modelscope/DiffSynth-Painter

1import torch2from .sd_text_encoder import CLIPEncoderLayer3    4 5class SDXLTextEncoder(torch.nn.Module):6    def __init__(self, embed_dim=768, vocab_size=49408, max_position_embeddings=77, num_encoder_layers=11, encoder_intermediate_size=3072):7        super().__init__()8 9        # token_embedding10        self.token_embedding = torch.nn.Embedding(vocab_size, embed_dim)11 12        # position_embeds (This is a fixed tensor)13        self.position_embeds = torch.nn.Parameter(torch.zeros(1, max_position_embeddings, embed_dim))14 15        # encoders16        self.encoders = torch.nn.ModuleList([CLIPEncoderLayer(embed_dim, encoder_intermediate_size) for _ in range(num_encoder_layers)])17 18        # attn_mask19        self.attn_mask = self.attention_mask(max_position_embeddings)20 21        # The text encoder is different to that in Stable Diffusion 1.x.22        # It does not include final_layer_norm.23 24    def attention_mask(self, length):25        mask = torch.empty(length, length)26        mask.fill_(float("-inf"))27        mask.triu_(1)28        return mask29 30    def forward(self, input_ids, clip_skip=1):31        embeds = self.token_embedding(input_ids) + self.position_embeds32        attn_mask = self.attn_mask.to(device=embeds.device, dtype=embeds.dtype)33        for encoder_id, encoder in enumerate(self.encoders):34            embeds = encoder(embeds, attn_mask=attn_mask)35            if encoder_id + clip_skip == len(self.encoders):36                break37        return embeds38    39    @staticmethod40    def state_dict_converter():41        return SDXLTextEncoderStateDictConverter()42    43 44class SDXLTextEncoder2(torch.nn.Module):45    def __init__(self, embed_dim=1280, vocab_size=49408, max_position_embeddings=77, num_encoder_layers=32, encoder_intermediate_size=5120):46        super().__init__()47 48        # token_embedding49        self.token_embedding = torch.nn.Embedding(vocab_size, embed_dim)50 51        # position_embeds (This is a fixed tensor)52        self.position_embeds = torch.nn.Parameter(torch.zeros(1, max_position_embeddings, embed_dim))53 54        # encoders55        self.encoders = torch.nn.ModuleList([CLIPEncoderLayer(embed_dim, encoder_intermediate_size, num_heads=20, head_dim=64, use_quick_gelu=False) for _ in range(num_encoder_layers)])56 57        # attn_mask58        self.attn_mask = self.attention_mask(max_position_embeddings)59 60        # final_layer_norm61        self.final_layer_norm = torch.nn.LayerNorm(embed_dim)62 63        # text_projection64        self.text_projection = torch.nn.Linear(embed_dim, embed_dim, bias=False)65 66    def attention_mask(self, length):67        mask = torch.empty(length, length)68        mask.fill_(float("-inf"))69        mask.triu_(1)70        return mask71 72    def forward(self, input_ids, clip_skip=2):73        embeds = self.token_embedding(input_ids) + self.position_embeds74        attn_mask = self.attn_mask.to(device=embeds.device, dtype=embeds.dtype)75        for encoder_id, encoder in enumerate(self.encoders):76            embeds = encoder(embeds, attn_mask=attn_mask)77            if encoder_id + clip_skip == len(self.encoders):78                hidden_states = embeds79        embeds = self.final_layer_norm(embeds)80        pooled_embeds = embeds[torch.arange(embeds.shape[0]), input_ids.to(dtype=torch.int).argmax(dim=-1)]81        pooled_embeds = self.text_projection(pooled_embeds)82        return pooled_embeds, hidden_states83    84    @staticmethod85    def state_dict_converter():86        return SDXLTextEncoder2StateDictConverter()87 88 89class SDXLTextEncoderStateDictConverter:90    def __init__(self):91        pass92 93    def from_diffusers(self, state_dict):94        rename_dict = {95            "text_model.embeddings.token_embedding.weight": "token_embedding.weight",96            "text_model.embeddings.position_embedding.weight": "position_embeds",97            "text_model.final_layer_norm.weight": "final_layer_norm.weight",98            "text_model.final_layer_norm.bias": "final_layer_norm.bias"99        }100        attn_rename_dict = {101            "self_attn.q_proj": "attn.to_q",102            "self_attn.k_proj": "attn.to_k",103            "self_attn.v_proj": "attn.to_v",104            "self_attn.out_proj": "attn.to_out",105            "layer_norm1": "layer_norm1",106            "layer_norm2": "layer_norm2",107            "mlp.fc1": "fc1",108            "mlp.fc2": "fc2",109        }110        state_dict_ = {}111        for name in state_dict:112            if name in rename_dict:113                param = state_dict[name]114                if name == "text_model.embeddings.position_embedding.weight":115                    param = param.reshape((1, param.shape[0], param.shape[1]))116                state_dict_[rename_dict[name]] = param117            elif name.startswith("text_model.encoder.layers."):118                param = state_dict[name]119                names = name.split(".")120                layer_id, layer_type, tail = names[3], ".".join(names[4:-1]), names[-1]121                name_ = ".".join(["encoders", layer_id, attn_rename_dict[layer_type], tail])122                state_dict_[name_] = param123        return state_dict_124    125    def from_civitai(self, state_dict):126        rename_dict = {127            "conditioner.embedders.0.transformer.text_model.embeddings.position_embedding.weight": "position_embeds",128            "conditioner.embedders.0.transformer.text_model.embeddings.token_embedding.weight": "token_embedding.weight",129            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.layer_norm1.bias": "encoders.0.layer_norm1.bias",130            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.layer_norm1.weight": "encoders.0.layer_norm1.weight",131            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.layer_norm2.bias": "encoders.0.layer_norm2.bias",132            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.layer_norm2.weight": "encoders.0.layer_norm2.weight",133            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.mlp.fc1.bias": "encoders.0.fc1.bias",134            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.mlp.fc1.weight": "encoders.0.fc1.weight",135            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.mlp.fc2.bias": "encoders.0.fc2.bias",136            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.mlp.fc2.weight": "encoders.0.fc2.weight",137            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.k_proj.bias": "encoders.0.attn.to_k.bias",138            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.k_proj.weight": "encoders.0.attn.to_k.weight",139            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.out_proj.bias": "encoders.0.attn.to_out.bias",140            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.out_proj.weight": "encoders.0.attn.to_out.weight",141            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.q_proj.bias": "encoders.0.attn.to_q.bias",142            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.q_proj.weight": "encoders.0.attn.to_q.weight",143            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.v_proj.bias": "encoders.0.attn.to_v.bias",144            "conditioner.embedders.0.transformer.text_model.encoder.layers.0.self_attn.v_proj.weight": "encoders.0.attn.to_v.weight",145            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.layer_norm1.bias": "encoders.1.layer_norm1.bias",146            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.layer_norm1.weight": "encoders.1.layer_norm1.weight",147            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.layer_norm2.bias": "encoders.1.layer_norm2.bias",148            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.layer_norm2.weight": "encoders.1.layer_norm2.weight",149            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.mlp.fc1.bias": "encoders.1.fc1.bias",150            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.mlp.fc1.weight": "encoders.1.fc1.weight",151            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.mlp.fc2.bias": "encoders.1.fc2.bias",152            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.mlp.fc2.weight": "encoders.1.fc2.weight",153            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.k_proj.bias": "encoders.1.attn.to_k.bias",154            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.k_proj.weight": "encoders.1.attn.to_k.weight",155            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.out_proj.bias": "encoders.1.attn.to_out.bias",156            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.out_proj.weight": "encoders.1.attn.to_out.weight",157            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.q_proj.bias": "encoders.1.attn.to_q.bias",158            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.q_proj.weight": "encoders.1.attn.to_q.weight",159            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.v_proj.bias": "encoders.1.attn.to_v.bias",160            "conditioner.embedders.0.transformer.text_model.encoder.layers.1.self_attn.v_proj.weight": "encoders.1.attn.to_v.weight",161            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.layer_norm1.bias": "encoders.10.layer_norm1.bias",162            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.layer_norm1.weight": "encoders.10.layer_norm1.weight",163            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.layer_norm2.bias": "encoders.10.layer_norm2.bias",164            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.layer_norm2.weight": "encoders.10.layer_norm2.weight",165            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.mlp.fc1.bias": "encoders.10.fc1.bias",166            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.mlp.fc1.weight": "encoders.10.fc1.weight",167            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.mlp.fc2.bias": "encoders.10.fc2.bias",168            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.mlp.fc2.weight": "encoders.10.fc2.weight",169            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.k_proj.bias": "encoders.10.attn.to_k.bias",170            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.k_proj.weight": "encoders.10.attn.to_k.weight",171            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.out_proj.bias": "encoders.10.attn.to_out.bias",172            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.out_proj.weight": "encoders.10.attn.to_out.weight",173            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.q_proj.bias": "encoders.10.attn.to_q.bias",174            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.q_proj.weight": "encoders.10.attn.to_q.weight",175            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.v_proj.bias": "encoders.10.attn.to_v.bias",176            "conditioner.embedders.0.transformer.text_model.encoder.layers.10.self_attn.v_proj.weight": "encoders.10.attn.to_v.weight",177            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.layer_norm1.bias": "encoders.2.layer_norm1.bias",178            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.layer_norm1.weight": "encoders.2.layer_norm1.weight",179            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.layer_norm2.bias": "encoders.2.layer_norm2.bias",180            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.layer_norm2.weight": "encoders.2.layer_norm2.weight",181            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.mlp.fc1.bias": "encoders.2.fc1.bias",182            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.mlp.fc1.weight": "encoders.2.fc1.weight",183            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.mlp.fc2.bias": "encoders.2.fc2.bias",184            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.mlp.fc2.weight": "encoders.2.fc2.weight",185            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.k_proj.bias": "encoders.2.attn.to_k.bias",186            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.k_proj.weight": "encoders.2.attn.to_k.weight",187            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.out_proj.bias": "encoders.2.attn.to_out.bias",188            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.out_proj.weight": "encoders.2.attn.to_out.weight",189            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.q_proj.bias": "encoders.2.attn.to_q.bias",190            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.q_proj.weight": "encoders.2.attn.to_q.weight",191            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.v_proj.bias": "encoders.2.attn.to_v.bias",192            "conditioner.embedders.0.transformer.text_model.encoder.layers.2.self_attn.v_proj.weight": "encoders.2.attn.to_v.weight",193            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.layer_norm1.bias": "encoders.3.layer_norm1.bias",194            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.layer_norm1.weight": "encoders.3.layer_norm1.weight",195            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.layer_norm2.bias": "encoders.3.layer_norm2.bias",196            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.layer_norm2.weight": "encoders.3.layer_norm2.weight",197            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.mlp.fc1.bias": "encoders.3.fc1.bias",198            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.mlp.fc1.weight": "encoders.3.fc1.weight",199            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.mlp.fc2.bias": "encoders.3.fc2.bias",200            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.mlp.fc2.weight": "encoders.3.fc2.weight",201            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.k_proj.bias": "encoders.3.attn.to_k.bias",202            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.k_proj.weight": "encoders.3.attn.to_k.weight",203            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.out_proj.bias": "encoders.3.attn.to_out.bias",204            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.out_proj.weight": "encoders.3.attn.to_out.weight",205            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.q_proj.bias": "encoders.3.attn.to_q.bias",206            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.q_proj.weight": "encoders.3.attn.to_q.weight",207            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.v_proj.bias": "encoders.3.attn.to_v.bias",208            "conditioner.embedders.0.transformer.text_model.encoder.layers.3.self_attn.v_proj.weight": "encoders.3.attn.to_v.weight",209            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.layer_norm1.bias": "encoders.4.layer_norm1.bias",210            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.layer_norm1.weight": "encoders.4.layer_norm1.weight",211            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.layer_norm2.bias": "encoders.4.layer_norm2.bias",212            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.layer_norm2.weight": "encoders.4.layer_norm2.weight",213            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.mlp.fc1.bias": "encoders.4.fc1.bias",214            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.mlp.fc1.weight": "encoders.4.fc1.weight",215            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.mlp.fc2.bias": "encoders.4.fc2.bias",216            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.mlp.fc2.weight": "encoders.4.fc2.weight",217            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.k_proj.bias": "encoders.4.attn.to_k.bias",218            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.k_proj.weight": "encoders.4.attn.to_k.weight",219            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.out_proj.bias": "encoders.4.attn.to_out.bias",220            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.out_proj.weight": "encoders.4.attn.to_out.weight",221            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.q_proj.bias": "encoders.4.attn.to_q.bias",222            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.q_proj.weight": "encoders.4.attn.to_q.weight",223            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.v_proj.bias": "encoders.4.attn.to_v.bias",224            "conditioner.embedders.0.transformer.text_model.encoder.layers.4.self_attn.v_proj.weight": "encoders.4.attn.to_v.weight",225            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.layer_norm1.bias": "encoders.5.layer_norm1.bias",226            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.layer_norm1.weight": "encoders.5.layer_norm1.weight",227            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.layer_norm2.bias": "encoders.5.layer_norm2.bias",228            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.layer_norm2.weight": "encoders.5.layer_norm2.weight",229            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.mlp.fc1.bias": "encoders.5.fc1.bias",230            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.mlp.fc1.weight": "encoders.5.fc1.weight",231            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.mlp.fc2.bias": "encoders.5.fc2.bias",232            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.mlp.fc2.weight": "encoders.5.fc2.weight",233            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.k_proj.bias": "encoders.5.attn.to_k.bias",234            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.k_proj.weight": "encoders.5.attn.to_k.weight",235            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.out_proj.bias": "encoders.5.attn.to_out.bias",236            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.out_proj.weight": "encoders.5.attn.to_out.weight",237            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.q_proj.bias": "encoders.5.attn.to_q.bias",238            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.q_proj.weight": "encoders.5.attn.to_q.weight",239            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.v_proj.bias": "encoders.5.attn.to_v.bias",240            "conditioner.embedders.0.transformer.text_model.encoder.layers.5.self_attn.v_proj.weight": "encoders.5.attn.to_v.weight",241            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.layer_norm1.bias": "encoders.6.layer_norm1.bias",242            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.layer_norm1.weight": "encoders.6.layer_norm1.weight",243            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.layer_norm2.bias": "encoders.6.layer_norm2.bias",244            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.layer_norm2.weight": "encoders.6.layer_norm2.weight",245            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.mlp.fc1.bias": "encoders.6.fc1.bias",246            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.mlp.fc1.weight": "encoders.6.fc1.weight",247            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.mlp.fc2.bias": "encoders.6.fc2.bias",248            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.mlp.fc2.weight": "encoders.6.fc2.weight",249            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.k_proj.bias": "encoders.6.attn.to_k.bias",250            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.k_proj.weight": "encoders.6.attn.to_k.weight",251            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.out_proj.bias": "encoders.6.attn.to_out.bias",252            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.out_proj.weight": "encoders.6.attn.to_out.weight",253            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.q_proj.bias": "encoders.6.attn.to_q.bias",254            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.q_proj.weight": "encoders.6.attn.to_q.weight",255            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.v_proj.bias": "encoders.6.attn.to_v.bias",256            "conditioner.embedders.0.transformer.text_model.encoder.layers.6.self_attn.v_proj.weight": "encoders.6.attn.to_v.weight",257            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.layer_norm1.bias": "encoders.7.layer_norm1.bias",258            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.layer_norm1.weight": "encoders.7.layer_norm1.weight",259            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.layer_norm2.bias": "encoders.7.layer_norm2.bias",260            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.layer_norm2.weight": "encoders.7.layer_norm2.weight",261            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.mlp.fc1.bias": "encoders.7.fc1.bias",262            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.mlp.fc1.weight": "encoders.7.fc1.weight",263            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.mlp.fc2.bias": "encoders.7.fc2.bias",264            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.mlp.fc2.weight": "encoders.7.fc2.weight",265            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.k_proj.bias": "encoders.7.attn.to_k.bias",266            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.k_proj.weight": "encoders.7.attn.to_k.weight",267            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.out_proj.bias": "encoders.7.attn.to_out.bias",268            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.out_proj.weight": "encoders.7.attn.to_out.weight",269            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.q_proj.bias": "encoders.7.attn.to_q.bias",270            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.q_proj.weight": "encoders.7.attn.to_q.weight",271            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.v_proj.bias": "encoders.7.attn.to_v.bias",272            "conditioner.embedders.0.transformer.text_model.encoder.layers.7.self_attn.v_proj.weight": "encoders.7.attn.to_v.weight",273            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.layer_norm1.bias": "encoders.8.layer_norm1.bias",274            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.layer_norm1.weight": "encoders.8.layer_norm1.weight",275            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.layer_norm2.bias": "encoders.8.layer_norm2.bias",276            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.layer_norm2.weight": "encoders.8.layer_norm2.weight",277            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.mlp.fc1.bias": "encoders.8.fc1.bias",278            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.mlp.fc1.weight": "encoders.8.fc1.weight",279            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.mlp.fc2.bias": "encoders.8.fc2.bias",280            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.mlp.fc2.weight": "encoders.8.fc2.weight",281            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.k_proj.bias": "encoders.8.attn.to_k.bias",282            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.k_proj.weight": "encoders.8.attn.to_k.weight",283            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.out_proj.bias": "encoders.8.attn.to_out.bias",284            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.out_proj.weight": "encoders.8.attn.to_out.weight",285            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.q_proj.bias": "encoders.8.attn.to_q.bias",286            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.q_proj.weight": "encoders.8.attn.to_q.weight",287            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.v_proj.bias": "encoders.8.attn.to_v.bias",288            "conditioner.embedders.0.transformer.text_model.encoder.layers.8.self_attn.v_proj.weight": "encoders.8.attn.to_v.weight",289            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.layer_norm1.bias": "encoders.9.layer_norm1.bias",290            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.layer_norm1.weight": "encoders.9.layer_norm1.weight",291            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.layer_norm2.bias": "encoders.9.layer_norm2.bias",292            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.layer_norm2.weight": "encoders.9.layer_norm2.weight",293            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.mlp.fc1.bias": "encoders.9.fc1.bias",294            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.mlp.fc1.weight": "encoders.9.fc1.weight",295            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.mlp.fc2.bias": "encoders.9.fc2.bias",296            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.mlp.fc2.weight": "encoders.9.fc2.weight",297            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.k_proj.bias": "encoders.9.attn.to_k.bias",298            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.k_proj.weight": "encoders.9.attn.to_k.weight",299            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.out_proj.bias": "encoders.9.attn.to_out.bias",300            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.out_proj.weight": "encoders.9.attn.to_out.weight",301            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.q_proj.bias": "encoders.9.attn.to_q.bias",302            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.q_proj.weight": "encoders.9.attn.to_q.weight",303            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.v_proj.bias": "encoders.9.attn.to_v.bias",304            "conditioner.embedders.0.transformer.text_model.encoder.layers.9.self_attn.v_proj.weight": "encoders.9.attn.to_v.weight",305        }306        state_dict_ = {}307        for name in state_dict:308            if name in rename_dict:309                param = state_dict[name]310                if name == "conditioner.embedders.0.transformer.text_model.embeddings.position_embedding.weight":311                    param = param.reshape((1, param.shape[0], param.shape[1]))312                state_dict_[rename_dict[name]] = param313        return state_dict_314 315 316class SDXLTextEncoder2StateDictConverter:317    def __init__(self):318        pass319 320    def from_diffusers(self, state_dict):321        rename_dict = {322            "text_model.embeddings.token_embedding.weight": "token_embedding.weight",323            "text_model.embeddings.position_embedding.weight": "position_embeds",324            "text_model.final_layer_norm.weight": "final_layer_norm.weight",325            "text_model.final_layer_norm.bias": "final_layer_norm.bias",326            "text_projection.weight": "text_projection.weight"327        }328        attn_rename_dict = {329            "self_attn.q_proj": "attn.to_q",330            "self_attn.k_proj": "attn.to_k",331            "self_attn.v_proj": "attn.to_v",332            "self_attn.out_proj": "attn.to_out",333            "layer_norm1": "layer_norm1",334            "layer_norm2": "layer_norm2",335            "mlp.fc1": "fc1",336            "mlp.fc2": "fc2",337        }338        state_dict_ = {}339        for name in state_dict:340            if name in rename_dict:341                param = state_dict[name]342                if name == "text_model.embeddings.position_embedding.weight":343                    param = param.reshape((1, param.shape[0], param.shape[1]))344                state_dict_[rename_dict[name]] = param345            elif name.startswith("text_model.encoder.layers."):346                param = state_dict[name]347                names = name.split(".")348                layer_id, layer_type, tail = names[3], ".".join(names[4:-1]), names[-1]349                name_ = ".".join(["encoders", layer_id, attn_rename_dict[layer_type], tail])350                state_dict_[name_] = param351        return state_dict_352    353    def from_civitai(self, state_dict):354        rename_dict = {355            "conditioner.embedders.1.model.ln_final.bias": "final_layer_norm.bias",356            "conditioner.embedders.1.model.ln_final.weight": "final_layer_norm.weight",357            "conditioner.embedders.1.model.positional_embedding": "position_embeds",358            "conditioner.embedders.1.model.token_embedding.weight": "token_embedding.weight",359            "conditioner.embedders.1.model.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'],360            "conditioner.embedders.1.model.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'],361            "conditioner.embedders.1.model.transformer.resblocks.0.attn.out_proj.bias": "encoders.0.attn.to_out.bias",362            "conditioner.embedders.1.model.transformer.resblocks.0.attn.out_proj.weight": "encoders.0.attn.to_out.weight",363            "conditioner.embedders.1.model.transformer.resblocks.0.ln_1.bias": "encoders.0.layer_norm1.bias",364            "conditioner.embedders.1.model.transformer.resblocks.0.ln_1.weight": "encoders.0.layer_norm1.weight",365            "conditioner.embedders.1.model.transformer.resblocks.0.ln_2.bias": "encoders.0.layer_norm2.bias",366            "conditioner.embedders.1.model.transformer.resblocks.0.ln_2.weight": "encoders.0.layer_norm2.weight",367            "conditioner.embedders.1.model.transformer.resblocks.0.mlp.c_fc.bias": "encoders.0.fc1.bias",368            "conditioner.embedders.1.model.transformer.resblocks.0.mlp.c_fc.weight": "encoders.0.fc1.weight",369            "conditioner.embedders.1.model.transformer.resblocks.0.mlp.c_proj.bias": "encoders.0.fc2.bias",370            "conditioner.embedders.1.model.transformer.resblocks.0.mlp.c_proj.weight": "encoders.0.fc2.weight",371            "conditioner.embedders.1.model.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'],372            "conditioner.embedders.1.model.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'],373            "conditioner.embedders.1.model.transformer.resblocks.1.attn.out_proj.bias": "encoders.1.attn.to_out.bias",374            "conditioner.embedders.1.model.transformer.resblocks.1.attn.out_proj.weight": "encoders.1.attn.to_out.weight",375            "conditioner.embedders.1.model.transformer.resblocks.1.ln_1.bias": "encoders.1.layer_norm1.bias",376            "conditioner.embedders.1.model.transformer.resblocks.1.ln_1.weight": "encoders.1.layer_norm1.weight",377            "conditioner.embedders.1.model.transformer.resblocks.1.ln_2.bias": "encoders.1.layer_norm2.bias",378            "conditioner.embedders.1.model.transformer.resblocks.1.ln_2.weight": "encoders.1.layer_norm2.weight",379            "conditioner.embedders.1.model.transformer.resblocks.1.mlp.c_fc.bias": "encoders.1.fc1.bias",380            "conditioner.embedders.1.model.transformer.resblocks.1.mlp.c_fc.weight": "encoders.1.fc1.weight",381            "conditioner.embedders.1.model.transformer.resblocks.1.mlp.c_proj.bias": "encoders.1.fc2.bias",382            "conditioner.embedders.1.model.transformer.resblocks.1.mlp.c_proj.weight": "encoders.1.fc2.weight",383            "conditioner.embedders.1.model.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'],384            "conditioner.embedders.1.model.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'],385            "conditioner.embedders.1.model.transformer.resblocks.10.attn.out_proj.bias": "encoders.10.attn.to_out.bias",386            "conditioner.embedders.1.model.transformer.resblocks.10.attn.out_proj.weight": "encoders.10.attn.to_out.weight",387            "conditioner.embedders.1.model.transformer.resblocks.10.ln_1.bias": "encoders.10.layer_norm1.bias",388            "conditioner.embedders.1.model.transformer.resblocks.10.ln_1.weight": "encoders.10.layer_norm1.weight",389            "conditioner.embedders.1.model.transformer.resblocks.10.ln_2.bias": "encoders.10.layer_norm2.bias",390            "conditioner.embedders.1.model.transformer.resblocks.10.ln_2.weight": "encoders.10.layer_norm2.weight",391            "conditioner.embedders.1.model.transformer.resblocks.10.mlp.c_fc.bias": "encoders.10.fc1.bias",392            "conditioner.embedders.1.model.transformer.resblocks.10.mlp.c_fc.weight": "encoders.10.fc1.weight",393            "conditioner.embedders.1.model.transformer.resblocks.10.mlp.c_proj.bias": "encoders.10.fc2.bias",394            "conditioner.embedders.1.model.transformer.resblocks.10.mlp.c_proj.weight": "encoders.10.fc2.weight",395            "conditioner.embedders.1.model.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'],396            "conditioner.embedders.1.model.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'],397            "conditioner.embedders.1.model.transformer.resblocks.11.attn.out_proj.bias": "encoders.11.attn.to_out.bias",398            "conditioner.embedders.1.model.transformer.resblocks.11.attn.out_proj.weight": "encoders.11.attn.to_out.weight",399            "conditioner.embedders.1.model.transformer.resblocks.11.ln_1.bias": "encoders.11.layer_norm1.bias",400            "conditioner.embedders.1.model.transformer.resblocks.11.ln_1.weight": "encoders.11.layer_norm1.weight",401            "conditioner.embedders.1.model.transformer.resblocks.11.ln_2.bias": "encoders.11.layer_norm2.bias",402            "conditioner.embedders.1.model.transformer.resblocks.11.ln_2.weight": "encoders.11.layer_norm2.weight",403            "conditioner.embedders.1.model.transformer.resblocks.11.mlp.c_fc.bias": "encoders.11.fc1.bias",404            "conditioner.embedders.1.model.transformer.resblocks.11.mlp.c_fc.weight": "encoders.11.fc1.weight",405            "conditioner.embedders.1.model.transformer.resblocks.11.mlp.c_proj.bias": "encoders.11.fc2.bias",406            "conditioner.embedders.1.model.transformer.resblocks.11.mlp.c_proj.weight": "encoders.11.fc2.weight",407            "conditioner.embedders.1.model.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'],408            "conditioner.embedders.1.model.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'],409            "conditioner.embedders.1.model.transformer.resblocks.12.attn.out_proj.bias": "encoders.12.attn.to_out.bias",410            "conditioner.embedders.1.model.transformer.resblocks.12.attn.out_proj.weight": "encoders.12.attn.to_out.weight",411            "conditioner.embedders.1.model.transformer.resblocks.12.ln_1.bias": "encoders.12.layer_norm1.bias",412            "conditioner.embedders.1.model.transformer.resblocks.12.ln_1.weight": "encoders.12.layer_norm1.weight",413            "conditioner.embedders.1.model.transformer.resblocks.12.ln_2.bias": "encoders.12.layer_norm2.bias",414            "conditioner.embedders.1.model.transformer.resblocks.12.ln_2.weight": "encoders.12.layer_norm2.weight",415            "conditioner.embedders.1.model.transformer.resblocks.12.mlp.c_fc.bias": "encoders.12.fc1.bias",416            "conditioner.embedders.1.model.transformer.resblocks.12.mlp.c_fc.weight": "encoders.12.fc1.weight",417            "conditioner.embedders.1.model.transformer.resblocks.12.mlp.c_proj.bias": "encoders.12.fc2.bias",418            "conditioner.embedders.1.model.transformer.resblocks.12.mlp.c_proj.weight": "encoders.12.fc2.weight",419            "conditioner.embedders.1.model.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'],420            "conditioner.embedders.1.model.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'],421            "conditioner.embedders.1.model.transformer.resblocks.13.attn.out_proj.bias": "encoders.13.attn.to_out.bias",422            "conditioner.embedders.1.model.transformer.resblocks.13.attn.out_proj.weight": "encoders.13.attn.to_out.weight",423            "conditioner.embedders.1.model.transformer.resblocks.13.ln_1.bias": "encoders.13.layer_norm1.bias",424            "conditioner.embedders.1.model.transformer.resblocks.13.ln_1.weight": "encoders.13.layer_norm1.weight",425            "conditioner.embedders.1.model.transformer.resblocks.13.ln_2.bias": "encoders.13.layer_norm2.bias",426            "conditioner.embedders.1.model.transformer.resblocks.13.ln_2.weight": "encoders.13.layer_norm2.weight",427            "conditioner.embedders.1.model.transformer.resblocks.13.mlp.c_fc.bias": "encoders.13.fc1.bias",428            "conditioner.embedders.1.model.transformer.resblocks.13.mlp.c_fc.weight": "encoders.13.fc1.weight",429            "conditioner.embedders.1.model.transformer.resblocks.13.mlp.c_proj.bias": "encoders.13.fc2.bias",430            "conditioner.embedders.1.model.transformer.resblocks.13.mlp.c_proj.weight": "encoders.13.fc2.weight",431            "conditioner.embedders.1.model.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'],432            "conditioner.embedders.1.model.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'],433            "conditioner.embedders.1.model.transformer.resblocks.14.attn.out_proj.bias": "encoders.14.attn.to_out.bias",434            "conditioner.embedders.1.model.transformer.resblocks.14.attn.out_proj.weight": "encoders.14.attn.to_out.weight",435            "conditioner.embedders.1.model.transformer.resblocks.14.ln_1.bias": "encoders.14.layer_norm1.bias",436            "conditioner.embedders.1.model.transformer.resblocks.14.ln_1.weight": "encoders.14.layer_norm1.weight",437            "conditioner.embedders.1.model.transformer.resblocks.14.ln_2.bias": "encoders.14.layer_norm2.bias",438            "conditioner.embedders.1.model.transformer.resblocks.14.ln_2.weight": "encoders.14.layer_norm2.weight",439            "conditioner.embedders.1.model.transformer.resblocks.14.mlp.c_fc.bias": "encoders.14.fc1.bias",440            "conditioner.embedders.1.model.transformer.resblocks.14.mlp.c_fc.weight": "encoders.14.fc1.weight",441            "conditioner.embedders.1.model.transformer.resblocks.14.mlp.c_proj.bias": "encoders.14.fc2.bias",442            "conditioner.embedders.1.model.transformer.resblocks.14.mlp.c_proj.weight": "encoders.14.fc2.weight",443            "conditioner.embedders.1.model.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'],444            "conditioner.embedders.1.model.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'],445            "conditioner.embedders.1.model.transformer.resblocks.15.attn.out_proj.bias": "encoders.15.attn.to_out.bias",446            "conditioner.embedders.1.model.transformer.resblocks.15.attn.out_proj.weight": "encoders.15.attn.to_out.weight",447            "conditioner.embedders.1.model.transformer.resblocks.15.ln_1.bias": "encoders.15.layer_norm1.bias",448            "conditioner.embedders.1.model.transformer.resblocks.15.ln_1.weight": "encoders.15.layer_norm1.weight",449            "conditioner.embedders.1.model.transformer.resblocks.15.ln_2.bias": "encoders.15.layer_norm2.bias",450            "conditioner.embedders.1.model.transformer.resblocks.15.ln_2.weight": "encoders.15.layer_norm2.weight",451            "conditioner.embedders.1.model.transformer.resblocks.15.mlp.c_fc.bias": "encoders.15.fc1.bias",452            "conditioner.embedders.1.model.transformer.resblocks.15.mlp.c_fc.weight": "encoders.15.fc1.weight",453            "conditioner.embedders.1.model.transformer.resblocks.15.mlp.c_proj.bias": "encoders.15.fc2.bias",454            "conditioner.embedders.1.model.transformer.resblocks.15.mlp.c_proj.weight": "encoders.15.fc2.weight",455            "conditioner.embedders.1.model.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'],456            "conditioner.embedders.1.model.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'],457            "conditioner.embedders.1.model.transformer.resblocks.16.attn.out_proj.bias": "encoders.16.attn.to_out.bias",458            "conditioner.embedders.1.model.transformer.resblocks.16.attn.out_proj.weight": "encoders.16.attn.to_out.weight",459            "conditioner.embedders.1.model.transformer.resblocks.16.ln_1.bias": "encoders.16.layer_norm1.bias",460            "conditioner.embedders.1.model.transformer.resblocks.16.ln_1.weight": "encoders.16.layer_norm1.weight",461            "conditioner.embedders.1.model.transformer.resblocks.16.ln_2.bias": "encoders.16.layer_norm2.bias",462            "conditioner.embedders.1.model.transformer.resblocks.16.ln_2.weight": "encoders.16.layer_norm2.weight",463            "conditioner.embedders.1.model.transformer.resblocks.16.mlp.c_fc.bias": "encoders.16.fc1.bias",464            "conditioner.embedders.1.model.transformer.resblocks.16.mlp.c_fc.weight": "encoders.16.fc1.weight",465            "conditioner.embedders.1.model.transformer.resblocks.16.mlp.c_proj.bias": "encoders.16.fc2.bias",466            "conditioner.embedders.1.model.transformer.resblocks.16.mlp.c_proj.weight": "encoders.16.fc2.weight",467            "conditioner.embedders.1.model.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'],468            "conditioner.embedders.1.model.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'],469            "conditioner.embedders.1.model.transformer.resblocks.17.attn.out_proj.bias": "encoders.17.attn.to_out.bias",470            "conditioner.embedders.1.model.transformer.resblocks.17.attn.out_proj.weight": "encoders.17.attn.to_out.weight",471            "conditioner.embedders.1.model.transformer.resblocks.17.ln_1.bias": "encoders.17.layer_norm1.bias",472            "conditioner.embedders.1.model.transformer.resblocks.17.ln_1.weight": "encoders.17.layer_norm1.weight",473            "conditioner.embedders.1.model.transformer.resblocks.17.ln_2.bias": "encoders.17.layer_norm2.bias",474            "conditioner.embedders.1.model.transformer.resblocks.17.ln_2.weight": "encoders.17.layer_norm2.weight",475            "conditioner.embedders.1.model.transformer.resblocks.17.mlp.c_fc.bias": "encoders.17.fc1.bias",476            "conditioner.embedders.1.model.transformer.resblocks.17.mlp.c_fc.weight": "encoders.17.fc1.weight",477            "conditioner.embedders.1.model.transformer.resblocks.17.mlp.c_proj.bias": "encoders.17.fc2.bias",478            "conditioner.embedders.1.model.transformer.resblocks.17.mlp.c_proj.weight": "encoders.17.fc2.weight",479            "conditioner.embedders.1.model.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'],480            "conditioner.embedders.1.model.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'],481            "conditioner.embedders.1.model.transformer.resblocks.18.attn.out_proj.bias": "encoders.18.attn.to_out.bias",482            "conditioner.embedders.1.model.transformer.resblocks.18.attn.out_proj.weight": "encoders.18.attn.to_out.weight",483            "conditioner.embedders.1.model.transformer.resblocks.18.ln_1.bias": "encoders.18.layer_norm1.bias",484            "conditioner.embedders.1.model.transformer.resblocks.18.ln_1.weight": "encoders.18.layer_norm1.weight",485            "conditioner.embedders.1.model.transformer.resblocks.18.ln_2.bias": "encoders.18.layer_norm2.bias",486            "conditioner.embedders.1.model.transformer.resblocks.18.ln_2.weight": "encoders.18.layer_norm2.weight",487            "conditioner.embedders.1.model.transformer.resblocks.18.mlp.c_fc.bias": "encoders.18.fc1.bias",488            "conditioner.embedders.1.model.transformer.resblocks.18.mlp.c_fc.weight": "encoders.18.fc1.weight",489            "conditioner.embedders.1.model.transformer.resblocks.18.mlp.c_proj.bias": "encoders.18.fc2.bias",490            "conditioner.embedders.1.model.transformer.resblocks.18.mlp.c_proj.weight": "encoders.18.fc2.weight",491            "conditioner.embedders.1.model.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'],492            "conditioner.embedders.1.model.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'],493            "conditioner.embedders.1.model.transformer.resblocks.19.attn.out_proj.bias": "encoders.19.attn.to_out.bias",494            "conditioner.embedders.1.model.transformer.resblocks.19.attn.out_proj.weight": "encoders.19.attn.to_out.weight",495            "conditioner.embedders.1.model.transformer.resblocks.19.ln_1.bias": "encoders.19.layer_norm1.bias",496            "conditioner.embedders.1.model.transformer.resblocks.19.ln_1.weight": "encoders.19.layer_norm1.weight",497            "conditioner.embedders.1.model.transformer.resblocks.19.ln_2.bias": "encoders.19.layer_norm2.bias",498            "conditioner.embedders.1.model.transformer.resblocks.19.ln_2.weight": "encoders.19.layer_norm2.weight",499            "conditioner.embedders.1.model.transformer.resblocks.19.mlp.c_fc.bias": "encoders.19.fc1.bias",500            "conditioner.embedders.1.model.transformer.resblocks.19.mlp.c_fc.weight": "encoders.19.fc1.weight",501            "conditioner.embedders.1.model.transformer.resblocks.19.mlp.c_proj.bias": "encoders.19.fc2.bias",502            "conditioner.embedders.1.model.transformer.resblocks.19.mlp.c_proj.weight": "encoders.19.fc2.weight",503            "conditioner.embedders.1.model.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'],504            "conditioner.embedders.1.model.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'],505            "conditioner.embedders.1.model.transformer.resblocks.2.attn.out_proj.bias": "encoders.2.attn.to_out.bias",506            "conditioner.embedders.1.model.transformer.resblocks.2.attn.out_proj.weight": "encoders.2.attn.to_out.weight",507            "conditioner.embedders.1.model.transformer.resblocks.2.ln_1.bias": "encoders.2.layer_norm1.bias",508            "conditioner.embedders.1.model.transformer.resblocks.2.ln_1.weight": "encoders.2.layer_norm1.weight",509            "conditioner.embedders.1.model.transformer.resblocks.2.ln_2.bias": "encoders.2.layer_norm2.bias",510            "conditioner.embedders.1.model.transformer.resblocks.2.ln_2.weight": "encoders.2.layer_norm2.weight",511            "conditioner.embedders.1.model.transformer.resblocks.2.mlp.c_fc.bias": "encoders.2.fc1.bias",512            "conditioner.embedders.1.model.transformer.resblocks.2.mlp.c_fc.weight": "encoders.2.fc1.weight",513            "conditioner.embedders.1.model.transformer.resblocks.2.mlp.c_proj.bias": "encoders.2.fc2.bias",514            "conditioner.embedders.1.model.transformer.resblocks.2.mlp.c_proj.weight": "encoders.2.fc2.weight",515            "conditioner.embedders.1.model.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'],516            "conditioner.embedders.1.model.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'],517            "conditioner.embedders.1.model.transformer.resblocks.20.attn.out_proj.bias": "encoders.20.attn.to_out.bias",518            "conditioner.embedders.1.model.transformer.resblocks.20.attn.out_proj.weight": "encoders.20.attn.to_out.weight",519            "conditioner.embedders.1.model.transformer.resblocks.20.ln_1.bias": "encoders.20.layer_norm1.bias",520            "conditioner.embedders.1.model.transformer.resblocks.20.ln_1.weight": "encoders.20.layer_norm1.weight",521            "conditioner.embedders.1.model.transformer.resblocks.20.ln_2.bias": "encoders.20.layer_norm2.bias",522            "conditioner.embedders.1.model.transformer.resblocks.20.ln_2.weight": "encoders.20.layer_norm2.weight",523            "conditioner.embedders.1.model.transformer.resblocks.20.mlp.c_fc.bias": "encoders.20.fc1.bias",524            "conditioner.embedders.1.model.transformer.resblocks.20.mlp.c_fc.weight": "encoders.20.fc1.weight",525            "conditioner.embedders.1.model.transformer.resblocks.20.mlp.c_proj.bias": "encoders.20.fc2.bias",526            "conditioner.embedders.1.model.transformer.resblocks.20.mlp.c_proj.weight": "encoders.20.fc2.weight",527            "conditioner.embedders.1.model.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'],528            "conditioner.embedders.1.model.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'],529            "conditioner.embedders.1.model.transformer.resblocks.21.attn.out_proj.bias": "encoders.21.attn.to_out.bias",530            "conditioner.embedders.1.model.transformer.resblocks.21.attn.out_proj.weight": "encoders.21.attn.to_out.weight",531            "conditioner.embedders.1.model.transformer.resblocks.21.ln_1.bias": "encoders.21.layer_norm1.bias",532            "conditioner.embedders.1.model.transformer.resblocks.21.ln_1.weight": "encoders.21.layer_norm1.weight",533            "conditioner.embedders.1.model.transformer.resblocks.21.ln_2.bias": "encoders.21.layer_norm2.bias",534            "conditioner.embedders.1.model.transformer.resblocks.21.ln_2.weight": "encoders.21.layer_norm2.weight",535            "conditioner.embedders.1.model.transformer.resblocks.21.mlp.c_fc.bias": "encoders.21.fc1.bias",536            "conditioner.embedders.1.model.transformer.resblocks.21.mlp.c_fc.weight": "encoders.21.fc1.weight",537            "conditioner.embedders.1.model.transformer.resblocks.21.mlp.c_proj.bias": "encoders.21.fc2.bias",538            "conditioner.embedders.1.model.transformer.resblocks.21.mlp.c_proj.weight": "encoders.21.fc2.weight",539            "conditioner.embedders.1.model.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'],540            "conditioner.embedders.1.model.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'],541            "conditioner.embedders.1.model.transformer.resblocks.22.attn.out_proj.bias": "encoders.22.attn.to_out.bias",542            "conditioner.embedders.1.model.transformer.resblocks.22.attn.out_proj.weight": "encoders.22.attn.to_out.weight",543            "conditioner.embedders.1.model.transformer.resblocks.22.ln_1.bias": "encoders.22.layer_norm1.bias",544            "conditioner.embedders.1.model.transformer.resblocks.22.ln_1.weight": "encoders.22.layer_norm1.weight",545            "conditioner.embedders.1.model.transformer.resblocks.22.ln_2.bias": "encoders.22.layer_norm2.bias",546            "conditioner.embedders.1.model.transformer.resblocks.22.ln_2.weight": "encoders.22.layer_norm2.weight",547            "conditioner.embedders.1.model.transformer.resblocks.22.mlp.c_fc.bias": "encoders.22.fc1.bias",548            "conditioner.embedders.1.model.transformer.resblocks.22.mlp.c_fc.weight": "encoders.22.fc1.weight",549            "conditioner.embedders.1.model.transformer.resblocks.22.mlp.c_proj.bias": "encoders.22.fc2.bias",550            "conditioner.embedders.1.model.transformer.resblocks.22.mlp.c_proj.weight": "encoders.22.fc2.weight",551            "conditioner.embedders.1.model.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'],552            "conditioner.embedders.1.model.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'],553            "conditioner.embedders.1.model.transformer.resblocks.23.attn.out_proj.bias": "encoders.23.attn.to_out.bias",554            "conditioner.embedders.1.model.transformer.resblocks.23.attn.out_proj.weight": "encoders.23.attn.to_out.weight",555            "conditioner.embedders.1.model.transformer.resblocks.23.ln_1.bias": "encoders.23.layer_norm1.bias",556            "conditioner.embedders.1.model.transformer.resblocks.23.ln_1.weight": "encoders.23.layer_norm1.weight",557            "conditioner.embedders.1.model.transformer.resblocks.23.ln_2.bias": "encoders.23.layer_norm2.bias",558            "conditioner.embedders.1.model.transformer.resblocks.23.ln_2.weight": "encoders.23.layer_norm2.weight",559            "conditioner.embedders.1.model.transformer.resblocks.23.mlp.c_fc.bias": "encoders.23.fc1.bias",560            "conditioner.embedders.1.model.transformer.resblocks.23.mlp.c_fc.weight": "encoders.23.fc1.weight",561            "conditioner.embedders.1.model.transformer.resblocks.23.mlp.c_proj.bias": "encoders.23.fc2.bias",562            "conditioner.embedders.1.model.transformer.resblocks.23.mlp.c_proj.weight": "encoders.23.fc2.weight",563            "conditioner.embedders.1.model.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'],564            "conditioner.embedders.1.model.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'],565            "conditioner.embedders.1.model.transformer.resblocks.24.attn.out_proj.bias": "encoders.24.attn.to_out.bias",566            "conditioner.embedders.1.model.transformer.resblocks.24.attn.out_proj.weight": "encoders.24.attn.to_out.weight",567            "conditioner.embedders.1.model.transformer.resblocks.24.ln_1.bias": "encoders.24.layer_norm1.bias",568            "conditioner.embedders.1.model.transformer.resblocks.24.ln_1.weight": "encoders.24.layer_norm1.weight",569            "conditioner.embedders.1.model.transformer.resblocks.24.ln_2.bias": "encoders.24.layer_norm2.bias",570            "conditioner.embedders.1.model.transformer.resblocks.24.ln_2.weight": "encoders.24.layer_norm2.weight",571            "conditioner.embedders.1.model.transformer.resblocks.24.mlp.c_fc.bias": "encoders.24.fc1.bias",572            "conditioner.embedders.1.model.transformer.resblocks.24.mlp.c_fc.weight": "encoders.24.fc1.weight",573            "conditioner.embedders.1.model.transformer.resblocks.24.mlp.c_proj.bias": "encoders.24.fc2.bias",574            "conditioner.embedders.1.model.transformer.resblocks.24.mlp.c_proj.weight": "encoders.24.fc2.weight",575            "conditioner.embedders.1.model.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'],576            "conditioner.embedders.1.model.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'],577            "conditioner.embedders.1.model.transformer.resblocks.25.attn.out_proj.bias": "encoders.25.attn.to_out.bias",578            "conditioner.embedders.1.model.transformer.resblocks.25.attn.out_proj.weight": "encoders.25.attn.to_out.weight",579            "conditioner.embedders.1.model.transformer.resblocks.25.ln_1.bias": "encoders.25.layer_norm1.bias",580            "conditioner.embedders.1.model.transformer.resblocks.25.ln_1.weight": "encoders.25.layer_norm1.weight",581            "conditioner.embedders.1.model.transformer.resblocks.25.ln_2.bias": "encoders.25.layer_norm2.bias",582            "conditioner.embedders.1.model.transformer.resblocks.25.ln_2.weight": "encoders.25.layer_norm2.weight",583            "conditioner.embedders.1.model.transformer.resblocks.25.mlp.c_fc.bias": "encoders.25.fc1.bias",584            "conditioner.embedders.1.model.transformer.resblocks.25.mlp.c_fc.weight": "encoders.25.fc1.weight",585            "conditioner.embedders.1.model.transformer.resblocks.25.mlp.c_proj.bias": "encoders.25.fc2.bias",586            "conditioner.embedders.1.model.transformer.resblocks.25.mlp.c_proj.weight": "encoders.25.fc2.weight",587            "conditioner.embedders.1.model.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'],588            "conditioner.embedders.1.model.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'],589            "conditioner.embedders.1.model.transformer.resblocks.26.attn.out_proj.bias": "encoders.26.attn.to_out.bias",590            "conditioner.embedders.1.model.transformer.resblocks.26.attn.out_proj.weight": "encoders.26.attn.to_out.weight",591            "conditioner.embedders.1.model.transformer.resblocks.26.ln_1.bias": "encoders.26.layer_norm1.bias",592            "conditioner.embedders.1.model.transformer.resblocks.26.ln_1.weight": "encoders.26.layer_norm1.weight",593            "conditioner.embedders.1.model.transformer.resblocks.26.ln_2.bias": "encoders.26.layer_norm2.bias",594            "conditioner.embedders.1.model.transformer.resblocks.26.ln_2.weight": "encoders.26.layer_norm2.weight",595            "conditioner.embedders.1.model.transformer.resblocks.26.mlp.c_fc.bias": "encoders.26.fc1.bias",596            "conditioner.embedders.1.model.transformer.resblocks.26.mlp.c_fc.weight": "encoders.26.fc1.weight",597            "conditioner.embedders.1.model.transformer.resblocks.26.mlp.c_proj.bias": "encoders.26.fc2.bias",598            "conditioner.embedders.1.model.transformer.resblocks.26.mlp.c_proj.weight": "encoders.26.fc2.weight",599            "conditioner.embedders.1.model.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'],600            "conditioner.embedders.1.model.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'],601            "conditioner.embedders.1.model.transformer.resblocks.27.attn.out_proj.bias": "encoders.27.attn.to_out.bias",602            "conditioner.embedders.1.model.transformer.resblocks.27.attn.out_proj.weight": "encoders.27.attn.to_out.weight",603            "conditioner.embedders.1.model.transformer.resblocks.27.ln_1.bias": "encoders.27.layer_norm1.bias",604            "conditioner.embedders.1.model.transformer.resblocks.27.ln_1.weight": "encoders.27.layer_norm1.weight",605            "conditioner.embedders.1.model.transformer.resblocks.27.ln_2.bias": "encoders.27.layer_norm2.bias",606            "conditioner.embedders.1.model.transformer.resblocks.27.ln_2.weight": "encoders.27.layer_norm2.weight",607            "conditioner.embedders.1.model.transformer.resblocks.27.mlp.c_fc.bias": "encoders.27.fc1.bias",608            "conditioner.embedders.1.model.transformer.resblocks.27.mlp.c_fc.weight": "encoders.27.fc1.weight",609            "conditioner.embedders.1.model.transformer.resblocks.27.mlp.c_proj.bias": "encoders.27.fc2.bias",610            "conditioner.embedders.1.model.transformer.resblocks.27.mlp.c_proj.weight": "encoders.27.fc2.weight",611            "conditioner.embedders.1.model.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'],612            "conditioner.embedders.1.model.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'],613            "conditioner.embedders.1.model.transformer.resblocks.28.attn.out_proj.bias": "encoders.28.attn.to_out.bias",614            "conditioner.embedders.1.model.transformer.resblocks.28.attn.out_proj.weight": "encoders.28.attn.to_out.weight",615            "conditioner.embedders.1.model.transformer.resblocks.28.ln_1.bias": "encoders.28.layer_norm1.bias",616            "conditioner.embedders.1.model.transformer.resblocks.28.ln_1.weight": "encoders.28.layer_norm1.weight",617            "conditioner.embedders.1.model.transformer.resblocks.28.ln_2.bias": "encoders.28.layer_norm2.bias",618            "conditioner.embedders.1.model.transformer.resblocks.28.ln_2.weight": "encoders.28.layer_norm2.weight",619            "conditioner.embedders.1.model.transformer.resblocks.28.mlp.c_fc.bias": "encoders.28.fc1.bias",620            "conditioner.embedders.1.model.transformer.resblocks.28.mlp.c_fc.weight": "encoders.28.fc1.weight",621            "conditioner.embedders.1.model.transformer.resblocks.28.mlp.c_proj.bias": "encoders.28.fc2.bias",622            "conditioner.embedders.1.model.transformer.resblocks.28.mlp.c_proj.weight": "encoders.28.fc2.weight",623            "conditioner.embedders.1.model.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'],624            "conditioner.embedders.1.model.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'],625            "conditioner.embedders.1.model.transformer.resblocks.29.attn.out_proj.bias": "encoders.29.attn.to_out.bias",626            "conditioner.embedders.1.model.transformer.resblocks.29.attn.out_proj.weight": "encoders.29.attn.to_out.weight",627            "conditioner.embedders.1.model.transformer.resblocks.29.ln_1.bias": "encoders.29.layer_norm1.bias",628            "conditioner.embedders.1.model.transformer.resblocks.29.ln_1.weight": "encoders.29.layer_norm1.weight",629            "conditioner.embedders.1.model.transformer.resblocks.29.ln_2.bias": "encoders.29.layer_norm2.bias",630            "conditioner.embedders.1.model.transformer.resblocks.29.ln_2.weight": "encoders.29.layer_norm2.weight",631            "conditioner.embedders.1.model.transformer.resblocks.29.mlp.c_fc.bias": "encoders.29.fc1.bias",632            "conditioner.embedders.1.model.transformer.resblocks.29.mlp.c_fc.weight": "encoders.29.fc1.weight",633            "conditioner.embedders.1.model.transformer.resblocks.29.mlp.c_proj.bias": "encoders.29.fc2.bias",634            "conditioner.embedders.1.model.transformer.resblocks.29.mlp.c_proj.weight": "encoders.29.fc2.weight",635            "conditioner.embedders.1.model.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'],636            "conditioner.embedders.1.model.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'],637            "conditioner.embedders.1.model.transformer.resblocks.3.attn.out_proj.bias": "encoders.3.attn.to_out.bias",638            "conditioner.embedders.1.model.transformer.resblocks.3.attn.out_proj.weight": "encoders.3.attn.to_out.weight",639            "conditioner.embedders.1.model.transformer.resblocks.3.ln_1.bias": "encoders.3.layer_norm1.bias",640            "conditioner.embedders.1.model.transformer.resblocks.3.ln_1.weight": "encoders.3.layer_norm1.weight",641            "conditioner.embedders.1.model.transformer.resblocks.3.ln_2.bias": "encoders.3.layer_norm2.bias",642            "conditioner.embedders.1.model.transformer.resblocks.3.ln_2.weight": "encoders.3.layer_norm2.weight",643            "conditioner.embedders.1.model.transformer.resblocks.3.mlp.c_fc.bias": "encoders.3.fc1.bias",644            "conditioner.embedders.1.model.transformer.resblocks.3.mlp.c_fc.weight": "encoders.3.fc1.weight",645            "conditioner.embedders.1.model.transformer.resblocks.3.mlp.c_proj.bias": "encoders.3.fc2.bias",646            "conditioner.embedders.1.model.transformer.resblocks.3.mlp.c_proj.weight": "encoders.3.fc2.weight",647            "conditioner.embedders.1.model.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'],648            "conditioner.embedders.1.model.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'],649            "conditioner.embedders.1.model.transformer.resblocks.30.attn.out_proj.bias": "encoders.30.attn.to_out.bias",650            "conditioner.embedders.1.model.transformer.resblocks.30.attn.out_proj.weight": "encoders.30.attn.to_out.weight",651            "conditioner.embedders.1.model.transformer.resblocks.30.ln_1.bias": "encoders.30.layer_norm1.bias",652            "conditioner.embedders.1.model.transformer.resblocks.30.ln_1.weight": "encoders.30.layer_norm1.weight",653            "conditioner.embedders.1.model.transformer.resblocks.30.ln_2.bias": "encoders.30.layer_norm2.bias",654            "conditioner.embedders.1.model.transformer.resblocks.30.ln_2.weight": "encoders.30.layer_norm2.weight",655            "conditioner.embedders.1.model.transformer.resblocks.30.mlp.c_fc.bias": "encoders.30.fc1.bias",656            "conditioner.embedders.1.model.transformer.resblocks.30.mlp.c_fc.weight": "encoders.30.fc1.weight",657            "conditioner.embedders.1.model.transformer.resblocks.30.mlp.c_proj.bias": "encoders.30.fc2.bias",658            "conditioner.embedders.1.model.transformer.resblocks.30.mlp.c_proj.weight": "encoders.30.fc2.weight",659            "conditioner.embedders.1.model.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'],660            "conditioner.embedders.1.model.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'],661            "conditioner.embedders.1.model.transformer.resblocks.31.attn.out_proj.bias": "encoders.31.attn.to_out.bias",662            "conditioner.embedders.1.model.transformer.resblocks.31.attn.out_proj.weight": "encoders.31.attn.to_out.weight",663            "conditioner.embedders.1.model.transformer.resblocks.31.ln_1.bias": "encoders.31.layer_norm1.bias",664            "conditioner.embedders.1.model.transformer.resblocks.31.ln_1.weight": "encoders.31.layer_norm1.weight",665            "conditioner.embedders.1.model.transformer.resblocks.31.ln_2.bias": "encoders.31.layer_norm2.bias",666            "conditioner.embedders.1.model.transformer.resblocks.31.ln_2.weight": "encoders.31.layer_norm2.weight",667            "conditioner.embedders.1.model.transformer.resblocks.31.mlp.c_fc.bias": "encoders.31.fc1.bias",668            "conditioner.embedders.1.model.transformer.resblocks.31.mlp.c_fc.weight": "encoders.31.fc1.weight",669            "conditioner.embedders.1.model.transformer.resblocks.31.mlp.c_proj.bias": "encoders.31.fc2.bias",670            "conditioner.embedders.1.model.transformer.resblocks.31.mlp.c_proj.weight": "encoders.31.fc2.weight",671            "conditioner.embedders.1.model.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'],672            "conditioner.embedders.1.model.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'],673            "conditioner.embedders.1.model.transformer.resblocks.4.attn.out_proj.bias": "encoders.4.attn.to_out.bias",674            "conditioner.embedders.1.model.transformer.resblocks.4.attn.out_proj.weight": "encoders.4.attn.to_out.weight",675            "conditioner.embedders.1.model.transformer.resblocks.4.ln_1.bias": "encoders.4.layer_norm1.bias",676            "conditioner.embedders.1.model.transformer.resblocks.4.ln_1.weight": "encoders.4.layer_norm1.weight",677            "conditioner.embedders.1.model.transformer.resblocks.4.ln_2.bias": "encoders.4.layer_norm2.bias",678            "conditioner.embedders.1.model.transformer.resblocks.4.ln_2.weight": "encoders.4.layer_norm2.weight",679            "conditioner.embedders.1.model.transformer.resblocks.4.mlp.c_fc.bias": "encoders.4.fc1.bias",680            "conditioner.embedders.1.model.transformer.resblocks.4.mlp.c_fc.weight": "encoders.4.fc1.weight",681            "conditioner.embedders.1.model.transformer.resblocks.4.mlp.c_proj.bias": "encoders.4.fc2.bias",682            "conditioner.embedders.1.model.transformer.resblocks.4.mlp.c_proj.weight": "encoders.4.fc2.weight",683            "conditioner.embedders.1.model.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'],684            "conditioner.embedders.1.model.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'],685            "conditioner.embedders.1.model.transformer.resblocks.5.attn.out_proj.bias": "encoders.5.attn.to_out.bias",686            "conditioner.embedders.1.model.transformer.resblocks.5.attn.out_proj.weight": "encoders.5.attn.to_out.weight",687            "conditioner.embedders.1.model.transformer.resblocks.5.ln_1.bias": "encoders.5.layer_norm1.bias",688            "conditioner.embedders.1.model.transformer.resblocks.5.ln_1.weight": "encoders.5.layer_norm1.weight",689            "conditioner.embedders.1.model.transformer.resblocks.5.ln_2.bias": "encoders.5.layer_norm2.bias",690            "conditioner.embedders.1.model.transformer.resblocks.5.ln_2.weight": "encoders.5.layer_norm2.weight",691            "conditioner.embedders.1.model.transformer.resblocks.5.mlp.c_fc.bias": "encoders.5.fc1.bias",692            "conditioner.embedders.1.model.transformer.resblocks.5.mlp.c_fc.weight": "encoders.5.fc1.weight",693            "conditioner.embedders.1.model.transformer.resblocks.5.mlp.c_proj.bias": "encoders.5.fc2.bias",694            "conditioner.embedders.1.model.transformer.resblocks.5.mlp.c_proj.weight": "encoders.5.fc2.weight",695            "conditioner.embedders.1.model.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'],696            "conditioner.embedders.1.model.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'],697            "conditioner.embedders.1.model.transformer.resblocks.6.attn.out_proj.bias": "encoders.6.attn.to_out.bias",698            "conditioner.embedders.1.model.transformer.resblocks.6.attn.out_proj.weight": "encoders.6.attn.to_out.weight",699            "conditioner.embedders.1.model.transformer.resblocks.6.ln_1.bias": "encoders.6.layer_norm1.bias",700            "conditioner.embedders.1.model.transformer.resblocks.6.ln_1.weight": "encoders.6.layer_norm1.weight",701            "conditioner.embedders.1.model.transformer.resblocks.6.ln_2.bias": "encoders.6.layer_norm2.bias",702            "conditioner.embedders.1.model.transformer.resblocks.6.ln_2.weight": "encoders.6.layer_norm2.weight",703            "conditioner.embedders.1.model.transformer.resblocks.6.mlp.c_fc.bias": "encoders.6.fc1.bias",704            "conditioner.embedders.1.model.transformer.resblocks.6.mlp.c_fc.weight": "encoders.6.fc1.weight",705            "conditioner.embedders.1.model.transformer.resblocks.6.mlp.c_proj.bias": "encoders.6.fc2.bias",706            "conditioner.embedders.1.model.transformer.resblocks.6.mlp.c_proj.weight": "encoders.6.fc2.weight",707            "conditioner.embedders.1.model.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'],708            "conditioner.embedders.1.model.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'],709            "conditioner.embedders.1.model.transformer.resblocks.7.attn.out_proj.bias": "encoders.7.attn.to_out.bias",710            "conditioner.embedders.1.model.transformer.resblocks.7.attn.out_proj.weight": "encoders.7.attn.to_out.weight",711            "conditioner.embedders.1.model.transformer.resblocks.7.ln_1.bias": "encoders.7.layer_norm1.bias",712            "conditioner.embedders.1.model.transformer.resblocks.7.ln_1.weight": "encoders.7.layer_norm1.weight",713            "conditioner.embedders.1.model.transformer.resblocks.7.ln_2.bias": "encoders.7.layer_norm2.bias",714            "conditioner.embedders.1.model.transformer.resblocks.7.ln_2.weight": "encoders.7.layer_norm2.weight",715            "conditioner.embedders.1.model.transformer.resblocks.7.mlp.c_fc.bias": "encoders.7.fc1.bias",716            "conditioner.embedders.1.model.transformer.resblocks.7.mlp.c_fc.weight": "encoders.7.fc1.weight",717            "conditioner.embedders.1.model.transformer.resblocks.7.mlp.c_proj.bias": "encoders.7.fc2.bias",718            "conditioner.embedders.1.model.transformer.resblocks.7.mlp.c_proj.weight": "encoders.7.fc2.weight",719            "conditioner.embedders.1.model.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'],720            "conditioner.embedders.1.model.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'],721            "conditioner.embedders.1.model.transformer.resblocks.8.attn.out_proj.bias": "encoders.8.attn.to_out.bias",722            "conditioner.embedders.1.model.transformer.resblocks.8.attn.out_proj.weight": "encoders.8.attn.to_out.weight",723            "conditioner.embedders.1.model.transformer.resblocks.8.ln_1.bias": "encoders.8.layer_norm1.bias",724            "conditioner.embedders.1.model.transformer.resblocks.8.ln_1.weight": "encoders.8.layer_norm1.weight",725            "conditioner.embedders.1.model.transformer.resblocks.8.ln_2.bias": "encoders.8.layer_norm2.bias",726            "conditioner.embedders.1.model.transformer.resblocks.8.ln_2.weight": "encoders.8.layer_norm2.weight",727            "conditioner.embedders.1.model.transformer.resblocks.8.mlp.c_fc.bias": "encoders.8.fc1.bias",728            "conditioner.embedders.1.model.transformer.resblocks.8.mlp.c_fc.weight": "encoders.8.fc1.weight",729            "conditioner.embedders.1.model.transformer.resblocks.8.mlp.c_proj.bias": "encoders.8.fc2.bias",730            "conditioner.embedders.1.model.transformer.resblocks.8.mlp.c_proj.weight": "encoders.8.fc2.weight",731            "conditioner.embedders.1.model.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'],732            "conditioner.embedders.1.model.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'],733            "conditioner.embedders.1.model.transformer.resblocks.9.attn.out_proj.bias": "encoders.9.attn.to_out.bias",734            "conditioner.embedders.1.model.transformer.resblocks.9.attn.out_proj.weight": "encoders.9.attn.to_out.weight",735            "conditioner.embedders.1.model.transformer.resblocks.9.ln_1.bias": "encoders.9.layer_norm1.bias",736            "conditioner.embedders.1.model.transformer.resblocks.9.ln_1.weight": "encoders.9.layer_norm1.weight",737            "conditioner.embedders.1.model.transformer.resblocks.9.ln_2.bias": "encoders.9.layer_norm2.bias",738            "conditioner.embedders.1.model.transformer.resblocks.9.ln_2.weight": "encoders.9.layer_norm2.weight",739            "conditioner.embedders.1.model.transformer.resblocks.9.mlp.c_fc.bias": "encoders.9.fc1.bias",740            "conditioner.embedders.1.model.transformer.resblocks.9.mlp.c_fc.weight": "encoders.9.fc1.weight",741            "conditioner.embedders.1.model.transformer.resblocks.9.mlp.c_proj.bias": "encoders.9.fc2.bias",742            "conditioner.embedders.1.model.transformer.resblocks.9.mlp.c_proj.weight": "encoders.9.fc2.weight",743            "conditioner.embedders.1.model.text_projection": "text_projection.weight",744        }745        state_dict_ = {}746        for name in state_dict:747            if name in rename_dict:748                param = state_dict[name]749                if name == "conditioner.embedders.1.model.positional_embedding":750                    param = param.reshape((1, param.shape[0], param.shape[1]))751                elif name == "conditioner.embedders.1.model.text_projection":752                    param = param.T753                if isinstance(rename_dict[name], str):754                    state_dict_[rename_dict[name]] = param755                else:756                    length = param.shape[0] // 3757                    for i, rename in enumerate(rename_dict[name]):758                        state_dict_[rename] = param[i*length: i*length+length]759        return state_dict_