norm2": "layer_norm2",74 "mlp.fc1": "fc1",75 "mlp.fc2": "fc2",76 }77 state_dict_ = {}78 for name in state_dict:79 if name in rename_dict:80 param = state_dict[name]81 if name == "vision_model.embeddings.class_embedding":82 param = state_dict[name].view(1, 1, -1)83 elif name == "vision_model.embeddings.position_embedding.weight":84 param = state_dict[name].unsqueeze(0)85 state_dict_[rename_dict[name]] = param86 elif name.startswith("vision_model.encoder.layers."):87 param = state_dict[name]88 names = name.split(".")89 layer_id, layer_type, tail = names[3], ".".join(names[4:-1]), names[-1]90 name_ = ".".join(["encoders", layer_id, attn_rename_dict[layer_type], tail])91 state_dict_[name_] = param92 return state_dict_93 94 def from_civitai(self, state_dict):95 rename_dict = {96 "conditioner.embedders.0.open_clip.model.visual.class_embedding": "embeddings.class_embedding",97 "conditioner.embedders.0.open_clip.model.visual.conv1.weight": "embeddings.patch_embedding.weight",98 "conditioner.embedders.0.open_clip.model.visual.ln_post.bias": "post_layernorm.bias",99 "conditioner.embedders.0.open_clip.model.visual.ln_post.weight": "post_layernorm.weight",100 "conditioner.embedders.0.open_clip.model.visual.ln_pre.bias": "pre_layernorm.bias",101 "conditioner.embedders.0.open_clip.model.visual.ln_pre.weight": "pre_layernorm.weight",102 "conditioner.embedders.0.open_clip.model.visual.positional_embedding": "embeddings.position_embeds",103 "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'],104 "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'],105 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.attn.out_proj.bias": "encoders.0.attn.to_out.bias",106 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.attn.out_proj.weight": "encoders.0.attn.to_out.weight",107 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_1.bias": "encoders.0.layer_norm1.bias",108 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_1.weight": "encoders.0.layer_norm1.weight",109 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_2.bias": "encoders.0.layer_norm2.bias",110 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.ln_2.weight": "encoders.0.layer_norm2.weight",111 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_fc.bias": "encoders.0.fc1.bias",112 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_fc.weight": "encoders.0.fc1.weight",113 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_proj.bias": "encoders.0.fc2.bias",114 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.0.mlp.c_proj.weight": "encoders.0.fc2.weight",115 "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'],116 "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'],117 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.attn.out_proj.bias": "encoders.1.attn.to_out.bias",118 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.attn.out_proj.weight": "encoders.1.attn.to_out.weight",119 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_1.bias": "encoders.1.layer_norm1.bias",120 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_1.weight": "encoders.1.layer_norm1.weight",121 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_2.bias": "encoders.1.layer_norm2.bias",122 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.ln_2.weight": "encoders.1.layer_norm2.weight",123 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_fc.bias": "encoders.1.fc1.bias",124 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_fc.weight": "encoders.1.fc1.weight",125 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_proj.bias": "encoders.1.fc2.bias",126 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.1.mlp.c_proj.weight": "encoders.1.fc2.weight",127 "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'],128 "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'],129 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.attn.out_proj.bias": "encoders.10.attn.to_out.bias",130 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.attn.out_proj.weight": "encoders.10.attn.to_out.weight",131 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_1.bias": "encoders.10.layer_norm1.bias",132 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_1.weight": "encoders.10.layer_norm1.weight",133 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_2.bias": "encoders.10.layer_norm2.bias",134 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.ln_2.weight": "encoders.10.layer_norm2.weight",135 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_fc.bias": "encoders.10.fc1.bias",136 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_fc.weight": "encoders.10.fc1.weight",137 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_proj.bias": "encoders.10.fc2.bias",138 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.10.mlp.c_proj.weight": "encoders.10.fc2.weight",139 "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'],140 "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'],141 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.attn.out_proj.bias": "encoders.11.attn.to_out.bias",142 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.attn.out_proj.weight": "encoders.11.attn.to_out.weight",143 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_1.bias": "encoders.11.layer_norm1.bias",144 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_1.weight": "encoders.11.layer_norm1.weight",145 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_2.bias": "encoders.11.layer_norm2.bias",146 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.ln_2.weight": "encoders.11.layer_norm2.weight",147 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_fc.bias": "encoders.11.fc1.bias",148 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_fc.weight": "encoders.11.fc1.weight",149 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_proj.bias": "encoders.11.fc2.bias",150 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.11.mlp.c_proj.weight": "encoders.11.fc2.weight",151 "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'],152 "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'],153 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.attn.out_proj.bias": "encoders.12.attn.to_out.bias",154 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.attn.out_proj.weight": "encoders.12.attn.to_out.weight",155 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_1.bias": "encoders.12.layer_norm1.bias",156 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_1.weight": "encoders.12.layer_norm1.weight",157 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_2.bias": "encoders.12.layer_norm2.bias",158 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.ln_2.weight": "encoders.12.layer_norm2.weight",159 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_fc.bias": "encoders.12.fc1.bias",160 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_fc.weight": "encoders.12.fc1.weight",161 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_proj.bias": "encoders.12.fc2.bias",162 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.12.mlp.c_proj.weight": "encoders.12.fc2.weight",163 "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'],164 "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'],165 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.attn.out_proj.bias": "encoders.13.attn.to_out.bias",166 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.attn.out_proj.weight": "encoders.13.attn.to_out.weight",167 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_1.bias": "encoders.13.layer_norm1.bias",168 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_1.weight": "encoders.13.layer_norm1.weight",169 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_2.bias": "encoders.13.layer_norm2.bias",170 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.ln_2.weight": "encoders.13.layer_norm2.weight",171 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_fc.bias": "encoders.13.fc1.bias",172 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_fc.weight": "encoders.13.fc1.weight",173 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_proj.bias": "encoders.13.fc2.bias",174 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.13.mlp.c_proj.weight": "encoders.13.fc2.weight",175 "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'],176 "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'],177 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.attn.out_proj.bias": "encoders.14.attn.to_out.bias",178 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.attn.out_proj.weight": "encoders.14.attn.to_out.weight",179 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_1.bias": "encoders.14.layer_norm1.bias",180 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_1.weight": "encoders.14.layer_norm1.weight",181 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_2.bias": "encoders.14.layer_norm2.bias",182 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.ln_2.weight": "encoders.14.layer_norm2.weight",183 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_fc.bias": "encoders.14.fc1.bias",184 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_fc.weight": "encoders.14.fc1.weight",185 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_proj.bias": "encoders.14.fc2.bias",186 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.14.mlp.c_proj.weight": "encoders.14.fc2.weight",187 "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'],188 "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'],189 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.attn.out_proj.bias": "encoders.15.attn.to_out.bias",190 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.attn.out_proj.weight": "encoders.15.attn.to_out.weight",191 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_1.bias": "encoders.15.layer_norm1.bias",192 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_1.weight": "encoders.15.layer_norm1.weight",193 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_2.bias": "encoders.15.layer_norm2.bias",194 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.ln_2.weight": "encoders.15.layer_norm2.weight",195 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_fc.bias": "encoders.15.fc1.bias",196 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_fc.weight": "encoders.15.fc1.weight",197 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_proj.bias": "encoders.15.fc2.bias",198 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.15.mlp.c_proj.weight": "encoders.15.fc2.weight",199 "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'],200 "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'],201 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.attn.out_proj.bias": "encoders.16.attn.to_out.bias",202 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.attn.out_proj.weight": "encoders.16.attn.to_out.weight",203 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_1.bias": "encoders.16.layer_norm1.bias",204 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_1.weight": "encoders.16.layer_norm1.weight",205 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_2.bias": "encoders.16.layer_norm2.bias",206 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.ln_2.weight": "encoders.16.layer_norm2.weight",207 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_fc.bias": "encoders.16.fc1.bias",208 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_fc.weight": "encoders.16.fc1.weight",209 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_proj.bias": "encoders.16.fc2.bias",210 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.16.mlp.c_proj.weight": "encoders.16.fc2.weight",211 "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'],212 "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'],213 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.attn.out_proj.bias": "encoders.17.attn.to_out.bias",214 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.attn.out_proj.weight": "encoders.17.attn.to_out.weight",215 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_1.bias": "encoders.17.layer_norm1.bias",216 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_1.weight": "encoders.17.layer_norm1.weight",217 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_2.bias": "encoders.17.layer_norm2.bias",218 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.ln_2.weight": "encoders.17.layer_norm2.weight",219 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_fc.bias": "encoders.17.fc1.bias",220 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_fc.weight": "encoders.17.fc1.weight",221 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_proj.bias": "encoders.17.fc2.bias",222 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.17.mlp.c_proj.weight": "encoders.17.fc2.weight",223 "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'],224 "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'],225 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.attn.out_proj.bias": "encoders.18.attn.to_out.bias",226 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.attn.out_proj.weight": "encoders.18.attn.to_out.weight",227 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_1.bias": "encoders.18.layer_norm1.bias",228 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_1.weight": "encoders.18.layer_norm1.weight",229 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_2.bias": "encoders.18.layer_norm2.bias",230 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.ln_2.weight": "encoders.18.layer_norm2.weight",231 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_fc.bias": "encoders.18.fc1.bias",232 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_fc.weight": "encoders.18.fc1.weight",233 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_proj.bias": "encoders.18.fc2.bias",234 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.18.mlp.c_proj.weight": "encoders.18.fc2.weight",235 "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'],236 "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'],237 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.attn.out_proj.bias": "encoders.19.attn.to_out.bias",238 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.attn.out_proj.weight": "encoders.19.attn.to_out.weight",239 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_1.bias": "encoders.19.layer_norm1.bias",240 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_1.weight": "encoders.19.layer_norm1.weight",241 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_2.bias": "encoders.19.layer_norm2.bias",242 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.ln_2.weight": "encoders.19.layer_norm2.weight",243 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_fc.bias": "encoders.19.fc1.bias",244 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_fc.weight": "encoders.19.fc1.weight",245 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_proj.bias": "encoders.19.fc2.bias",246 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.19.mlp.c_proj.weight": "encoders.19.fc2.weight",247 "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'],248 "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'],249 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.attn.out_proj.bias": "encoders.2.attn.to_out.bias",250 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.attn.out_proj.weight": "encoders.2.attn.to_out.weight",251 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_1.bias": "encoders.2.layer_norm1.bias",252 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_1.weight": "encoders.2.layer_norm1.weight",253 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_2.bias": "encoders.2.layer_norm2.bias",254 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.ln_2.weight": "encoders.2.layer_norm2.weight",255 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_fc.bias": "encoders.2.fc1.bias",256 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_fc.weight": "encoders.2.fc1.weight",257 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_proj.bias": "encoders.2.fc2.bias",258 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.2.mlp.c_proj.weight": "encoders.2.fc2.weight",259 "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'],260 "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'],261 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.attn.out_proj.bias": "encoders.20.attn.to_out.bias",262 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.attn.out_proj.weight": "encoders.20.attn.to_out.weight",263 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_1.bias": "encoders.20.layer_norm1.bias",264 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_1.weight": "encoders.20.layer_norm1.weight",265 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_2.bias": "encoders.20.layer_norm2.bias",266 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.ln_2.weight": "encoders.20.layer_norm2.weight",267 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_fc.bias": "encoders.20.fc1.bias",268 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_fc.weight": "encoders.20.fc1.weight",269 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_proj.bias": "encoders.20.fc2.bias",270 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.20.mlp.c_proj.weight": "encoders.20.fc2.weight",271 "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'],272 "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'],273 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.attn.out_proj.bias": "encoders.21.attn.to_out.bias",274 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.attn.out_proj.weight": "encoders.21.attn.to_out.weight",275 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_1.bias": "encoders.21.layer_norm1.bias",276 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_1.weight": "encoders.21.layer_norm1.weight",277 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_2.bias": "encoders.21.layer_norm2.bias",278 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.ln_2.weight": "encoders.21.layer_norm2.weight",279 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_fc.bias": "encoders.21.fc1.bias",280 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_fc.weight": "encoders.21.fc1.weight",281 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_proj.bias": "encoders.21.fc2.bias",282 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.21.mlp.c_proj.weight": "encoders.21.fc2.weight",283 "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'],284 "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'],285 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.attn.out_proj.bias": "encoders.22.attn.to_out.bias",286 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.attn.out_proj.weight": "encoders.22.attn.to_out.weight",287 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_1.bias": "encoders.22.layer_norm1.bias",288 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_1.weight": "encoders.22.layer_norm1.weight",289 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_2.bias": "encoders.22.layer_norm2.bias",290 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.ln_2.weight": "encoders.22.layer_norm2.weight",291 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_fc.bias": "encoders.22.fc1.bias",292 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_fc.weight": "encoders.22.fc1.weight",293 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_proj.bias": "encoders.22.fc2.bias",294 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.22.mlp.c_proj.weight": "encoders.22.fc2.weight",295 "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'],296 "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'],297 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.attn.out_proj.bias": "encoders.23.attn.to_out.bias",298 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.attn.out_proj.weight": "encoders.23.attn.to_out.weight",299 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_1.bias": "encoders.23.layer_norm1.bias",300 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_1.weight": "encoders.23.layer_norm1.weight",301 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_2.bias": "encoders.23.layer_norm2.bias",302 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.ln_2.weight": "encoders.23.layer_norm2.weight",303 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_fc.bias": "encoders.23.fc1.bias",304 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_fc.weight": "encoders.23.fc1.weight",305 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_proj.bias": "encoders.23.fc2.bias",306 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.23.mlp.c_proj.weight": "encoders.23.fc2.weight",307 "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'],308 "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'],309 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.attn.out_proj.bias": "encoders.24.attn.to_out.bias",310 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.attn.out_proj.weight": "encoders.24.attn.to_out.weight",311 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_1.bias": "encoders.24.layer_norm1.bias",312 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_1.weight": "encoders.24.layer_norm1.weight",313 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_2.bias": "encoders.24.layer_norm2.bias",314 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.ln_2.weight": "encoders.24.layer_norm2.weight",315 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_fc.bias": "encoders.24.fc1.bias",316 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_fc.weight": "encoders.24.fc1.weight",317 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_proj.bias": "encoders.24.fc2.bias",318 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.24.mlp.c_proj.weight": "encoders.24.fc2.weight",319 "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'],320 "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'],321 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.attn.out_proj.bias": "encoders.25.attn.to_out.bias",322 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.attn.out_proj.weight": "encoders.25.attn.to_out.weight",323 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_1.bias": "encoders.25.layer_norm1.bias",324 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_1.weight": "encoders.25.layer_norm1.weight",325 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_2.bias": "encoders.25.layer_norm2.bias",326 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.ln_2.weight": "encoders.25.layer_norm2.weight",327 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_fc.bias": "encoders.25.fc1.bias",328 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_fc.weight": "encoders.25.fc1.weight",329 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_proj.bias": "encoders.25.fc2.bias",330 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.25.mlp.c_proj.weight": "encoders.25.fc2.weight",331 "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'],332 "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'],333 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.attn.out_proj.bias": "encoders.26.attn.to_out.bias",334 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.attn.out_proj.weight": "encoders.26.attn.to_out.weight",335 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_1.bias": "encoders.26.layer_norm1.bias",336 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_1.weight": "encoders.26.layer_norm1.weight",337 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_2.bias": "encoders.26.layer_norm2.bias",338 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.ln_2.weight": "encoders.26.layer_norm2.weight",339 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_fc.bias": "encoders.26.fc1.bias",340 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_fc.weight": "encoders.26.fc1.weight",341 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_proj.bias": "encoders.26.fc2.bias",342 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.26.mlp.c_proj.weight": "encoders.26.fc2.weight",343 "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'],344 "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'],345 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.attn.out_proj.bias": "encoders.27.attn.to_out.bias",346 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.attn.out_proj.weight": "encoders.27.attn.to_out.weight",347 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_1.bias": "encoders.27.layer_norm1.bias",348 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_1.weight": "encoders.27.layer_norm1.weight",349 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_2.bias": "encoders.27.layer_norm2.bias",350 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.ln_2.weight": "encoders.27.layer_norm2.weight",351 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_fc.bias": "encoders.27.fc1.bias",352 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_fc.weight": "encoders.27.fc1.weight",353 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_proj.bias": "encoders.27.fc2.bias",354 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.27.mlp.c_proj.weight": "encoders.27.fc2.weight",355 "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'],356 "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'],357 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.attn.out_proj.bias": "encoders.28.attn.to_out.bias",358 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.attn.out_proj.weight": "encoders.28.attn.to_out.weight",359 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_1.bias": "encoders.28.layer_norm1.bias",360 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_1.weight": "encoders.28.layer_norm1.weight",361 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_2.bias": "encoders.28.layer_norm2.bias",362 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.ln_2.weight": "encoders.28.layer_norm2.weight",363 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_fc.bias": "encoders.28.fc1.bias",364 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_fc.weight": "encoders.28.fc1.weight",365 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_proj.bias": "encoders.28.fc2.bias",366 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.28.mlp.c_proj.weight": "encoders.28.fc2.weight",367 "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'],368 "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'],369 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.attn.out_proj.bias": "encoders.29.attn.to_out.bias",370 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.attn.out_proj.weight": "encoders.29.attn.to_out.weight",371 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_1.bias": "encoders.29.layer_norm1.bias",372 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_1.weight": "encoders.29.layer_norm1.weight",373 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_2.bias": "encoders.29.layer_norm2.bias",374 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.ln_2.weight": "encoders.29.layer_norm2.weight",375 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_fc.bias": "encoders.29.fc1.bias",376 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_fc.weight": "encoders.29.fc1.weight",377 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_proj.bias": "encoders.29.fc2.bias",378 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.29.mlp.c_proj.weight": "encoders.29.fc2.weight",379 "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'],380 "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'],381 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.attn.out_proj.bias": "encoders.3.attn.to_out.bias",382 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.attn.out_proj.weight": "encoders.3.attn.to_out.weight",383 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_1.bias": "encoders.3.layer_norm1.bias",384 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_1.weight": "encoders.3.layer_norm1.weight",385 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_2.bias": "encoders.3.layer_norm2.bias",386 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.ln_2.weight": "encoders.3.layer_norm2.weight",387 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_fc.bias": "encoders.3.fc1.bias",388 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_fc.weight": "encoders.3.fc1.weight",389 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_proj.bias": "encoders.3.fc2.bias",390 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.3.mlp.c_proj.weight": "encoders.3.fc2.weight",391 "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'],392 "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'],393 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.attn.out_proj.bias": "encoders.30.attn.to_out.bias",394 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.attn.out_proj.weight": "encoders.30.attn.to_out.weight",395 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_1.bias": "encoders.30.layer_norm1.bias",396 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_1.weight": "encoders.30.layer_norm1.weight",397 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_2.bias": "encoders.30.layer_norm2.bias",398 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.ln_2.weight": "encoders.30.layer_norm2.weight",399 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_fc.bias": "encoders.30.fc1.bias",400 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_fc.weight": "encoders.30.fc1.weight",401 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_proj.bias": "encoders.30.fc2.bias",402 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.30.mlp.c_proj.weight": "encoders.30.fc2.weight",403 "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'],404 "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'],405 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.attn.out_proj.bias": "encoders.31.attn.to_out.bias",406 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.attn.out_proj.weight": "encoders.31.attn.to_out.weight",407 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_1.bias": "encoders.31.layer_norm1.bias",408 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_1.weight": "encoders.31.layer_norm1.weight",409 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_2.bias": "encoders.31.layer_norm2.bias",410 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.ln_2.weight": "encoders.31.layer_norm2.weight",411 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_fc.bias": "encoders.31.fc1.bias",412 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_fc.weight": "encoders.31.fc1.weight",413 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_proj.bias": "encoders.31.fc2.bias",414 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.31.mlp.c_proj.weight": "encoders.31.fc2.weight",415 "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'],416 "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'],417 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.attn.out_proj.bias": "encoders.4.attn.to_out.bias",418 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.attn.out_proj.weight": "encoders.4.attn.to_out.weight",419 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_1.bias": "encoders.4.layer_norm1.bias",420 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_1.weight": "encoders.4.layer_norm1.weight",421 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_2.bias": "encoders.4.layer_norm2.bias",422 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.ln_2.weight": "encoders.4.layer_norm2.weight",423 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_fc.bias": "encoders.4.fc1.bias",424 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_fc.weight": "encoders.4.fc1.weight",425 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_proj.bias": "encoders.4.fc2.bias",426 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.4.mlp.c_proj.weight": "encoders.4.fc2.weight",427 "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'],428 "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'],429 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.attn.out_proj.bias": "encoders.5.attn.to_out.bias",430 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.attn.out_proj.weight": "encoders.5.attn.to_out.weight",431 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_1.bias": "encoders.5.layer_norm1.bias",432 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_1.weight": "encoders.5.layer_norm1.weight",433 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_2.bias": "encoders.5.layer_norm2.bias",434 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.ln_2.weight": "encoders.5.layer_norm2.weight",435 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_fc.bias": "encoders.5.fc1.bias",436 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_fc.weight": "encoders.5.fc1.weight",437 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_proj.bias": "encoders.5.fc2.bias",438 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.5.mlp.c_proj.weight": "encoders.5.fc2.weight",439 "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'],440 "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'],441 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.attn.out_proj.bias": "encoders.6.attn.to_out.bias",442 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.attn.out_proj.weight": "encoders.6.attn.to_out.weight",443 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_1.bias": "encoders.6.layer_norm1.bias",444 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_1.weight": "encoders.6.layer_norm1.weight",445 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_2.bias": "encoders.6.layer_norm2.bias",446 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.ln_2.weight": "encoders.6.layer_norm2.weight",447 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_fc.bias": "encoders.6.fc1.bias",448 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_fc.weight": "encoders.6.fc1.weight",449 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_proj.bias": "encoders.6.fc2.bias",450 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.6.mlp.c_proj.weight": "encoders.6.fc2.weight",451 "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'],452 "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'],453 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.attn.out_proj.bias": "encoders.7.attn.to_out.bias",454 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.attn.out_proj.weight": "encoders.7.attn.to_out.weight",455 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_1.bias": "encoders.7.layer_norm1.bias",456 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_1.weight": "encoders.7.layer_norm1.weight",457 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_2.bias": "encoders.7.layer_norm2.bias",458 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.ln_2.weight": "encoders.7.layer_norm2.weight",459 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_fc.bias": "encoders.7.fc1.bias",460 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_fc.weight": "encoders.7.fc1.weight",461 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_proj.bias": "encoders.7.fc2.bias",462 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.7.mlp.c_proj.weight": "encoders.7.fc2.weight",463 "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'],464 "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'],465 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.attn.out_proj.bias": "encoders.8.attn.to_out.bias",466 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.attn.out_proj.weight": "encoders.8.attn.to_out.weight",467 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_1.bias": "encoders.8.layer_norm1.bias",468 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_1.weight": "encoders.8.layer_norm1.weight",469 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_2.bias": "encoders.8.layer_norm2.bias",470 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.ln_2.weight": "encoders.8.layer_norm2.weight",471 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_fc.bias": "encoders.8.fc1.bias",472 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_fc.weight": "encoders.8.fc1.weight",473 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_proj.bias": "encoders.8.fc2.bias",474 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.8.mlp.c_proj.weight": "encoders.8.fc2.weight",475 "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'],476 "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'],477 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.attn.out_proj.bias": "encoders.9.attn.to_out.bias",478 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.attn.out_proj.weight": "encoders.9.attn.to_out.weight",479 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_1.bias": "encoders.9.layer_norm1.bias",480 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_1.weight": "encoders.9.layer_norm1.weight",481 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_2.bias": "encoders.9.layer_norm2.bias",482 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.ln_2.weight": "encoders.9.layer_norm2.weight",483 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_fc.bias": "encoders.9.fc1.bias",484 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_fc.weight": "encoders.9.fc1.weight",485 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_proj.bias": "encoders.9.fc2.bias",486 "conditioner.embedders.0.open_clip.model.visual.transformer.resblocks.9.mlp.c_proj.weight": "encoders.9.fc2.weight",487 "conditioner.embedders.0.open_clip.model.visual.proj": "visual_projection.weight",488 }489 state_dict_ = {}490 for name in state_dict:491 if name in rename_dict:492 param = state_dict[name]493 if name == "conditioner.embedders.0.open_clip.model.visual.class_embedding":494 param = param.reshape((1, 1, param.shape[0]))495 elif name == "conditioner.embedders.0.open_clip.model.visual.positional_embedding":496 param = param.reshape((1, param.shape[0], param.shape[1]))497 elif name == "conditioner.embedders.0.open_clip.model.visual.proj":498 param = param.T499 if isinstance(rename_dict[name], str):500 state_dict_[rename_dict[name]] = param501 else:502 length = param.shape[0] // 3503 for i, rename in enumerate(rename_dict[name]):504 state_dict_[rename] = param[i*length: i*length+length]505 return state_dict_506