initial consisid test

__init__.py

@@ -1,7 +1,9 @@
 from .nodes import NODE_CLASS_MAPPINGS as NODES_CLASS, NODE_DISPLAY_NAME_MAPPINGS as NODES_DISPLAY
 from .model_loading import NODE_CLASS_MAPPINGS as MODEL_CLASS, NODE_DISPLAY_NAME_MAPPINGS as MODEL_DISPLAY
+from .nodes_consis_id import NODE_CLASS_MAPPINGS as CONSID_ID_CLASS, NODE_DISPLAY_NAME_MAPPINGS as CONSID_ID_DISPLAY
+
+NODE_CLASS_MAPPINGS = {**NODES_CLASS, **MODEL_CLASS, **CONSID_ID_CLASS}
+NODE_DISPLAY_NAME_MAPPINGS = {**NODES_DISPLAY, **MODEL_DISPLAY, **CONSID_ID_DISPLAY}
 
-NODE_CLASS_MAPPINGS = {**NODES_CLASS, **MODEL_CLASS}
-NODE_DISPLAY_NAME_MAPPINGS = {**NODES_DISPLAY, **MODEL_DISPLAY}
 
 __all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]

consis_id/models/eva_clip/__init__.py

@@ -0,0 +1,11 @@
+from .constants import OPENAI_DATASET_MEAN, OPENAI_DATASET_STD
+from .factory import create_model, create_model_and_transforms, create_model_from_pretrained, get_tokenizer, create_transforms
+from .factory import list_models, add_model_config, get_model_config, load_checkpoint
+from .loss import ClipLoss
+from .model import CLIP, CustomCLIP, CLIPTextCfg, CLIPVisionCfg,\
+    convert_weights_to_lp, convert_weights_to_fp16, trace_model, get_cast_dtype
+from .openai import load_openai_model, list_openai_models
+from .pretrained import list_pretrained, list_pretrained_models_by_tag, list_pretrained_tags_by_model,\
+    get_pretrained_url, download_pretrained_from_url, is_pretrained_cfg, get_pretrained_cfg, download_pretrained
+from .tokenizer import SimpleTokenizer, tokenize
+from .transform import image_transform

consis_id/models/eva_clip/constants.py

@@ -0,0 +1,2 @@
+OPENAI_DATASET_MEAN = (0.48145466, 0.4578275, 0.40821073)
+OPENAI_DATASET_STD = (0.26862954, 0.26130258, 0.27577711)

consis_id/models/eva_clip/eva_vit_model.py

@@ -0,0 +1,548 @@
+# --------------------------------------------------------
+# Adapted from  https://github.com/microsoft/unilm/tree/master/beit
+# --------------------------------------------------------
+import math
+import os
+from functools import partial
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+try:
+    from timm.models.layers import drop_path, to_2tuple, trunc_normal_
+except:
+    from timm.layers import drop_path, to_2tuple, trunc_normal_
+    
+from .transformer import PatchDropout
+from .rope import VisionRotaryEmbedding, VisionRotaryEmbeddingFast
+
+if os.getenv('ENV_TYPE') == 'deepspeed':
+    try:
+        from deepspeed.runtime.activation_checkpointing.checkpointing import checkpoint
+    except:
+        from torch.utils.checkpoint import checkpoint
+else:
+    from torch.utils.checkpoint import checkpoint
+
+try:
+    import xformers
+    import xformers.ops as xops
+    XFORMERS_IS_AVAILBLE = True
+except:
+    XFORMERS_IS_AVAILBLE = False
+
+class DropPath(nn.Module):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
+    """
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training)
+    
+    def extra_repr(self) -> str:
+        return 'p={}'.format(self.drop_prob)
+
+
+class Mlp(nn.Module):
+    def __init__(
+        self, 
+        in_features, 
+        hidden_features=None, 
+        out_features=None, 
+        act_layer=nn.GELU, 
+        norm_layer=nn.LayerNorm, 
+        drop=0.,
+        subln=False,
+
+        ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+
+        self.ffn_ln = norm_layer(hidden_features) if subln else nn.Identity()
+
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        # x = self.drop(x)
+        # commit this for the orignal BERT implement 
+        x = self.ffn_ln(x)
+
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+class SwiGLU(nn.Module):
+    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.SiLU, drop=0., 
+                norm_layer=nn.LayerNorm, subln=False):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+
+        self.w1 = nn.Linear(in_features, hidden_features)
+        self.w2 = nn.Linear(in_features, hidden_features)
+
+        self.act = act_layer()
+        self.ffn_ln = norm_layer(hidden_features) if subln else nn.Identity()
+        self.w3 = nn.Linear(hidden_features, out_features)
+        
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x1 = self.w1(x)
+        x2 = self.w2(x)
+        hidden = self.act(x1) * x2
+        x = self.ffn_ln(hidden)
+        x = self.w3(x)
+        x = self.drop(x)
+        return x
+
+class Attention(nn.Module):
+    def __init__(
+            self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0.,
+            proj_drop=0., window_size=None, attn_head_dim=None, xattn=False, rope=None, subln=False, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        if attn_head_dim is not None:
+            head_dim = attn_head_dim
+        all_head_dim = head_dim * self.num_heads
+        self.scale = qk_scale or head_dim ** -0.5
+
+        self.subln = subln
+        if self.subln:
+            self.q_proj = nn.Linear(dim, all_head_dim, bias=False)
+            self.k_proj = nn.Linear(dim, all_head_dim, bias=False)
+            self.v_proj = nn.Linear(dim, all_head_dim, bias=False)
+        else:
+            self.qkv = nn.Linear(dim, all_head_dim * 3, bias=False)
+
+        if qkv_bias:
+            self.q_bias = nn.Parameter(torch.zeros(all_head_dim))
+            self.v_bias = nn.Parameter(torch.zeros(all_head_dim))
+        else:
+            self.q_bias = None
+            self.v_bias = None
+
+        if window_size:
+            self.window_size = window_size
+            self.num_relative_distance = (2 * window_size[0] - 1) * (2 * window_size[1] - 1) + 3
+            self.relative_position_bias_table = nn.Parameter(
+                torch.zeros(self.num_relative_distance, num_heads))  # 2*Wh-1 * 2*Ww-1, nH
+            # cls to token & token 2 cls & cls to cls
+
+            # get pair-wise relative position index for each token inside the window
+            coords_h = torch.arange(window_size[0])
+            coords_w = torch.arange(window_size[1])
+            coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
+            coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
+            relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
+            relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
+            relative_coords[:, :, 0] += window_size[0] - 1  # shift to start from 0
+            relative_coords[:, :, 1] += window_size[1] - 1
+            relative_coords[:, :, 0] *= 2 * window_size[1] - 1
+            relative_position_index = \
+                torch.zeros(size=(window_size[0] * window_size[1] + 1, ) * 2, dtype=relative_coords.dtype)
+            relative_position_index[1:, 1:] = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
+            relative_position_index[0, 0:] = self.num_relative_distance - 3
+            relative_position_index[0:, 0] = self.num_relative_distance - 2
+            relative_position_index[0, 0] = self.num_relative_distance - 1
+
+            self.register_buffer("relative_position_index", relative_position_index)
+        else:
+            self.window_size = None
+            self.relative_position_bias_table = None
+            self.relative_position_index = None
+
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.inner_attn_ln = norm_layer(all_head_dim) if subln else nn.Identity()
+        # self.proj = nn.Linear(all_head_dim, all_head_dim)
+        self.proj = nn.Linear(all_head_dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+        self.xattn = xattn
+        self.xattn_drop = attn_drop
+
+        self.rope = rope
+
+    def forward(self, x, rel_pos_bias=None, attn_mask=None):
+        B, N, C = x.shape
+        if self.subln: 
+            q = F.linear(input=x, weight=self.q_proj.weight, bias=self.q_bias)
+            k = F.linear(input=x, weight=self.k_proj.weight, bias=None)
+            v = F.linear(input=x, weight=self.v_proj.weight, bias=self.v_bias)
+
+            q = q.reshape(B, N, self.num_heads, -1).permute(0, 2, 1, 3)     # B, num_heads, N, C
+            k = k.reshape(B, N, self.num_heads, -1).permute(0, 2, 1, 3)  
+            v = v.reshape(B, N, self.num_heads, -1).permute(0, 2, 1, 3) 
+        else: 
+
+            qkv_bias = None
+            if self.q_bias is not None:
+                qkv_bias = torch.cat((self.q_bias, torch.zeros_like(self.v_bias, requires_grad=False), self.v_bias))
+            
+            qkv = F.linear(input=x, weight=self.qkv.weight, bias=qkv_bias)
+            qkv = qkv.reshape(B, N, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)   # 3, B, num_heads, N, C
+            q, k, v = qkv[0], qkv[1], qkv[2]
+
+        if self.rope:
+            # slightly fast impl
+            q_t = q[:, :, 1:, :]
+            ro_q_t = self.rope(q_t)
+            q = torch.cat((q[:, :, :1, :], ro_q_t), -2).type_as(v)
+
+            k_t = k[:, :, 1:, :]
+            ro_k_t = self.rope(k_t)
+            k = torch.cat((k[:, :, :1, :], ro_k_t), -2).type_as(v)
+
+        if self.xattn:
+            q = q.permute(0, 2, 1, 3)   # B, num_heads, N, C -> B, N, num_heads, C
+            k = k.permute(0, 2, 1, 3)
+            v = v.permute(0, 2, 1, 3)
+
+            x = xops.memory_efficient_attention(
+                q, k, v,
+                p=self.xattn_drop,
+                scale=self.scale,
+                )
+            x = x.reshape(B, N, -1)
+            x = self.inner_attn_ln(x)
+            x = self.proj(x)
+            x = self.proj_drop(x)
+        else:
+            q = q * self.scale
+            attn = (q @ k.transpose(-2, -1))
+
+            if self.relative_position_bias_table is not None:
+                relative_position_bias = \
+                    self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
+                        self.window_size[0] * self.window_size[1] + 1,
+                        self.window_size[0] * self.window_size[1] + 1, -1)  # Wh*Ww,Wh*Ww,nH
+                relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
+                attn = attn + relative_position_bias.unsqueeze(0).type_as(attn)
+
+            if rel_pos_bias is not None:
+                attn = attn + rel_pos_bias.type_as(attn)
+
+            if attn_mask is not None:
+                attn_mask = attn_mask.bool()
+                attn = attn.masked_fill(~attn_mask[:, None, None, :], float("-inf"))
+            
+            attn = attn.softmax(dim=-1)
+            attn = self.attn_drop(attn)
+
+            x = (attn @ v).transpose(1, 2).reshape(B, N, -1)
+            x = self.inner_attn_ln(x)
+            x = self.proj(x)
+            x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Module):
+
+    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., init_values=None, act_layer=nn.GELU, norm_layer=nn.LayerNorm,
+                 window_size=None, attn_head_dim=None, xattn=False, rope=None, postnorm=False,
+                 subln=False, naiveswiglu=False):
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,
+            attn_drop=attn_drop, proj_drop=drop, window_size=window_size, attn_head_dim=attn_head_dim,
+            xattn=xattn, rope=rope, subln=subln, norm_layer=norm_layer)
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+
+        if naiveswiglu:
+            self.mlp = SwiGLU(
+                in_features=dim, 
+                hidden_features=mlp_hidden_dim, 
+                subln=subln,
+                norm_layer=norm_layer,
+            )
+        else:
+            self.mlp = Mlp(
+                in_features=dim, 
+                hidden_features=mlp_hidden_dim, 
+                act_layer=act_layer,
+                subln=subln,
+                drop=drop
+            )
+
+        if init_values is not None and init_values > 0:
+            self.gamma_1 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)
+            self.gamma_2 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)
+        else:
+            self.gamma_1, self.gamma_2 = None, None
+
+        self.postnorm = postnorm
+
+    def forward(self, x, rel_pos_bias=None, attn_mask=None):
+        if self.gamma_1 is None:
+            if self.postnorm:
+                x = x + self.drop_path(self.norm1(self.attn(x, rel_pos_bias=rel_pos_bias, attn_mask=attn_mask)))
+                x = x + self.drop_path(self.norm2(self.mlp(x)))
+            else:
+                x = x + self.drop_path(self.attn(self.norm1(x), rel_pos_bias=rel_pos_bias, attn_mask=attn_mask))
+                x = x + self.drop_path(self.mlp(self.norm2(x)))
+        else:
+            if self.postnorm:
+                x = x + self.drop_path(self.gamma_1 * self.norm1(self.attn(x, rel_pos_bias=rel_pos_bias, attn_mask=attn_mask)))
+                x = x + self.drop_path(self.gamma_2 * self.norm2(self.mlp(x)))
+            else:
+                x = x + self.drop_path(self.gamma_1 * self.attn(self.norm1(x), rel_pos_bias=rel_pos_bias, attn_mask=attn_mask))
+                x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x)))
+        return x
+
+
+class PatchEmbed(nn.Module):
+    """ Image to Patch Embedding
+    """
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
+        super().__init__()
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        num_patches = (img_size[1] // patch_size[1]) * (img_size[0] // patch_size[0])
+        self.patch_shape = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.num_patches = num_patches
+
+        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
+
+    def forward(self, x, **kwargs):
+        B, C, H, W = x.shape
+        # FIXME look at relaxing size constraints
+        assert H == self.img_size[0] and W == self.img_size[1], \
+            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
+        x = self.proj(x).flatten(2).transpose(1, 2)
+        return x
+
+
+class RelativePositionBias(nn.Module):
+
+    def __init__(self, window_size, num_heads):
+        super().__init__()
+        self.window_size = window_size
+        self.num_relative_distance = (2 * window_size[0] - 1) * (2 * window_size[1] - 1) + 3
+        self.relative_position_bias_table = nn.Parameter(
+            torch.zeros(self.num_relative_distance, num_heads))  # 2*Wh-1 * 2*Ww-1, nH
+        # cls to token & token 2 cls & cls to cls
+
+        # get pair-wise relative position index for each token inside the window
+        coords_h = torch.arange(window_size[0])
+        coords_w = torch.arange(window_size[1])
+        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
+        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
+        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
+        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
+        relative_coords[:, :, 0] += window_size[0] - 1  # shift to start from 0
+        relative_coords[:, :, 1] += window_size[1] - 1
+        relative_coords[:, :, 0] *= 2 * window_size[1] - 1
+        relative_position_index = \
+            torch.zeros(size=(window_size[0] * window_size[1] + 1,) * 2, dtype=relative_coords.dtype)
+        relative_position_index[1:, 1:] = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
+        relative_position_index[0, 0:] = self.num_relative_distance - 3
+        relative_position_index[0:, 0] = self.num_relative_distance - 2
+        relative_position_index[0, 0] = self.num_relative_distance - 1
+
+        self.register_buffer("relative_position_index", relative_position_index)
+
+    def forward(self):
+        relative_position_bias = \
+            self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
+                self.window_size[0] * self.window_size[1] + 1,
+                self.window_size[0] * self.window_size[1] + 1, -1)  # Wh*Ww,Wh*Ww,nH
+        return relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
+
+
+class EVAVisionTransformer(nn.Module):
+    """ Vision Transformer with support for patch or hybrid CNN input stage
+    """
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
+                 num_heads=12, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop_rate=0., attn_drop_rate=0.,
+                 drop_path_rate=0., norm_layer=nn.LayerNorm, init_values=None, patch_dropout=0.,
+                 use_abs_pos_emb=True, use_rel_pos_bias=False, use_shared_rel_pos_bias=False, rope=False,
+                 use_mean_pooling=True, init_scale=0.001, grad_checkpointing=False, xattn=False, postnorm=False,
+                 pt_hw_seq_len=16, intp_freq=False, naiveswiglu=False, subln=False):
+        super().__init__()
+
+        if not XFORMERS_IS_AVAILBLE:
+            xattn = False
+
+        self.image_size = img_size
+        self.num_classes = num_classes
+        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
+
+        self.patch_embed = PatchEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)
+        num_patches = self.patch_embed.num_patches
+
+        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
+        # self.mask_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
+        if use_abs_pos_emb:
+            self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
+        else:
+            self.pos_embed = None
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        if use_shared_rel_pos_bias:
+            self.rel_pos_bias = RelativePositionBias(window_size=self.patch_embed.patch_shape, num_heads=num_heads)
+        else:
+            self.rel_pos_bias = None
+
+        if rope:
+            half_head_dim = embed_dim // num_heads // 2
+            hw_seq_len = img_size // patch_size
+            self.rope = VisionRotaryEmbeddingFast(
+                dim=half_head_dim,
+                pt_seq_len=pt_hw_seq_len,
+                ft_seq_len=hw_seq_len if intp_freq else None,
+                # patch_dropout=patch_dropout
+            )
+        else: 
+            self.rope = None
+
+        self.naiveswiglu = naiveswiglu
+
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
+        self.use_rel_pos_bias = use_rel_pos_bias
+        self.blocks = nn.ModuleList([
+            Block(
+                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
+                init_values=init_values, window_size=self.patch_embed.patch_shape if use_rel_pos_bias else None,
+                xattn=xattn, rope=self.rope, postnorm=postnorm, subln=subln, naiveswiglu=naiveswiglu)
+            for i in range(depth)])
+        self.norm = nn.Identity() if use_mean_pooling else norm_layer(embed_dim)
+        self.fc_norm = norm_layer(embed_dim) if use_mean_pooling else None
+        self.head = nn.Linear(embed_dim, num_classes) if num_classes > 0 else nn.Identity()
+
+        if self.pos_embed is not None:
+            trunc_normal_(self.pos_embed, std=.02)
+
+        trunc_normal_(self.cls_token, std=.02)
+        # trunc_normal_(self.mask_token, std=.02)
+
+        self.apply(self._init_weights)
+        self.fix_init_weight()
+
+        if isinstance(self.head, nn.Linear):
+            trunc_normal_(self.head.weight, std=.02)
+            self.head.weight.data.mul_(init_scale)
+            self.head.bias.data.mul_(init_scale)
+
+        # setting a patch_dropout of 0. would mean it is disabled and this function would be the identity fn
+        self.patch_dropout = PatchDropout(patch_dropout) if patch_dropout > 0. else nn.Identity()
+
+        self.grad_checkpointing = grad_checkpointing
+
+    def fix_init_weight(self):
+        def rescale(param, layer_id):
+            param.div_(math.sqrt(2.0 * layer_id))
+
+        for layer_id, layer in enumerate(self.blocks):
+            rescale(layer.attn.proj.weight.data, layer_id + 1)
+            if self.naiveswiglu:
+                rescale(layer.mlp.w3.weight.data, layer_id + 1)
+            else:
+                rescale(layer.mlp.fc2.weight.data, layer_id + 1)
+
+    def get_cast_dtype(self) -> torch.dtype:
+        return self.blocks[0].mlp.fc2.weight.dtype
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=.02)
+            if m.bias is not None:
+                nn.init.constant_(m.bias, 0)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+
+    def get_num_layers(self):
+        return len(self.blocks)
+    
+    def lock(self, unlocked_groups=0, freeze_bn_stats=False):
+        assert unlocked_groups == 0, 'partial locking not currently supported for this model'
+        for param in self.parameters():
+            param.requires_grad = False
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        self.grad_checkpointing = enable
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'pos_embed', 'cls_token'}
+
+    def get_classifier(self):
+        return self.head
+
+    def reset_classifier(self, num_classes, global_pool=''):
+        self.num_classes = num_classes
+        self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()
+
+    def forward_features(self, x, return_all_features=False, return_hidden=False, shuffle=False):
+        
+        x = self.patch_embed(x)
+        batch_size, seq_len, _ = x.size()
+
+        if shuffle:
+            idx = torch.randperm(x.shape[1]) + 1
+            zero = torch.LongTensor([0, ])
+            idx = torch.cat([zero, idx])
+            pos_embed = self.pos_embed[:, idx]
+
+        cls_tokens = self.cls_token.expand(batch_size, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
+        x = torch.cat((cls_tokens, x), dim=1)
+        if shuffle:
+            x = x + pos_embed
+        elif self.pos_embed is not None:
+            x = x + self.pos_embed
+        x = self.pos_drop(x)
+
+        # a patch_dropout of 0. would mean it is disabled and this function would do nothing but return what was passed in
+        if os.getenv('RoPE') == '1':
+            if self.training and not isinstance(self.patch_dropout, nn.Identity):
+                x, patch_indices_keep = self.patch_dropout(x)
+                self.rope.forward = partial(self.rope.forward, patch_indices_keep=patch_indices_keep)
+            else:
+                self.rope.forward = partial(self.rope.forward, patch_indices_keep=None)
+                x = self.patch_dropout(x)
+        else:
+            x = self.patch_dropout(x)
+
+        rel_pos_bias = self.rel_pos_bias() if self.rel_pos_bias is not None else None
+        hidden_states = []
+        for idx, blk in enumerate(self.blocks):
+            if (0 < idx <= 20) and (idx % 4 == 0) and return_hidden:
+                hidden_states.append(x)
+            if self.grad_checkpointing:
+                x = checkpoint(blk, x, (rel_pos_bias,))
+            else:
+                x = blk(x, rel_pos_bias=rel_pos_bias)
+
+        if not return_all_features:
+            x = self.norm(x)
+            if self.fc_norm is not None:
+                return self.fc_norm(x.mean(1)), hidden_states
+            else:
+                return x[:, 0], hidden_states
+        return x
+
+    def forward(self, x, return_all_features=False, return_hidden=False, shuffle=False):
+        if return_all_features:
+            return self.forward_features(x, return_all_features, return_hidden, shuffle)
+        x, hidden_states = self.forward_features(x, return_all_features, return_hidden, shuffle)
+        x = self.head(x)
+        if return_hidden:
+            return x, hidden_states
+        return x

consis_id/models/eva_clip/factory.py

@@ -0,0 +1,517 @@
+import json
+import logging
+import os
+import pathlib
+import re
+from copy import deepcopy
+from pathlib import Path
+from typing import Optional, Tuple, Union, Dict, Any
+import torch
+
+from .constants import OPENAI_DATASET_MEAN, OPENAI_DATASET_STD
+from .model import CLIP, CustomCLIP, convert_weights_to_lp, convert_to_custom_text_state_dict,\
+    get_cast_dtype
+from .openai import load_openai_model
+from .pretrained import is_pretrained_cfg, get_pretrained_cfg, download_pretrained, list_pretrained_tags_by_model
+from .transform import image_transform
+from .tokenizer import HFTokenizer, tokenize
+from .utils import resize_clip_pos_embed, resize_evaclip_pos_embed, resize_visual_pos_embed, resize_eva_pos_embed
+
+
+_MODEL_CONFIG_PATHS = [Path(__file__).parent / f"model_configs/"]
+_MODEL_CONFIGS = {}  # directory (model_name: config) of model architecture configs
+
+
+def _natural_key(string_):
+    return [int(s) if s.isdigit() else s for s in re.split(r'(\d+)', string_.lower())]
+
+
+def _rescan_model_configs():
+    global _MODEL_CONFIGS
+
+    config_ext = ('.json',)
+    config_files = []
+    for config_path in _MODEL_CONFIG_PATHS:
+        if config_path.is_file() and config_path.suffix in config_ext:
+            config_files.append(config_path)
+        elif config_path.is_dir():
+            for ext in config_ext:
+                config_files.extend(config_path.glob(f'*{ext}'))
+
+    for cf in config_files:
+        with open(cf, "r", encoding="utf8") as f:
+            model_cfg = json.load(f)
+            if all(a in model_cfg for a in ('embed_dim', 'vision_cfg', 'text_cfg')):
+                _MODEL_CONFIGS[cf.stem] = model_cfg
+
+    _MODEL_CONFIGS = dict(sorted(_MODEL_CONFIGS.items(), key=lambda x: _natural_key(x[0])))
+
+
+_rescan_model_configs()  # initial populate of model config registry
+
+
+def list_models():
+    """ enumerate available model architectures based on config files """
+    return list(_MODEL_CONFIGS.keys())
+
+
+def add_model_config(path):
+    """ add model config path or file and update registry """
+    if not isinstance(path, Path):
+        path = Path(path)
+    _MODEL_CONFIG_PATHS.append(path)
+    _rescan_model_configs()
+
+
+def get_model_config(model_name):
+    if model_name in _MODEL_CONFIGS:
+        return deepcopy(_MODEL_CONFIGS[model_name])
+    else:
+        return None
+
+
+def get_tokenizer(model_name):
+    config = get_model_config(model_name)
+    tokenizer = HFTokenizer(config['text_cfg']['hf_tokenizer_name']) if 'hf_tokenizer_name' in config['text_cfg'] else tokenize
+    return tokenizer
+
+
+# loading openai CLIP weights when is_openai=True for training
+def load_state_dict(checkpoint_path: str, map_location: str='cpu', model_key: str='model|module|state_dict', is_openai: bool=False, skip_list: list=[]):
+    if is_openai:
+        model = torch.jit.load(checkpoint_path, map_location="cpu").eval()
+        state_dict = model.state_dict()
+        for key in ["input_resolution", "context_length", "vocab_size"]:
+            state_dict.pop(key, None)
+    else:
+        checkpoint = torch.load(checkpoint_path, map_location=map_location)
+        for mk in model_key.split('|'):
+            if isinstance(checkpoint, dict) and mk in checkpoint:
+                state_dict = checkpoint[mk]
+                break
+            else:
+                state_dict = checkpoint
+        if next(iter(state_dict.items()))[0].startswith('module'):
+            state_dict = {k[7:]: v for k, v in state_dict.items()}
+    
+    for k in skip_list:
+        if k in list(state_dict.keys()):
+            logging.info(f"Removing key {k} from pretrained checkpoint")
+            del state_dict[k]
+
+    if os.getenv('RoPE') == '1':
+        for k in list(state_dict.keys()):
+            if 'freqs_cos' in k or 'freqs_sin' in k:
+                del state_dict[k]
+    return state_dict
+
+
+
+def load_checkpoint(model, checkpoint_path, model_key="model|module|state_dict", strict=True):
+    state_dict = load_state_dict(checkpoint_path, model_key=model_key, is_openai=False)
+    # detect old format and make compatible with new format
+    if 'positional_embedding' in state_dict and not hasattr(model, 'positional_embedding'):
+        state_dict = convert_to_custom_text_state_dict(state_dict)
+    if 'text.logit_scale' in state_dict and hasattr(model, 'logit_scale'):
+        state_dict['logit_scale'] = state_dict['text.logit_scale']
+        del state_dict['text.logit_scale']
+
+    # resize_clip_pos_embed for CLIP and open CLIP
+    if 'visual.positional_embedding' in state_dict:
+        resize_clip_pos_embed(state_dict, model)
+    # specified to eva_vit_model
+    elif 'visual.pos_embed' in state_dict:
+        resize_evaclip_pos_embed(state_dict, model)
+
+    # resize_clip_pos_embed(state_dict, model)
+    incompatible_keys = model.load_state_dict(state_dict, strict=strict)
+    logging.info(f"incompatible_keys.missing_keys: {incompatible_keys.missing_keys}")
+    return incompatible_keys
+
+def load_clip_visual_state_dict(checkpoint_path: str, map_location: str='cpu', is_openai: bool=False, skip_list:list=[]):
+    state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
+
+    for k in list(state_dict.keys()):
+        if not k.startswith('visual.'):
+            del state_dict[k]
+    for k in list(state_dict.keys()):
+        if k.startswith('visual.'):
+            new_k = k[7:]
+            state_dict[new_k] = state_dict[k]
+            del state_dict[k]
+    return state_dict
+
+def load_clip_text_state_dict(checkpoint_path: str, map_location: str='cpu', is_openai: bool=False, skip_list:list=[]):
+    state_dict = load_state_dict(checkpoint_path, map_location=map_location, is_openai=is_openai, skip_list=skip_list)
+
+    for k in list(state_dict.keys()):
+        if k.startswith('visual.'):
+            del state_dict[k]
+    return state_dict
+
+def get_pretrained_tag(pretrained_model):
+    pretrained_model = pretrained_model.lower()
+    if "laion" in pretrained_model or "open_clip" in pretrained_model:
+        return "open_clip"
+    elif "openai" in pretrained_model:
+        return "clip"
+    elif "eva" in pretrained_model and "clip" in pretrained_model:
+        return "eva_clip"
+    else:
+        return "other"
+
+def load_pretrained_checkpoint(
+        model,
+        visual_checkpoint_path,
+        text_checkpoint_path,
+        strict=True,
+        visual_model=None,
+        text_model=None,
+        model_key="model|module|state_dict",
+        skip_list=[]):
+    visual_tag = get_pretrained_tag(visual_model)
+    text_tag = get_pretrained_tag(text_model)
+
+    logging.info(f"num of model state_dict keys: {len(model.state_dict().keys())}")
+    visual_incompatible_keys, text_incompatible_keys = None, None
+    if visual_checkpoint_path:
+        if visual_tag == "eva_clip" or visual_tag == "open_clip":
+            visual_state_dict = load_clip_visual_state_dict(visual_checkpoint_path, is_openai=False, skip_list=skip_list)
+        elif visual_tag == "clip":
+            visual_state_dict = load_clip_visual_state_dict(visual_checkpoint_path, is_openai=True, skip_list=skip_list)
+        else:
+            visual_state_dict = load_state_dict(visual_checkpoint_path, model_key=model_key, is_openai=False, skip_list=skip_list)
+    
+        # resize_clip_pos_embed for CLIP and open CLIP
+        if 'positional_embedding' in visual_state_dict:
+            resize_visual_pos_embed(visual_state_dict, model)
+        # specified to EVA model
+        elif 'pos_embed' in visual_state_dict:
+            resize_eva_pos_embed(visual_state_dict, model)
+
+        visual_incompatible_keys = model.visual.load_state_dict(visual_state_dict, strict=strict)
+        logging.info(f"num of loaded visual_state_dict keys: {len(visual_state_dict.keys())}")
+        logging.info(f"visual_incompatible_keys.missing_keys: {visual_incompatible_keys.missing_keys}")
+
+    if text_checkpoint_path:
+        if text_tag == "eva_clip" or text_tag == "open_clip":
+            text_state_dict = load_clip_text_state_dict(text_checkpoint_path, is_openai=False, skip_list=skip_list)
+        elif text_tag == "clip":
+            text_state_dict = load_clip_text_state_dict(text_checkpoint_path, is_openai=True, skip_list=skip_list)
+        else:
+            text_state_dict = load_state_dict(visual_checkpoint_path, model_key=model_key, is_openai=False, skip_list=skip_list)
+
+        text_incompatible_keys = model.text.load_state_dict(text_state_dict, strict=strict)
+        
+        logging.info(f"num of loaded text_state_dict keys: {len(text_state_dict.keys())}")
+        logging.info(f"text_incompatible_keys.missing_keys: {text_incompatible_keys.missing_keys}")
+
+    return visual_incompatible_keys, text_incompatible_keys
+
+def create_model(
+        model_name: str,
+        pretrained: Optional[str] = None,
+        precision: str = 'fp32',
+        device: Union[str, torch.device] = 'cpu',
+        jit: bool = False,
+        force_quick_gelu: bool = False,
+        force_custom_clip: bool = False,
+        force_patch_dropout: Optional[float] = None,
+        pretrained_image: str = '',
+        pretrained_text: str = '',
+        pretrained_hf: bool = True,
+        pretrained_visual_model: str = None,
+        pretrained_text_model: str = None,
+        cache_dir: Optional[str] = None,
+        skip_list: list  = [],
+):
+    model_name = model_name.replace('/', '-')  # for callers using old naming with / in ViT names
+    if isinstance(device, str):
+        device = torch.device(device)
+
+    if pretrained and pretrained.lower() == 'openai':
+        logging.info(f'Loading pretrained {model_name} from OpenAI.')
+        model = load_openai_model(
+            model_name,
+            precision=precision,
+            device=device,
+            jit=jit,
+            cache_dir=cache_dir,
+        )
+    else:
+        model_cfg = get_model_config(model_name)
+        if model_cfg is not None:
+            logging.info(f'Loaded {model_name} model config.')
+        else:
+            logging.error(f'Model config for {model_name} not found; available models {list_models()}.')
+            raise RuntimeError(f'Model config for {model_name} not found.')
+
+        if 'rope' in model_cfg.get('vision_cfg', {}):
+            if model_cfg['vision_cfg']['rope']:
+                os.environ['RoPE'] = "1"
+        else:
+            os.environ['RoPE'] = "0"
+
+        if force_quick_gelu:
+            # override for use of QuickGELU on non-OpenAI transformer models
+            model_cfg["quick_gelu"] = True
+        
+        if force_patch_dropout is not None:
+            # override the default patch dropout value
+            model_cfg['vision_cfg']["patch_dropout"] = force_patch_dropout
+
+        cast_dtype = get_cast_dtype(precision)
+        custom_clip = model_cfg.pop('custom_text', False) or force_custom_clip or ('hf_model_name' in model_cfg['text_cfg'])
+
+
+        if custom_clip:
+            if 'hf_model_name' in model_cfg.get('text_cfg', {}):
+                model_cfg['text_cfg']['hf_model_pretrained'] = pretrained_hf
+            model = CustomCLIP(**model_cfg, cast_dtype=cast_dtype)
+        else:
+            model = CLIP(**model_cfg, cast_dtype=cast_dtype)
+
+        pretrained_cfg = {}
+        if pretrained:
+            checkpoint_path = ''
+            pretrained_cfg = get_pretrained_cfg(model_name, pretrained)
+            if pretrained_cfg:
+                checkpoint_path = download_pretrained(pretrained_cfg, cache_dir=cache_dir)
+            elif os.path.exists(pretrained):
+                checkpoint_path = pretrained
+
+            if checkpoint_path:
+                logging.info(f'Loading pretrained {model_name} weights ({pretrained}).')
+                load_checkpoint(model,
+                               checkpoint_path,
+                               model_key="model|module|state_dict",
+                               strict=False
+                               ) 
+            else:
+                error_str = (
+                    f'Pretrained weights ({pretrained}) not found for model {model_name}.'
+                    f'Available pretrained tags ({list_pretrained_tags_by_model(model_name)}.')
+                logging.warning(error_str)
+                raise RuntimeError(error_str)
+        else:
+            visual_checkpoint_path = ''
+            text_checkpoint_path = ''
+            
+            if pretrained_image:
+                pretrained_visual_model = pretrained_visual_model.replace('/', '-')  # for callers using old naming with / in ViT names
+                pretrained_image_cfg = get_pretrained_cfg(pretrained_visual_model, pretrained_image)
+                if 'timm_model_name' in model_cfg.get('vision_cfg', {}):
+                    # pretrained weight loading for timm models set via vision_cfg
+                    model_cfg['vision_cfg']['timm_model_pretrained'] = True
+                elif pretrained_image_cfg:
+                    visual_checkpoint_path = download_pretrained(pretrained_image_cfg, cache_dir=cache_dir)
+                elif os.path.exists(pretrained_image):
+                    visual_checkpoint_path = pretrained_image
+                else:
+                    logging.warning(f'Pretrained weights ({visual_checkpoint_path}) not found for model {model_name}.visual.')
+                    raise RuntimeError(f'Pretrained weights ({visual_checkpoint_path}) not found for model {model_name}.visual.')
+
+            if pretrained_text:
+                pretrained_text_model = pretrained_text_model.replace('/', '-')  # for callers using old naming with / in ViT names
+                pretrained_text_cfg = get_pretrained_cfg(pretrained_text_model, pretrained_text)
+                if pretrained_image_cfg:
+                    text_checkpoint_path = download_pretrained(pretrained_text_cfg, cache_dir=cache_dir)
+                elif os.path.exists(pretrained_text):
+                    text_checkpoint_path = pretrained_text
+                else:
+                    logging.warning(f'Pretrained weights ({text_checkpoint_path}) not found for model {model_name}.text.')
+                    raise RuntimeError(f'Pretrained weights ({text_checkpoint_path}) not found for model {model_name}.text.')
+            
+            if visual_checkpoint_path:
+                logging.info(f'Loading pretrained {model_name}.visual weights ({visual_checkpoint_path}).')
+            if text_checkpoint_path:
+                logging.info(f'Loading pretrained {model_name}.text weights ({text_checkpoint_path}).')
+
+            if visual_checkpoint_path or text_checkpoint_path:
+                load_pretrained_checkpoint(
+                    model,
+                    visual_checkpoint_path,
+                    text_checkpoint_path,
+                    strict=False,
+                    visual_model=pretrained_visual_model,
+                    text_model=pretrained_text_model,
+                    model_key="model|module|state_dict",
+                    skip_list=skip_list
+                )
+        
+        if "fp16" in precision or "bf16" in precision:
+            logging.info(f'convert precision to {precision}')
+            model = model.to(torch.bfloat16) if 'bf16' in precision else model.to(torch.float16)
+
+        model.to(device=device)
+
+        # set image / mean metadata from pretrained_cfg if available, or use default
+        model.visual.image_mean = pretrained_cfg.get('mean', None) or OPENAI_DATASET_MEAN
+        model.visual.image_std = pretrained_cfg.get('std', None) or OPENAI_DATASET_STD
+
+        if jit:
+            model = torch.jit.script(model)
+
+    return model
+
+
+def create_model_and_transforms(
+        model_name: str,
+        pretrained: Optional[str] = None,
+        precision: str = 'fp32',
+        device: Union[str, torch.device] = 'cpu',
+        jit: bool = False,
+        force_quick_gelu: bool = False,
+        force_custom_clip: bool = False,
+        force_patch_dropout: Optional[float] = None,
+        pretrained_image: str = '',
+        pretrained_text: str = '',
+        pretrained_hf: bool = True,
+        pretrained_visual_model: str = None,
+        pretrained_text_model: str = None,
+        image_mean: Optional[Tuple[float, ...]] = None,
+        image_std: Optional[Tuple[float, ...]] = None,
+        cache_dir: Optional[str] = None,
+        skip_list: list = [],
+):
+    model = create_model(
+        model_name,
+        pretrained,
+        precision=precision,
+        device=device,
+        jit=jit,
+        force_quick_gelu=force_quick_gelu,
+        force_custom_clip=force_custom_clip,
+        force_patch_dropout=force_patch_dropout,
+        pretrained_image=pretrained_image,
+        pretrained_text=pretrained_text,
+        pretrained_hf=pretrained_hf,
+        pretrained_visual_model=pretrained_visual_model,
+        pretrained_text_model=pretrained_text_model,
+        cache_dir=cache_dir,
+        skip_list=skip_list,
+    )
+
+    image_mean = image_mean or getattr(model.visual, 'image_mean', None)
+    image_std = image_std or getattr(model.visual, 'image_std', None)
+    preprocess_train = image_transform(
+        model.visual.image_size,
+        is_train=True,
+        mean=image_mean,
+        std=image_std
+    )
+    preprocess_val = image_transform(
+        model.visual.image_size,
+        is_train=False,
+        mean=image_mean,
+        std=image_std
+    )
+
+    return model, preprocess_train, preprocess_val
+
+
+def create_transforms(
+        model_name: str,
+        pretrained: Optional[str] = None,
+        precision: str = 'fp32',
+        device: Union[str, torch.device] = 'cpu',
+        jit: bool = False,
+        force_quick_gelu: bool = False,
+        force_custom_clip: bool = False,
+        force_patch_dropout: Optional[float] = None,
+        pretrained_image: str = '',
+        pretrained_text: str = '',
+        pretrained_hf: bool = True,
+        pretrained_visual_model: str = None,
+        pretrained_text_model: str = None,
+        image_mean: Optional[Tuple[float, ...]] = None,
+        image_std: Optional[Tuple[float, ...]] = None,
+        cache_dir: Optional[str] = None,
+        skip_list: list = [],
+):
+    model = create_model(
+        model_name,
+        pretrained,
+        precision=precision,
+        device=device,
+        jit=jit,
+        force_quick_gelu=force_quick_gelu,
+        force_custom_clip=force_custom_clip,
+        force_patch_dropout=force_patch_dropout,
+        pretrained_image=pretrained_image,
+        pretrained_text=pretrained_text,
+        pretrained_hf=pretrained_hf,
+        pretrained_visual_model=pretrained_visual_model,
+        pretrained_text_model=pretrained_text_model,
+        cache_dir=cache_dir,
+        skip_list=skip_list,
+    )
+
+
+    image_mean = image_mean or getattr(model.visual, 'image_mean', None)
+    image_std = image_std or getattr(model.visual, 'image_std', None)
+    preprocess_train = image_transform(
+        model.visual.image_size,
+        is_train=True,
+        mean=image_mean,
+        std=image_std
+    )
+    preprocess_val = image_transform(
+        model.visual.image_size,
+        is_train=False,
+        mean=image_mean,
+        std=image_std
+    )
+    del model
+
+    return preprocess_train, preprocess_val
+
+def create_model_from_pretrained(
+        model_name: str,
+        pretrained: str,
+        precision: str = 'fp32',
+        device: Union[str, torch.device] = 'cpu',
+        jit: bool = False,
+        force_quick_gelu: bool = False,
+        force_custom_clip: bool = False,
+        force_patch_dropout: Optional[float] = None,
+        return_transform: bool = True,
+        image_mean: Optional[Tuple[float, ...]] = None,
+        image_std: Optional[Tuple[float, ...]] = None,
+        cache_dir: Optional[str] = None,
+        is_frozen: bool = False,
+):
+    if not is_pretrained_cfg(model_name, pretrained) and not os.path.exists(pretrained):
+        raise RuntimeError(
+            f'{pretrained} is not a valid pretrained cfg or checkpoint for {model_name}.'
+            f' Use open_clip.list_pretrained() to find one.')
+
+    model = create_model(
+        model_name,
+        pretrained,
+        precision=precision,
+        device=device,
+        jit=jit,
+        force_quick_gelu=force_quick_gelu,
+        force_custom_clip=force_custom_clip,
+        force_patch_dropout=force_patch_dropout,
+        cache_dir=cache_dir,
+    )
+
+    if is_frozen:
+        for param in model.parameters():
+            param.requires_grad = False
+
+    if not return_transform:
+        return model
+
+    image_mean = image_mean or getattr(model.visual, 'image_mean', None)
+    image_std = image_std or getattr(model.visual, 'image_std', None)
+    preprocess = image_transform(
+        model.visual.image_size,
+        is_train=False,
+        mean=image_mean,
+        std=image_std
+    )
+
+    return model, preprocess

consis_id/models/eva_clip/hf_configs.py

@@ -0,0 +1,57 @@
+# HF architecture dict:
+arch_dict = {
+  # https://huggingface.co/docs/transformers/model_doc/roberta#roberta
+  "roberta": {
+      "config_names": {
+          "context_length": "max_position_embeddings",
+          "vocab_size": "vocab_size",
+          "width": "hidden_size",
+          "heads": "num_attention_heads",
+          "layers": "num_hidden_layers",
+          "layer_attr": "layer",
+          "token_embeddings_attr": "embeddings"
+      },
+      "pooler": "mean_pooler",
+  },
+  # https://huggingface.co/docs/transformers/model_doc/xlm-roberta#transformers.XLMRobertaConfig
+  "xlm-roberta": {
+      "config_names": {
+          "context_length": "max_position_embeddings",
+          "vocab_size": "vocab_size",
+          "width": "hidden_size",
+          "heads": "num_attention_heads",
+          "layers": "num_hidden_layers",
+          "layer_attr": "layer",
+          "token_embeddings_attr": "embeddings"
+      },
+      "pooler": "mean_pooler",
+  },
+  # https://huggingface.co/docs/transformers/model_doc/mt5#mt5
+  "mt5": {
+      "config_names": {
+          # unlimited seqlen
+          # https://github.com/google-research/text-to-text-transfer-transformer/issues/273
+          # https://github.com/huggingface/transformers/blob/v4.24.0/src/transformers/models/t5/modeling_t5.py#L374
+          "context_length": "",
+          "vocab_size": "vocab_size",
+          "width": "d_model",
+          "heads": "num_heads",
+          "layers": "num_layers",
+          "layer_attr": "block",
+          "token_embeddings_attr": "embed_tokens"
+      },
+      "pooler": "mean_pooler",
+  },
+  "bert": {
+    "config_names": {
+      "context_length": "max_position_embeddings",
+      "vocab_size": "vocab_size",
+      "width": "hidden_size",
+      "heads": "num_attention_heads",
+      "layers": "num_hidden_layers",
+      "layer_attr": "layer",
+      "token_embeddings_attr": "embeddings"
+    },
+    "pooler": "mean_pooler",
+  }
+}

consis_id/models/eva_clip/hf_model.py

@@ -0,0 +1,248 @@
+""" huggingface model adapter
+
+Wraps HuggingFace transformers (https://github.com/huggingface/transformers) models for use as a text tower in CLIP model.
+"""
+
+import re
+
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from torch import TensorType
+try:
+    import transformers
+    from transformers import AutoModel, AutoModelForMaskedLM, AutoTokenizer, AutoConfig, PretrainedConfig
+    from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling, \
+        BaseModelOutputWithPoolingAndCrossAttentions
+except ImportError as e:
+    transformers = None
+
+
+    class BaseModelOutput:
+        pass
+
+
+    class PretrainedConfig:
+        pass
+
+from .hf_configs import arch_dict
+
+# utils
+def _camel2snake(s):
+    return re.sub(r'(?<!^)(?=[A-Z])', '_', s).lower()
+
+# TODO: ?last - for gpt-like models
+_POOLERS = {}
+
+def register_pooler(cls):
+    """Decorator registering pooler class"""
+    _POOLERS[_camel2snake(cls.__name__)] = cls
+    return cls
+
+
+@register_pooler
+class MeanPooler(nn.Module):
+    """Mean pooling"""
+    def forward(self, x:BaseModelOutput, attention_mask:TensorType):
+        masked_output = x.last_hidden_state * attention_mask.unsqueeze(-1)
+        return masked_output.sum(dim=1) / attention_mask.sum(-1, keepdim=True)
+
+@register_pooler
+class MaxPooler(nn.Module):
+    """Max pooling"""
+    def forward(self, x:BaseModelOutput, attention_mask:TensorType):
+        masked_output = x.last_hidden_state.masked_fill(attention_mask.unsqueeze(-1), -torch.inf)
+        return masked_output.max(1).values
+
+@register_pooler
+class ClsPooler(nn.Module):
+    """CLS token pooling"""
+    def __init__(self, use_pooler_output=True):
+        super().__init__()
+        self.cls_token_position = 0
+        self.use_pooler_output = use_pooler_output
+
+    def forward(self, x:BaseModelOutput, attention_mask:TensorType):
+        
+        if (self.use_pooler_output and 
+            isinstance(x, (BaseModelOutputWithPooling, BaseModelOutputWithPoolingAndCrossAttentions)) and
+            (x.pooler_output is not None)
+            ):
+            return x.pooler_output
+        
+        return x.last_hidden_state[:, self.cls_token_position, :]
+
+class HFTextEncoder(nn.Module):
+    """HuggingFace model adapter"""
+    def __init__(
+            self, 
+            model_name_or_path: str,
+            output_dim: int,
+            tokenizer_name: str = None,
+            config: PretrainedConfig = None,
+            pooler_type: str = None,
+            proj: str = None,
+            pretrained: bool = True,
+            masked_language_modeling: bool = False):
+        super().__init__()
+
+        self.output_dim = output_dim
+
+        # TODO: find better way to get this information
+        uses_transformer_pooler = (pooler_type == "cls_pooler")
+
+        if transformers is None:
+            raise RuntimeError("Please `pip install transformers` to use pre-trained HuggingFace models")
+        if config is None:
+            self.config = AutoConfig.from_pretrained(model_name_or_path)
+            if masked_language_modeling:
+                create_func, model_args = (AutoModelForMaskedLM.from_pretrained, model_name_or_path) if pretrained else (
+                    AutoModelForMaskedLM.from_config, self.config)
+            else:
+                create_func, model_args = (AutoModel.from_pretrained, model_name_or_path) if pretrained else (
+                    AutoModel.from_config, self.config)
+            # TODO: do all model configs have this attribute? PretrainedConfig does so yes??
+            if hasattr(self.config, "is_encoder_decoder") and self.config.is_encoder_decoder:
+                self.transformer = create_func(model_args)
+                self.transformer = self.transformer.encoder
+            else:
+                self.transformer = create_func(model_args, add_pooling_layer=uses_transformer_pooler)
+        else:
+            self.config = config
+            if masked_language_modeling:
+                self.transformer = AutoModelForMaskedLM.from_config(config)
+            else:
+                self.transformer = AutoModel.from_config(config)
+
+        if pooler_type is None: # get default arch pooler
+            self.pooler = _POOLERS[(arch_dict[self.config.model_type]["pooler"])]()
+        else:
+            self.pooler = _POOLERS[pooler_type]()
+
+        d_model = getattr(self.config, arch_dict[self.config.model_type]["config_names"]["width"])
+        if (d_model == output_dim) and (proj is None): # do we always need a proj?
+            self.proj = nn.Identity()
+        elif proj == 'linear':
+            self.proj = nn.Linear(d_model, output_dim, bias=False)
+        elif proj == 'mlp':
+            hidden_size = (d_model + output_dim) // 2
+            self.proj = nn.Sequential(
+                nn.Linear(d_model, hidden_size, bias=False),
+                nn.GELU(),
+                nn.Linear(hidden_size, output_dim, bias=False),
+            )
+
+        # self.itm_proj = nn.Linear(d_model, 2, bias=False)
+        # self.mlm_proj = nn.Linear(d_model, self.config.vocab_size), bias=False)
+        self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
+
+    # def forward_itm(self, x:TensorType, image_embeds:TensorType) -> TensorType:
+    #     image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(x.device)  
+    #     attn_mask = (x != self.config.pad_token_id).long()
+    #     out = self.transformer(
+    #         input_ids=x, 
+    #         attention_mask=attn_mask,
+    #         encoder_hidden_states = image_embeds,
+    #         encoder_attention_mask = image_atts,
+    #         )
+    #     pooled_out = self.pooler(out, attn_mask)
+
+    #     return self.itm_proj(pooled_out)
+
+    def mask(self, input_ids, vocab_size, device, targets=None, masked_indices=None, probability_matrix=None):
+        if masked_indices is None:                                       
+            masked_indices = torch.bernoulli(probability_matrix).bool()
+                                               
+        masked_indices[input_ids == self.tokenizer.pad_token_id] = False
+        masked_indices[input_ids == self.tokenizer.cls_token_id] = False
+        
+        if targets is not None:
+            targets[~masked_indices] = -100 # We only compute loss on masked tokens            
+
+        # 80% of the time, we replace masked input tokens with tokenizer.mask_token ([MASK])
+        indices_replaced = torch.bernoulli(torch.full(input_ids.shape, 0.8)).bool() & masked_indices
+        input_ids[indices_replaced] = self.tokenizer.mask_token_id
+
+        # 10% of the time, we replace masked input tokens with random word
+        indices_random = torch.bernoulli(torch.full(input_ids.shape, 0.5)).bool() & masked_indices & ~indices_replaced
+        random_words = torch.randint(vocab_size, input_ids.shape, dtype=torch.long).to(device)
+        input_ids[indices_random] = random_words[indices_random]                     
+        # The rest of the time (10% of the time) we keep the masked input tokens unchanged   
+        
+        if targets is not None:
+            return input_ids, targets
+        else:
+            return input_ids
+
+    def forward_mlm(self, input_ids, image_embeds, mlm_probability=0.25):
+        labels = input_ids.clone()
+        attn_mask = (input_ids != self.config.pad_token_id).long()
+        image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(input_ids.device) 
+        vocab_size = getattr(self.config, arch_dict[self.config.model_type]["config_names"]["vocab_size"])
+        probability_matrix = torch.full(labels.shape, mlm_probability)
+        input_ids, labels = self.mask(input_ids, vocab_size, input_ids.device, targets=labels,
+                                      probability_matrix = probability_matrix)
+        mlm_output = self.transformer(input_ids,
+                        attention_mask = attn_mask,
+                        encoder_hidden_states = image_embeds,
+                        encoder_attention_mask = image_atts,
+                        return_dict = True,
+                        labels = labels,
+                    )
+        return mlm_output.loss
+        # mlm_output = self.transformer(input_ids,
+        #                 attention_mask = attn_mask,
+        #                 encoder_hidden_states = image_embeds,
+        #                 encoder_attention_mask = image_atts,
+        #                 return_dict = True,
+        #             ).last_hidden_state
+        # logits = self.mlm_proj(mlm_output)
+
+        # # logits = logits[:, :-1, :].contiguous().view(-1, vocab_size)
+        # logits = logits[:, 1:, :].contiguous().view(-1, vocab_size)
+        # labels = labels[:, 1:].contiguous().view(-1)
+
+        # mlm_loss = F.cross_entropy(
+        #     logits,
+        #     labels,
+        #     # label_smoothing=0.1,
+        # )
+        # return mlm_loss
+
+
+    def forward(self, x:TensorType) -> TensorType:
+        attn_mask = (x != self.config.pad_token_id).long()
+        out = self.transformer(input_ids=x, attention_mask=attn_mask)
+        pooled_out = self.pooler(out, attn_mask)
+
+        return self.proj(pooled_out)
+
+    def lock(self, unlocked_layers:int=0, freeze_layer_norm:bool=True):
+        if not unlocked_layers: # full freezing
+             for n, p in self.transformer.named_parameters():
+                 p.requires_grad = (not freeze_layer_norm) if "LayerNorm" in n.split(".") else False
+             return
+
+        encoder = self.transformer.encoder if hasattr(self.transformer, 'encoder') else self.transformer
+        layer_list = getattr(encoder, arch_dict[self.config.model_type]["config_names"]["layer_attr"])
+        print(f"Unlocking {unlocked_layers}/{len(layer_list) + 1} layers of hf model")
+        embeddings = getattr(
+            self.transformer, arch_dict[self.config.model_type]["config_names"]["token_embeddings_attr"])
+        modules = [embeddings, *layer_list][:-unlocked_layers]
+        # freeze layers
+        for module in modules:
+            for n, p in module.named_parameters():
+                p.requires_grad = (not freeze_layer_norm) if "LayerNorm" in n.split(".") else False
+
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        self.transformer.gradient_checkpointing_enable()
+
+    def get_num_layers(self):
+        encoder = self.transformer.encoder if hasattr(self.transformer, 'encoder') else self.transformer
+        layer_list = getattr(encoder, arch_dict[self.config.model_type]["config_names"]["layer_attr"])
+        return len(layer_list)
+
+    def init_parameters(self):
+        pass

consis_id/models/eva_clip/loss.py

@@ -0,0 +1,138 @@
+import math
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+
+try:
+    import torch.distributed.nn
+    from torch import distributed as dist
+    has_distributed = True
+except ImportError:
+    has_distributed = False
+
+try:
+    import horovod.torch as hvd
+except ImportError:
+    hvd = None
+
+from timm.loss import LabelSmoothingCrossEntropy
+
+
+def gather_features(
+        image_features,
+        text_features,
+        local_loss=False,
+        gather_with_grad=False,
+        rank=0,
+        world_size=1,
+        use_horovod=False
+):
+    assert has_distributed, 'torch.distributed did not import correctly, please use a PyTorch version with support.'
+    if use_horovod:
+        assert hvd is not None, 'Please install horovod'
+        if gather_with_grad:
+            all_image_features = hvd.allgather(image_features)
+            all_text_features = hvd.allgather(text_features)
+        else:
+            with torch.no_grad():
+                all_image_features = hvd.allgather(image_features)
+                all_text_features = hvd.allgather(text_features)
+            if not local_loss:
+                # ensure grads for local rank when all_* features don't have a gradient
+                gathered_image_features = list(all_image_features.chunk(world_size, dim=0))
+                gathered_text_features = list(all_text_features.chunk(world_size, dim=0))
+                gathered_image_features[rank] = image_features
+                gathered_text_features[rank] = text_features
+                all_image_features = torch.cat(gathered_image_features, dim=0)
+                all_text_features = torch.cat(gathered_text_features, dim=0)
+    else:
+        # We gather tensors from all gpus
+        if gather_with_grad:
+            all_image_features = torch.cat(torch.distributed.nn.all_gather(image_features), dim=0)
+            all_text_features = torch.cat(torch.distributed.nn.all_gather(text_features), dim=0)
+            # all_image_features = torch.cat(torch.distributed.nn.all_gather(image_features, async_op=True), dim=0)
+            # all_text_features = torch.cat(torch.distributed.nn.all_gather(text_features, async_op=True), dim=0)
+        else:
+            gathered_image_features = [torch.zeros_like(image_features) for _ in range(world_size)]
+            gathered_text_features = [torch.zeros_like(text_features) for _ in range(world_size)]
+            dist.all_gather(gathered_image_features, image_features)
+            dist.all_gather(gathered_text_features, text_features)
+            if not local_loss:
+                # ensure grads for local rank when all_* features don't have a gradient
+                gathered_image_features[rank] = image_features
+                gathered_text_features[rank] = text_features
+            all_image_features = torch.cat(gathered_image_features, dim=0)
+            all_text_features = torch.cat(gathered_text_features, dim=0)
+
+    return all_image_features, all_text_features
+
+
+class ClipLoss(nn.Module):
+
+    def __init__(
+            self,
+            local_loss=False,
+            gather_with_grad=False,
+            cache_labels=False,
+            rank=0,
+            world_size=1,
+            use_horovod=False,
+            smoothing=0.,
+    ):
+        super().__init__()
+        self.local_loss = local_loss
+        self.gather_with_grad = gather_with_grad
+        self.cache_labels = cache_labels
+        self.rank = rank
+        self.world_size = world_size
+        self.use_horovod = use_horovod
+        self.label_smoothing_cross_entropy = LabelSmoothingCrossEntropy(smoothing=smoothing) if smoothing > 0 else None
+
+        # cache state
+        self.prev_num_logits = 0
+        self.labels = {}
+
+    def forward(self, image_features, text_features, logit_scale=1.):
+        device = image_features.device
+        if self.world_size > 1:
+            all_image_features, all_text_features = gather_features(
+                image_features, text_features,
+                self.local_loss, self.gather_with_grad, self.rank, self.world_size, self.use_horovod)
+
+            if self.local_loss:
+                logits_per_image = logit_scale * image_features @ all_text_features.T
+                logits_per_text = logit_scale * text_features @ all_image_features.T
+            else:
+                logits_per_image = logit_scale * all_image_features @ all_text_features.T
+                logits_per_text = logits_per_image.T
+        else:
+            logits_per_image = logit_scale * image_features @ text_features.T
+            logits_per_text = logit_scale * text_features @ image_features.T
+        # calculated ground-truth and cache if enabled
+        num_logits = logits_per_image.shape[0]
+        if self.prev_num_logits != num_logits or device not in self.labels:
+            labels = torch.arange(num_logits, device=device, dtype=torch.long)
+            if self.world_size > 1 and self.local_loss:
+                labels = labels + num_logits * self.rank
+            if self.cache_labels:
+                self.labels[device] = labels
+                self.prev_num_logits = num_logits
+        else:
+            labels = self.labels[device]
+        
+        if self.label_smoothing_cross_entropy:
+            total_loss = (
+                self.label_smoothing_cross_entropy(logits_per_image, labels) +
+                self.label_smoothing_cross_entropy(logits_per_text, labels)
+                ) / 2
+        else:
+            total_loss = (
+                F.cross_entropy(logits_per_image, labels) +
+                F.cross_entropy(logits_per_text, labels)
+                ) / 2
+            
+        acc = None
+        i2t_acc = (logits_per_image.argmax(-1) == labels).sum() / len(logits_per_image)
+        t2i_acc = (logits_per_text.argmax(-1) == labels).sum() / len(logits_per_text)
+        acc = {"i2t": i2t_acc, "t2i": t2i_acc}
+        return total_loss, acc

consis_id/models/eva_clip/model.py

@@ -0,0 +1,439 @@
+""" CLIP Model
+
+Adapted from https://github.com/openai/CLIP. Originally MIT License, Copyright (c) 2021 OpenAI.
+"""
+import os
+from dataclasses import dataclass
+from typing import Optional, Tuple, Union
+from functools import partial
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+from torch import nn
+
+try:
+    from .hf_model import HFTextEncoder
+except:
+    HFTextEncoder = None
+from .modified_resnet import ModifiedResNet
+from .timm_model import TimmModel
+from .eva_vit_model import EVAVisionTransformer
+from .transformer import LayerNorm, QuickGELU, Attention, VisionTransformer, TextTransformer
+
+try:
+    from apex.normalization import FusedLayerNorm
+except:
+    FusedLayerNorm = LayerNorm
+    print("Please 'pip install apex'")
+
+try:
+    import xformers.ops as xops
+except ImportError:
+    xops = None
+    print("Please 'pip install xformers'")
+
+@dataclass
+class CLIPVisionCfg:
+    layers: Union[Tuple[int, int, int, int], int] = 12
+    width: int = 768
+    head_width: int = 64
+    mlp_ratio: float = 4.0
+    patch_size: int = 16
+    image_size: Union[Tuple[int, int], int] = 224
+    ls_init_value: Optional[float] = None  # layer scale initial value
+    patch_dropout: float = 0. # what fraction of patches to dropout during training (0 would mean disabled and no patches dropped) - 0.5 to 0.75 recommended in the paper for optimal results
+    global_average_pool: bool = False # whether to global average pool the last embedding layer, instead of using CLS token (https://arxiv.org/abs/2205.01580)
+    drop_path_rate: Optional[float] = None  # drop path rate
+    timm_model_name: str = None  # a valid model name overrides layers, width, patch_size
+    timm_model_pretrained: bool = False  # use (imagenet) pretrained weights for named model
+    timm_pool: str = 'avg'  # feature pooling for timm model ('abs_attn', 'rot_attn', 'avg', '')
+    timm_proj: str = 'linear'  # linear projection for timm model output ('linear', 'mlp', '')
+    timm_proj_bias: bool = False  # enable bias final projection
+    eva_model_name: str = None # a valid eva model name overrides layers, width, patch_size
+    qkv_bias: bool = True
+    fusedLN: bool = False
+    xattn: bool = False
+    postnorm: bool = False
+    rope: bool = False
+    pt_hw_seq_len: int = 16   # 224/14
+    intp_freq: bool = False
+    naiveswiglu: bool = False
+    subln: bool = False
+
+
+@dataclass
+class CLIPTextCfg:
+    context_length: int = 77
+    vocab_size: int = 49408
+    width: int = 512
+    heads: int = 8
+    layers: int = 12
+    ls_init_value: Optional[float] = None  # layer scale initial value
+    hf_model_name: str = None
+    hf_tokenizer_name: str = None
+    hf_model_pretrained: bool = True
+    proj: str = 'mlp'
+    pooler_type: str = 'mean_pooler'
+    masked_language_modeling: bool = False
+    fusedLN: bool = False
+    xattn: bool = False
+    attn_mask: bool = True
+
+def get_cast_dtype(precision: str):
+    cast_dtype = None
+    if precision == 'bf16':
+        cast_dtype = torch.bfloat16
+    elif precision == 'fp16':
+        cast_dtype = torch.float16
+    return cast_dtype
+
+
+def _build_vision_tower(
+        embed_dim: int,
+        vision_cfg: CLIPVisionCfg,
+        quick_gelu: bool = False,
+        cast_dtype: Optional[torch.dtype] = None
+):
+    if isinstance(vision_cfg, dict):
+        vision_cfg = CLIPVisionCfg(**vision_cfg)
+
+    # OpenAI models are pretrained w/ QuickGELU but native nn.GELU is both faster and more
+    # memory efficient in recent PyTorch releases (>= 1.10).
+    # NOTE: timm models always use native GELU regardless of quick_gelu flag.
+    act_layer = QuickGELU if quick_gelu else nn.GELU
+
+    if vision_cfg.eva_model_name:
+        vision_heads = vision_cfg.width // vision_cfg.head_width
+        norm_layer = LayerNorm
+        
+        visual = EVAVisionTransformer(
+            img_size=vision_cfg.image_size,
+            patch_size=vision_cfg.patch_size,
+            num_classes=embed_dim,
+            use_mean_pooling=vision_cfg.global_average_pool, #False
+            init_values=vision_cfg.ls_init_value,
+            patch_dropout=vision_cfg.patch_dropout,
+            embed_dim=vision_cfg.width,
+            depth=vision_cfg.layers,
+            num_heads=vision_heads,
+            mlp_ratio=vision_cfg.mlp_ratio,
+            qkv_bias=vision_cfg.qkv_bias,
+            drop_path_rate=vision_cfg.drop_path_rate,
+            norm_layer= partial(FusedLayerNorm, eps=1e-6) if vision_cfg.fusedLN else partial(norm_layer, eps=1e-6),
+            xattn=vision_cfg.xattn,
+            rope=vision_cfg.rope,
+            postnorm=vision_cfg.postnorm,
+            pt_hw_seq_len= vision_cfg.pt_hw_seq_len,   # 224/14
+            intp_freq= vision_cfg.intp_freq,
+            naiveswiglu= vision_cfg.naiveswiglu,
+            subln= vision_cfg.subln
+        )
+    elif vision_cfg.timm_model_name:
+        visual = TimmModel(
+            vision_cfg.timm_model_name,
+            pretrained=vision_cfg.timm_model_pretrained,
+            pool=vision_cfg.timm_pool,
+            proj=vision_cfg.timm_proj,
+            proj_bias=vision_cfg.timm_proj_bias,
+            embed_dim=embed_dim,
+            image_size=vision_cfg.image_size
+        )
+        act_layer = nn.GELU  # so that text transformer doesn't use QuickGELU w/ timm models
+    elif isinstance(vision_cfg.layers, (tuple, list)):
+        vision_heads = vision_cfg.width * 32 // vision_cfg.head_width
+        visual = ModifiedResNet(
+            layers=vision_cfg.layers,
+            output_dim=embed_dim,
+            heads=vision_heads,
+            image_size=vision_cfg.image_size,
+            width=vision_cfg.width
+        )
+    else:
+        vision_heads = vision_cfg.width // vision_cfg.head_width
+        norm_layer = LayerNormFp32 if cast_dtype in (torch.float16, torch.bfloat16) else LayerNorm
+        visual = VisionTransformer(
+            image_size=vision_cfg.image_size,
+            patch_size=vision_cfg.patch_size,
+            width=vision_cfg.width,
+            layers=vision_cfg.layers,
+            heads=vision_heads,
+            mlp_ratio=vision_cfg.mlp_ratio,
+            ls_init_value=vision_cfg.ls_init_value,
+            patch_dropout=vision_cfg.patch_dropout,
+            global_average_pool=vision_cfg.global_average_pool,
+            output_dim=embed_dim,
+            act_layer=act_layer,
+            norm_layer=norm_layer,
+        )
+
+    return visual
+
+
+def _build_text_tower(
+        embed_dim: int,
+        text_cfg: CLIPTextCfg,
+        quick_gelu: bool = False,
+        cast_dtype: Optional[torch.dtype] = None,
+):
+    if isinstance(text_cfg, dict):
+        text_cfg = CLIPTextCfg(**text_cfg)
+
+    if text_cfg.hf_model_name:
+        text = HFTextEncoder(
+            text_cfg.hf_model_name,
+            output_dim=embed_dim,
+            tokenizer_name=text_cfg.hf_tokenizer_name,
+            proj=text_cfg.proj,
+            pooler_type=text_cfg.pooler_type,
+            masked_language_modeling=text_cfg.masked_language_modeling
+       )
+    else:
+        act_layer = QuickGELU if quick_gelu else nn.GELU
+        norm_layer = LayerNorm
+
+        text = TextTransformer(
+            context_length=text_cfg.context_length,
+            vocab_size=text_cfg.vocab_size,
+            width=text_cfg.width,
+            heads=text_cfg.heads,
+            layers=text_cfg.layers,
+            ls_init_value=text_cfg.ls_init_value,
+            output_dim=embed_dim,
+            act_layer=act_layer,
+            norm_layer= FusedLayerNorm if text_cfg.fusedLN else norm_layer,
+            xattn=text_cfg.xattn,
+            attn_mask=text_cfg.attn_mask,
+        )
+    return text
+
+class CLIP(nn.Module):
+    def __init__(
+            self,
+            embed_dim: int,
+            vision_cfg: CLIPVisionCfg,
+            text_cfg: CLIPTextCfg,
+            quick_gelu: bool = False,
+            cast_dtype: Optional[torch.dtype] = None,
+    ):
+        super().__init__()
+        self.visual = _build_vision_tower(embed_dim, vision_cfg, quick_gelu, cast_dtype)
+
+        text = _build_text_tower(embed_dim, text_cfg, quick_gelu, cast_dtype)
+        self.transformer = text.transformer
+        self.vocab_size = text.vocab_size
+        self.token_embedding = text.token_embedding
+        self.positional_embedding = text.positional_embedding
+        self.ln_final = text.ln_final
+        self.text_projection = text.text_projection
+        self.register_buffer('attn_mask', text.attn_mask, persistent=False)
+
+        self.logit_scale = nn.Parameter(torch.ones([]) * np.log(1 / 0.07))
+
+    def lock_image_tower(self, unlocked_groups=0, freeze_bn_stats=False):
+        # lock image tower as per LiT - https://arxiv.org/abs/2111.07991
+        self.visual.lock(unlocked_groups=unlocked_groups, freeze_bn_stats=freeze_bn_stats)
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        self.visual.set_grad_checkpointing(enable)
+        self.transformer.grad_checkpointing = enable
+    
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'logit_scale'}
+
+    def encode_image(self, image, normalize: bool = False):
+        features = self.visual(image)
+        return F.normalize(features, dim=-1) if normalize else features
+
+    def encode_text(self, text, normalize: bool = False):
+        cast_dtype = self.transformer.get_cast_dtype()
+
+        x = self.token_embedding(text).to(cast_dtype)  # [batch_size, n_ctx, d_model]
+
+        x = x + self.positional_embedding.to(cast_dtype)
+        x = x.permute(1, 0, 2)  # NLD -> LND
+        x = self.transformer(x, attn_mask=self.attn_mask)
+        x = x.permute(1, 0, 2)  # LND -> NLD
+        x = self.ln_final(x)  # [batch_size, n_ctx, transformer.width]
+        # take features from the eot embedding (eot_token is the highest number in each sequence)
+        x = x[torch.arange(x.shape[0]), text.argmax(dim=-1)] @ self.text_projection
+        return F.normalize(x, dim=-1) if normalize else x
+
+    def forward(self, image, text):
+        image_features = self.encode_image(image, normalize=True)
+        text_features = self.encode_text(text, normalize=True)
+        return image_features, text_features, self.logit_scale.exp()
+
+
+class CustomCLIP(nn.Module):
+    def __init__(
+            self,
+            embed_dim: int,
+            vision_cfg: CLIPVisionCfg,
+            text_cfg: CLIPTextCfg,
+            quick_gelu: bool = False,
+            cast_dtype: Optional[torch.dtype] = None,
+            itm_task: bool = False,
+    ):
+        super().__init__()
+        self.visual = _build_vision_tower(embed_dim, vision_cfg, quick_gelu, cast_dtype)
+        self.text = _build_text_tower(embed_dim, text_cfg, quick_gelu, cast_dtype)
+        self.logit_scale = nn.Parameter(torch.ones([]) * np.log(1 / 0.07))
+
+    def lock_image_tower(self, unlocked_groups=0, freeze_bn_stats=False):
+        # lock image tower as per LiT - https://arxiv.org/abs/2111.07991
+        self.visual.lock(unlocked_groups=unlocked_groups, freeze_bn_stats=freeze_bn_stats)
+
+    def lock_text_tower(self, unlocked_layers:int=0, freeze_layer_norm:bool=True):
+        self.text.lock(unlocked_layers, freeze_layer_norm)
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        self.visual.set_grad_checkpointing(enable)
+        self.text.set_grad_checkpointing(enable)
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'logit_scale'}
+
+    def encode_image(self, image, normalize: bool = False):
+        features = self.visual(image)
+        return F.normalize(features, dim=-1) if normalize else features
+
+    def encode_text(self, text, normalize: bool = False):
+        features = self.text(text)
+        return F.normalize(features, dim=-1) if normalize else features
+
+    def forward(self, image, text):
+        image_features = self.encode_image(image, normalize=True)
+        text_features = self.encode_text(text, normalize=True)
+        return image_features, text_features, self.logit_scale.exp()
+
+
+def convert_weights_to_lp(model: nn.Module, dtype=torch.float16):
+    """Convert applicable model parameters to low-precision (bf16 or fp16)"""
+
+    def _convert_weights(l):
+        
+        if isinstance(l, (nn.Conv1d, nn.Conv2d, nn.Linear)):
+            l.weight.data = l.weight.data.to(dtype)
+            if l.bias is not None:
+                l.bias.data = l.bias.data.to(dtype)
+
+        if isinstance(l, (nn.MultiheadAttention, Attention)):
+            for attr in [*[f"{s}_proj_weight" for s in ["in", "q", "k", "v"]], "in_proj_bias", "bias_k", "bias_v"]:
+                tensor = getattr(l, attr, None)
+                if tensor is not None:
+                    tensor.data = tensor.data.to(dtype)
+
+        if isinstance(l, nn.Parameter):
+            l.data = l.data.to(dtype)
+
+        for name in ["text_projection", "proj"]:
+            if hasattr(l, name) and isinstance(l, nn.Parameter):
+                attr = getattr(l, name, None)
+                if attr is not None:
+                    attr.data = attr.data.to(dtype)
+
+    model.apply(_convert_weights)
+
+
+convert_weights_to_fp16 = convert_weights_to_lp  # backwards compat
+
+
+# used to maintain checkpoint compatibility
+def convert_to_custom_text_state_dict(state_dict: dict):
+    if 'text_projection' in state_dict:
+        # old format state_dict, move text tower -> .text
+        new_state_dict = {}
+        for k, v in state_dict.items():
+            if any(k.startswith(p) for p in (
+                'text_projection',
+                'positional_embedding',
+                'token_embedding',
+                'transformer',
+                'ln_final',
+                'logit_scale'
+            )):
+                k = 'text.' + k
+            new_state_dict[k] = v
+        return new_state_dict
+    return state_dict
+
+
+def build_model_from_openai_state_dict(
+        state_dict: dict,
+        quick_gelu=True,
+        cast_dtype=torch.float16,
+):
+    vit = "visual.proj" in state_dict
+
+    if vit:
+        vision_width = state_dict["visual.conv1.weight"].shape[0]
+        vision_layers = len(
+            [k for k in state_dict.keys() if k.startswith("visual.") and k.endswith(".attn.in_proj_weight")])
+        vision_patch_size = state_dict["visual.conv1.weight"].shape[-1]
+        grid_size = round((state_dict["visual.positional_embedding"].shape[0] - 1) ** 0.5)
+        image_size = vision_patch_size * grid_size
+    else:
+        counts: list = [
+            len(set(k.split(".")[2] for k in state_dict if k.startswith(f"visual.layer{b}"))) for b in [1, 2, 3, 4]]
+        vision_layers = tuple(counts)
+        vision_width = state_dict["visual.layer1.0.conv1.weight"].shape[0]
+        output_width = round((state_dict["visual.attnpool.positional_embedding"].shape[0] - 1) ** 0.5)
+        vision_patch_size = None
+        assert output_width ** 2 + 1 == state_dict["visual.attnpool.positional_embedding"].shape[0]
+        image_size = output_width * 32
+
+    embed_dim = state_dict["text_projection"].shape[1]
+    context_length = state_dict["positional_embedding"].shape[0]
+    vocab_size = state_dict["token_embedding.weight"].shape[0]
+    transformer_width = state_dict["ln_final.weight"].shape[0]
+    transformer_heads = transformer_width // 64
+    transformer_layers = len(set(k.split(".")[2] for k in state_dict if k.startswith(f"transformer.resblocks")))
+
+    vision_cfg = CLIPVisionCfg(
+        layers=vision_layers,
+        width=vision_width,
+        patch_size=vision_patch_size,
+        image_size=image_size,
+    )
+    text_cfg = CLIPTextCfg(
+        context_length=context_length,
+        vocab_size=vocab_size,
+        width=transformer_width,
+        heads=transformer_heads,
+        layers=transformer_layers
+    )
+    model = CLIP(
+        embed_dim,
+        vision_cfg=vision_cfg,
+        text_cfg=text_cfg,
+        quick_gelu=quick_gelu,  # OpenAI models were trained with QuickGELU
+        cast_dtype=cast_dtype,
+    )
+
+    for key in ["input_resolution", "context_length", "vocab_size"]:
+        state_dict.pop(key, None)
+
+    convert_weights_to_fp16(model)  # OpenAI state dicts are partially converted to float16
+    model.load_state_dict(state_dict)
+    return model.eval()
+
+
+def trace_model(model, batch_size=256, device=torch.device('cpu')):
+    model.eval()
+    image_size = model.visual.image_size
+    example_images = torch.ones((batch_size, 3, image_size, image_size), device=device)
+    example_text = torch.zeros((batch_size, model.context_length), dtype=torch.int, device=device)
+    model = torch.jit.trace_module(
+        model,
+        inputs=dict(
+            forward=(example_images, example_text),
+            encode_text=(example_text,),
+            encode_image=(example_images,)
+        ))
+    model.visual.image_size = image_size
+    return model

consis_id/models/eva_clip/model_configs/EVA01-CLIP-B-16.json

@@ -0,0 +1,19 @@
+{
+    "embed_dim": 512,
+    "vision_cfg": {
+        "image_size": 224,
+        "layers": 12,
+        "width": 768,
+        "patch_size": 16,
+        "eva_model_name": "eva-clip-b-16",
+        "ls_init_value": 0.1,
+        "drop_path_rate": 0.0
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 512,
+        "heads": 8,
+        "layers": 12
+    }
+}

consis_id/models/eva_clip/model_configs/EVA01-CLIP-g-14-plus.json

@@ -0,0 +1,24 @@
+{
+    "embed_dim": 1024,
+    "vision_cfg": {
+        "image_size": 224,
+        "layers": 40,
+        "width": 1408,
+        "head_width": 88,
+        "mlp_ratio": 4.3637,
+        "patch_size": 14,
+        "eva_model_name": "eva-clip-g-14-x",
+        "drop_path_rate": 0,
+        "xattn": true,
+        "fusedLN": true
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 1024,
+        "heads": 16,
+        "layers": 24,
+        "xattn": false,
+        "fusedLN": true
+    }
+}

consis_id/models/eva_clip/model_configs/EVA01-CLIP-g-14.json

@@ -0,0 +1,24 @@
+{
+    "embed_dim": 1024,
+    "vision_cfg": {
+        "image_size": 224,
+        "layers": 40,
+        "width": 1408,
+        "head_width": 88,
+        "mlp_ratio": 4.3637,
+        "patch_size": 14,
+        "eva_model_name": "eva-clip-g-14-x",
+        "drop_path_rate": 0.4,
+        "xattn": true,
+        "fusedLN": true
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 768,
+        "heads": 12,
+        "layers": 12,
+        "xattn": false,
+        "fusedLN": true
+    }
+}

consis_id/models/eva_clip/model_configs/EVA02-CLIP-B-16.json

@@ -0,0 +1,29 @@
+{
+    "embed_dim": 512,
+    "vision_cfg": {
+        "image_size": 224,
+        "layers": 12,
+        "width": 768,
+        "head_width": 64,
+        "patch_size": 16,
+        "mlp_ratio": 2.6667,
+        "eva_model_name": "eva-clip-b-16-X",
+        "drop_path_rate": 0.0,
+        "xattn": true,
+        "fusedLN": true,
+        "rope": true,
+        "pt_hw_seq_len": 16,
+        "intp_freq": true,
+        "naiveswiglu": true,
+        "subln": true
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 512,
+        "heads": 8,
+        "layers": 12,
+        "xattn": true,
+        "fusedLN": true
+    }
+}

consis_id/models/eva_clip/model_configs/EVA02-CLIP-L-14-336.json

@@ -0,0 +1,29 @@
+{
+    "embed_dim": 768,
+    "vision_cfg": {
+        "image_size": 336,
+        "layers": 24,
+        "width": 1024,
+        "drop_path_rate": 0,
+        "head_width": 64,
+        "mlp_ratio": 2.6667,
+        "patch_size": 14,
+        "eva_model_name": "eva-clip-l-14-336",
+        "xattn": true,
+        "fusedLN": true,
+        "rope": true,
+        "pt_hw_seq_len": 16,
+        "intp_freq": true,
+        "naiveswiglu": true,
+        "subln": true
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 768,
+        "heads": 12,
+        "layers": 12,
+        "xattn": false,
+        "fusedLN": true
+    }
+}

consis_id/models/eva_clip/model_configs/EVA02-CLIP-L-14.json

@@ -0,0 +1,29 @@
+{
+    "embed_dim": 768,
+    "vision_cfg": {
+        "image_size": 224,
+        "layers": 24,
+        "width": 1024,
+        "drop_path_rate": 0,
+        "head_width": 64,
+        "mlp_ratio": 2.6667,
+        "patch_size": 14,
+        "eva_model_name": "eva-clip-l-14",
+        "xattn": true,
+        "fusedLN": true,
+        "rope": true,
+        "pt_hw_seq_len": 16,
+        "intp_freq": true,
+        "naiveswiglu": true,
+        "subln": true
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 768,
+        "heads": 12,
+        "layers": 12,
+        "xattn": false,
+        "fusedLN": true
+    }
+}

consis_id/models/eva_clip/model_configs/EVA02-CLIP-bigE-14-plus.json

@@ -0,0 +1,25 @@
+{
+    "embed_dim": 1024,
+    "vision_cfg": {
+        "image_size": 224,
+        "layers": 64,
+        "width": 1792,
+        "head_width": 112,
+        "mlp_ratio": 8.571428571428571,
+        "patch_size": 14,
+        "eva_model_name": "eva-clip-4b-14-x",
+        "drop_path_rate": 0,
+        "xattn": true,
+        "postnorm": true,
+        "fusedLN": true
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 1280,
+        "heads": 20,
+        "layers": 32,
+        "xattn": false,
+        "fusedLN": true
+    }
+}

consis_id/models/eva_clip/model_configs/EVA02-CLIP-bigE-14.json

@@ -0,0 +1,25 @@
+{
+    "embed_dim": 1024,
+    "vision_cfg": {
+        "image_size": 224,
+        "layers": 64,
+        "width": 1792,
+        "head_width": 112,
+        "mlp_ratio": 8.571428571428571,
+        "patch_size": 14,
+        "eva_model_name": "eva-clip-4b-14-x",
+        "drop_path_rate": 0,
+        "xattn": true,
+        "postnorm": true,
+        "fusedLN": true
+    },
+    "text_cfg": {
+        "context_length": 77,
+        "vocab_size": 49408,
+        "width": 1024,
+        "heads": 16,
+        "layers": 24,
+        "xattn": false,
+        "fusedLN": true
+    }
+}

consis_id/models/eva_clip/modified_resnet.py

@@ -0,0 +1,188 @@
+import os
+import sys
+
+import torch
+from torch import nn
+from torch.nn import functional as F
+from collections import OrderedDict
+
+current_file_path = os.path.abspath(__file__)
+project_roots = [os.path.dirname(current_file_path)]
+for project_root in project_roots:
+    sys.path.insert(0, project_root) if project_root not in sys.path else None
+
+from .utils import freeze_batch_norm_2d
+
+
+class Bottleneck(nn.Module):
+    expansion = 4
+
+    def __init__(self, inplanes, planes, stride=1):
+        super().__init__()
+
+        # all conv layers have stride 1. an avgpool is performed after the second convolution when stride > 1
+        self.conv1 = nn.Conv2d(inplanes, planes, 1, bias=False)
+        self.bn1 = nn.BatchNorm2d(planes)
+        self.act1 = nn.ReLU(inplace=True)
+
+        self.conv2 = nn.Conv2d(planes, planes, 3, padding=1, bias=False)
+        self.bn2 = nn.BatchNorm2d(planes)
+        self.act2 = nn.ReLU(inplace=True)
+
+        self.avgpool = nn.AvgPool2d(stride) if stride > 1 else nn.Identity()
+
+        self.conv3 = nn.Conv2d(planes, planes * self.expansion, 1, bias=False)
+        self.bn3 = nn.BatchNorm2d(planes * self.expansion)
+        self.act3 = nn.ReLU(inplace=True)
+
+        self.downsample = None
+        self.stride = stride
+
+        if stride > 1 or inplanes != planes * Bottleneck.expansion:
+            # downsampling layer is prepended with an avgpool, and the subsequent convolution has stride 1
+            self.downsample = nn.Sequential(OrderedDict([
+                ("-1", nn.AvgPool2d(stride)),
+                ("0", nn.Conv2d(inplanes, planes * self.expansion, 1, stride=1, bias=False)),
+                ("1", nn.BatchNorm2d(planes * self.expansion))
+            ]))
+
+    def forward(self, x: torch.Tensor):
+        identity = x
+
+        out = self.act1(self.bn1(self.conv1(x)))
+        out = self.act2(self.bn2(self.conv2(out)))
+        out = self.avgpool(out)
+        out = self.bn3(self.conv3(out))
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+        out += identity
+        out = self.act3(out)
+        return out
+
+
+class AttentionPool2d(nn.Module):
+    def __init__(self, spacial_dim: int, embed_dim: int, num_heads: int, output_dim: int = None):
+        super().__init__()
+        self.positional_embedding = nn.Parameter(torch.randn(spacial_dim ** 2 + 1, embed_dim) / embed_dim ** 0.5)
+        self.k_proj = nn.Linear(embed_dim, embed_dim)
+        self.q_proj = nn.Linear(embed_dim, embed_dim)
+        self.v_proj = nn.Linear(embed_dim, embed_dim)
+        self.c_proj = nn.Linear(embed_dim, output_dim or embed_dim)
+        self.num_heads = num_heads
+
+    def forward(self, x):
+        x = x.reshape(x.shape[0], x.shape[1], x.shape[2] * x.shape[3]).permute(2, 0, 1)  # NCHW -> (HW)NC
+        x = torch.cat([x.mean(dim=0, keepdim=True), x], dim=0)  # (HW+1)NC
+        x = x + self.positional_embedding[:, None, :].to(x.dtype)  # (HW+1)NC
+        x, _ = F.multi_head_attention_forward(
+            query=x, key=x, value=x,
+            embed_dim_to_check=x.shape[-1],
+            num_heads=self.num_heads,
+            q_proj_weight=self.q_proj.weight,
+            k_proj_weight=self.k_proj.weight,
+            v_proj_weight=self.v_proj.weight,
+            in_proj_weight=None,
+            in_proj_bias=torch.cat([self.q_proj.bias, self.k_proj.bias, self.v_proj.bias]),
+            bias_k=None,
+            bias_v=None,
+            add_zero_attn=False,
+            dropout_p=0.,
+            out_proj_weight=self.c_proj.weight,
+            out_proj_bias=self.c_proj.bias,
+            use_separate_proj_weight=True,
+            training=self.training,
+            need_weights=False
+        )
+
+        return x[0]
+
+
+class ModifiedResNet(nn.Module):
+    """
+    A ResNet class that is similar to torchvision's but contains the following changes:
+    - There are now 3 "stem" convolutions as opposed to 1, with an average pool instead of a max pool.
+    - Performs anti-aliasing strided convolutions, where an avgpool is prepended to convolutions with stride > 1
+    - The final pooling layer is a QKV attention instead of an average pool
+    """
+
+    def __init__(self, layers, output_dim, heads, image_size=224, width=64):
+        super().__init__()
+        self.output_dim = output_dim
+        self.image_size = image_size
+
+        # the 3-layer stem
+        self.conv1 = nn.Conv2d(3, width // 2, kernel_size=3, stride=2, padding=1, bias=False)
+        self.bn1 = nn.BatchNorm2d(width // 2)
+        self.act1 = nn.ReLU(inplace=True)
+        self.conv2 = nn.Conv2d(width // 2, width // 2, kernel_size=3, padding=1, bias=False)
+        self.bn2 = nn.BatchNorm2d(width // 2)
+        self.act2 = nn.ReLU(inplace=True)
+        self.conv3 = nn.Conv2d(width // 2, width, kernel_size=3, padding=1, bias=False)
+        self.bn3 = nn.BatchNorm2d(width)
+        self.act3 = nn.ReLU(inplace=True)
+        self.avgpool = nn.AvgPool2d(2)
+
+        # residual layers
+        self._inplanes = width  # this is a *mutable* variable used during construction
+        self.layer1 = self._make_layer(width, layers[0])
+        self.layer2 = self._make_layer(width * 2, layers[1], stride=2)
+        self.layer3 = self._make_layer(width * 4, layers[2], stride=2)
+        self.layer4 = self._make_layer(width * 8, layers[3], stride=2)
+
+        embed_dim = width * 32  # the ResNet feature dimension
+        self.attnpool = AttentionPool2d(image_size // 32, embed_dim, heads, output_dim)
+
+        self.init_parameters()
+
+    def _make_layer(self, planes, blocks, stride=1):
+        layers = [Bottleneck(self._inplanes, planes, stride)]
+
+        self._inplanes = planes * Bottleneck.expansion
+        for _ in range(1, blocks):
+            layers.append(Bottleneck(self._inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def init_parameters(self):
+        if self.attnpool is not None:
+            std = self.attnpool.c_proj.in_features ** -0.5
+            nn.init.normal_(self.attnpool.q_proj.weight, std=std)
+            nn.init.normal_(self.attnpool.k_proj.weight, std=std)
+            nn.init.normal_(self.attnpool.v_proj.weight, std=std)
+            nn.init.normal_(self.attnpool.c_proj.weight, std=std)
+
+        for resnet_block in [self.layer1, self.layer2, self.layer3, self.layer4]:
+            for name, param in resnet_block.named_parameters():
+                if name.endswith("bn3.weight"):
+                    nn.init.zeros_(param)
+
+    def lock(self, unlocked_groups=0, freeze_bn_stats=False):
+        assert unlocked_groups == 0, 'partial locking not currently supported for this model'
+        for param in self.parameters():
+            param.requires_grad = False
+        if freeze_bn_stats:
+            freeze_batch_norm_2d(self)
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        # FIXME support for non-transformer
+        pass
+
+    def stem(self, x):
+        x = self.act1(self.bn1(self.conv1(x)))
+        x = self.act2(self.bn2(self.conv2(x)))
+        x = self.act3(self.bn3(self.conv3(x)))
+        x = self.avgpool(x)
+        return x
+
+    def forward(self, x):
+        x = self.stem(x)
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        x = self.layer4(x)
+        x = self.attnpool(x)
+
+        return x

consis_id/models/eva_clip/openai.py

@@ -0,0 +1,144 @@
+""" OpenAI pretrained model functions
+
+Adapted from https://github.com/openai/CLIP. Originally MIT License, Copyright (c) 2021 OpenAI.
+"""
+
+import os
+import warnings
+from typing import List, Optional, Union
+
+import torch
+
+from .model import build_model_from_openai_state_dict, convert_weights_to_lp, get_cast_dtype
+from .pretrained import get_pretrained_url, list_pretrained_models_by_tag, download_pretrained_from_url
+
+__all__ = ["list_openai_models", "load_openai_model"]
+
+
+def list_openai_models() -> List[str]:
+    """Returns the names of available CLIP models"""
+    return list_pretrained_models_by_tag('openai')
+
+
+def load_openai_model(
+        name: str,
+        precision: Optional[str] = None,
+        device: Optional[Union[str, torch.device]] = None,
+        jit: bool = True,
+        cache_dir: Optional[str] = None,
+):
+    """Load a CLIP model
+
+    Parameters
+    ----------
+    name : str
+        A model name listed by `clip.available_models()`, or the path to a model checkpoint containing the state_dict
+    precision: str
+        Model precision, if None defaults to 'fp32' if device == 'cpu' else 'fp16'.
+    device : Union[str, torch.device]
+        The device to put the loaded model
+    jit : bool
+        Whether to load the optimized JIT model (default) or more hackable non-JIT model.
+    cache_dir : Optional[str]
+        The directory to cache the downloaded model weights
+
+    Returns
+    -------
+    model : torch.nn.Module
+        The CLIP model
+    preprocess : Callable[[PIL.Image], torch.Tensor]
+        A torchvision transform that converts a PIL image into a tensor that the returned model can take as its input
+    """
+    if device is None:
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+    if precision is None:
+        precision = 'fp32' if device == 'cpu' else 'fp16'
+
+    if get_pretrained_url(name, 'openai'):
+        model_path = download_pretrained_from_url(get_pretrained_url(name, 'openai'), cache_dir=cache_dir)
+    elif os.path.isfile(name):
+        model_path = name
+    else:
+        raise RuntimeError(f"Model {name} not found; available models = {list_openai_models()}")
+
+    try:
+        # loading JIT archive
+        model = torch.jit.load(model_path, map_location=device if jit else "cpu").eval()
+        state_dict = None
+    except RuntimeError:
+        # loading saved state dict
+        if jit:
+            warnings.warn(f"File {model_path} is not a JIT archive. Loading as a state dict instead")
+            jit = False
+        state_dict = torch.load(model_path, map_location="cpu")
+
+    if not jit:
+        # Build a non-jit model from the OpenAI jitted model state dict
+        cast_dtype = get_cast_dtype(precision)
+        try:
+            model = build_model_from_openai_state_dict(state_dict or model.state_dict(), cast_dtype=cast_dtype)
+        except KeyError:
+            sd = {k[7:]: v for k, v in state_dict["state_dict"].items()}
+            model = build_model_from_openai_state_dict(sd, cast_dtype=cast_dtype)
+
+        # model from OpenAI state dict is in manually cast fp16 mode, must be converted for AMP/fp32/bf16 use
+        model = model.to(device)
+        if precision.startswith('amp') or precision == 'fp32':
+            model.float()
+        elif precision == 'bf16':
+            convert_weights_to_lp(model, dtype=torch.bfloat16)
+
+        return model
+
+    # patch the device names
+    device_holder = torch.jit.trace(lambda: torch.ones([]).to(torch.device(device)), example_inputs=[])
+    device_node = [n for n in device_holder.graph.findAllNodes("prim::Constant") if "Device" in repr(n)][-1]
+
+    def patch_device(module):
+        try:
+            graphs = [module.graph] if hasattr(module, "graph") else []
+        except RuntimeError:
+            graphs = []
+
+        if hasattr(module, "forward1"):
+            graphs.append(module.forward1.graph)
+
+        for graph in graphs:
+            for node in graph.findAllNodes("prim::Constant"):
+                if "value" in node.attributeNames() and str(node["value"]).startswith("cuda"):
+                    node.copyAttributes(device_node)
+
+    model.apply(patch_device)
+    patch_device(model.encode_image)
+    patch_device(model.encode_text)
+
+    # patch dtype to float32 (typically for CPU)
+    if precision == 'fp32':
+        float_holder = torch.jit.trace(lambda: torch.ones([]).float(), example_inputs=[])
+        float_input = list(float_holder.graph.findNode("aten::to").inputs())[1]
+        float_node = float_input.node()
+
+        def patch_float(module):
+            try:
+                graphs = [module.graph] if hasattr(module, "graph") else []
+            except RuntimeError:
+                graphs = []
+
+            if hasattr(module, "forward1"):
+                graphs.append(module.forward1.graph)
+
+            for graph in graphs:
+                for node in graph.findAllNodes("aten::to"):
+                    inputs = list(node.inputs())
+                    for i in [1, 2]:  # dtype can be the second or third argument to aten::to()
+                        if inputs[i].node()["value"] == 5:
+                            inputs[i].node().copyAttributes(float_node)
+
+        model.apply(patch_float)
+        patch_float(model.encode_image)
+        patch_float(model.encode_text)
+        model.float()
+
+    # ensure image_size attr available at consistent location for both jit and non-jit
+    model.visual.image_size = model.input_resolution.item()
+    return model

consis_id/models/eva_clip/pretrained.py

@@ -0,0 +1,332 @@
+import hashlib
+import os
+import urllib
+import warnings
+from functools import partial
+from typing import Dict, Union
+
+from tqdm import tqdm
+
+try:
+    from huggingface_hub import hf_hub_download
+    _has_hf_hub = True
+except ImportError:
+    hf_hub_download = None
+    _has_hf_hub = False
+
+
+def _pcfg(url='', hf_hub='', filename='', mean=None, std=None):
+    return dict(
+        url=url,
+        hf_hub=hf_hub,
+        mean=mean,
+        std=std,
+    )
+
+_VITB32 = dict(
+    openai=_pcfg(
+        "https://openaipublic.azureedge.net/clip/models/40d365715913c9da98579312b702a82c18be219cc2a73407c4526f58eba950af/ViT-B-32.pt"),
+    laion400m_e31=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_32-quickgelu-laion400m_e31-d867053b.pt"),
+    laion400m_e32=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_32-quickgelu-laion400m_e32-46683a32.pt"),
+    laion2b_e16=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_32-laion2b_e16-af8dbd0c.pth"),
+    laion2b_s34b_b79k=_pcfg(hf_hub='laion/CLIP-ViT-B-32-laion2B-s34B-b79K/')
+)
+
+_VITB32_quickgelu = dict(
+    openai=_pcfg(
+        "https://openaipublic.azureedge.net/clip/models/40d365715913c9da98579312b702a82c18be219cc2a73407c4526f58eba950af/ViT-B-32.pt"),
+    laion400m_e31=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_32-quickgelu-laion400m_e31-d867053b.pt"),
+    laion400m_e32=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_32-quickgelu-laion400m_e32-46683a32.pt"),
+)
+
+_VITB16 = dict(
+    openai=_pcfg(
+        "https://openaipublic.azureedge.net/clip/models/5806e77cd80f8b59890b7e101eabd078d9fb84e6937f9e85e4ecb61988df416f/ViT-B-16.pt"),
+    laion400m_e31=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_16-laion400m_e31-00efa78f.pt"),
+    laion400m_e32=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_16-laion400m_e32-55e67d44.pt"),
+    laion2b_s34b_b88k=_pcfg(hf_hub='laion/CLIP-ViT-B-16-laion2B-s34B-b88K/'),
+)
+
+_EVAB16 = dict(
+    eva=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_B_psz14to16.pt'),
+    eva02=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_B_psz14to16.pt'),
+    eva_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_B_psz16_s8B.pt'),
+    eva02_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_B_psz16_s8B.pt'),
+)
+
+_VITB16_PLUS_240 = dict(
+    laion400m_e31=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_16_plus_240-laion400m_e31-8fb26589.pt"),
+    laion400m_e32=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_b_16_plus_240-laion400m_e32-699c4b84.pt"),
+)
+
+_VITL14 = dict(
+    openai=_pcfg(
+        "https://openaipublic.azureedge.net/clip/models/b8cca3fd41ae0c99ba7e8951adf17d267cdb84cd88be6f7c2e0eca1737a03836/ViT-L-14.pt"),
+    laion400m_e31=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_l_14-laion400m_e31-69988bb6.pt"),
+    laion400m_e32=_pcfg(
+        "https://github.com/mlfoundations/open_clip/releases/download/v0.2-weights/vit_l_14-laion400m_e32-3d133497.pt"),
+    laion2b_s32b_b82k=_pcfg(
+        hf_hub='laion/CLIP-ViT-L-14-laion2B-s32B-b82K/',
+        mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)),
+)
+
+_EVAL14 = dict(
+    eva=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_L_psz14.pt'),
+    eva02=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_L_psz14.pt'),
+    eva_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_L_psz14_s4B.pt'),
+    eva02_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_L_psz14_s4B.pt'),
+)
+
+_VITL14_336 = dict(
+    openai=_pcfg(
+        "https://openaipublic.azureedge.net/clip/models/3035c92b350959924f9f00213499208652fc7ea050643e8b385c2dac08641f02/ViT-L-14-336px.pt"),
+)
+
+_EVAL14_336 = dict(
+    eva_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_L_336_psz14_s6B.pt'),
+    eva02_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_L_336_psz14_s6B.pt'),
+    eva_clip_224to336=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_L_psz14_224to336.pt'),
+    eva02_clip_224to336=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_L_psz14_224to336.pt'),
+)
+
+_VITH14 = dict(
+    laion2b_s32b_b79k=_pcfg(hf_hub='laion/CLIP-ViT-H-14-laion2B-s32B-b79K/'),
+)
+
+_VITg14 = dict(
+    laion2b_s12b_b42k=_pcfg(hf_hub='laion/CLIP-ViT-g-14-laion2B-s12B-b42K/'),
+    laion2b_s34b_b88k=_pcfg(hf_hub='laion/CLIP-ViT-g-14-laion2B-s34B-b88K/'),
+)
+
+_EVAg14 = dict(
+    eva=_pcfg(hf_hub='QuanSun/EVA-CLIP/'),
+    eva01=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA01_g_psz14.pt'),
+    eva_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA01_CLIP_g_14_psz14_s11B.pt'),
+    eva01_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA01_CLIP_g_14_psz14_s11B.pt'),
+)
+
+_EVAg14_PLUS = dict(
+    eva=_pcfg(hf_hub='QuanSun/EVA-CLIP/'),
+    eva01=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA01_g_psz14.pt'),
+    eva_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA01_CLIP_g_14_plus_psz14_s11B.pt'),
+    eva01_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA01_CLIP_g_14_plus_psz14_s11B.pt'),
+)
+
+_VITbigG14 = dict(
+    laion2b_s39b_b160k=_pcfg(hf_hub='laion/CLIP-ViT-bigG-14-laion2B-39B-b160k/'),
+)
+
+_EVAbigE14 = dict(
+    eva=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_E_psz14.pt'),
+    eva02=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_E_psz14.pt'),
+    eva_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_E_psz14_s4B.pt'),
+    eva02_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_E_psz14_s4B.pt'),
+)
+
+_EVAbigE14_PLUS = dict(
+    eva=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_E_psz14.pt'),
+    eva02=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_E_psz14.pt'),
+    eva_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_E_psz14_plus_s9B.pt'),
+    eva02_clip=_pcfg(hf_hub='QuanSun/EVA-CLIP/EVA02_CLIP_E_psz14_plus_s9B.pt'),
+)
+
+
+_PRETRAINED = {
+    # "ViT-B-32": _VITB32,
+    "OpenaiCLIP-B-32": _VITB32,
+    "OpenCLIP-B-32": _VITB32,
+
+    # "ViT-B-32-quickgelu": _VITB32_quickgelu,
+    "OpenaiCLIP-B-32-quickgelu": _VITB32_quickgelu,
+    "OpenCLIP-B-32-quickgelu": _VITB32_quickgelu,
+
+    # "ViT-B-16": _VITB16,
+    "OpenaiCLIP-B-16": _VITB16,
+    "OpenCLIP-B-16": _VITB16,
+
+    "EVA02-B-16": _EVAB16,
+    "EVA02-CLIP-B-16": _EVAB16,
+
+    # "ViT-B-16-plus-240": _VITB16_PLUS_240,
+    "OpenCLIP-B-16-plus-240": _VITB16_PLUS_240,
+
+    # "ViT-L-14": _VITL14,
+    "OpenaiCLIP-L-14": _VITL14,
+    "OpenCLIP-L-14": _VITL14,
+
+    "EVA02-L-14": _EVAL14,
+    "EVA02-CLIP-L-14": _EVAL14,
+
+    # "ViT-L-14-336": _VITL14_336,
+    "OpenaiCLIP-L-14-336": _VITL14_336,
+
+    "EVA02-CLIP-L-14-336": _EVAL14_336,
+
+    # "ViT-H-14": _VITH14,
+    # "ViT-g-14": _VITg14,
+    "OpenCLIP-H-14": _VITH14,
+    "OpenCLIP-g-14": _VITg14,
+
+    "EVA01-CLIP-g-14": _EVAg14,
+    "EVA01-CLIP-g-14-plus": _EVAg14_PLUS,
+
+    # "ViT-bigG-14": _VITbigG14,
+    "OpenCLIP-bigG-14": _VITbigG14,
+
+    "EVA02-CLIP-bigE-14": _EVAbigE14,
+    "EVA02-CLIP-bigE-14-plus": _EVAbigE14_PLUS,
+}
+
+
+def _clean_tag(tag: str):
+    # normalize pretrained tags
+    return tag.lower().replace('-', '_')
+
+
+def list_pretrained(as_str: bool = False):
+    """ returns list of pretrained models
+    Returns a tuple (model_name, pretrain_tag) by default or 'name:tag' if as_str == True
+    """
+    return [':'.join([k, t]) if as_str else (k, t) for k in _PRETRAINED.keys() for t in _PRETRAINED[k].keys()]
+
+
+def list_pretrained_models_by_tag(tag: str):
+    """ return all models having the specified pretrain tag """
+    models = []
+    tag = _clean_tag(tag)
+    for k in _PRETRAINED.keys():
+        if tag in _PRETRAINED[k]:
+            models.append(k)
+    return models
+
+
+def list_pretrained_tags_by_model(model: str):
+    """ return all pretrain tags for the specified model architecture """
+    tags = []
+    if model in _PRETRAINED:
+        tags.extend(_PRETRAINED[model].keys())
+    return tags
+
+
+def is_pretrained_cfg(model: str, tag: str):
+    if model not in _PRETRAINED:
+        return False
+    return _clean_tag(tag) in _PRETRAINED[model]
+
+
+def get_pretrained_cfg(model: str, tag: str):
+    if model not in _PRETRAINED:
+        return {}
+    model_pretrained = _PRETRAINED[model]
+    return model_pretrained.get(_clean_tag(tag), {})
+
+
+def get_pretrained_url(model: str, tag: str):
+    cfg = get_pretrained_cfg(model, _clean_tag(tag))
+    return cfg.get('url', '')
+
+
+def download_pretrained_from_url(
+        url: str,
+        cache_dir: Union[str, None] = None,
+):
+    if not cache_dir:
+        cache_dir = os.path.expanduser("~/.cache/clip")
+    os.makedirs(cache_dir, exist_ok=True)
+    filename = os.path.basename(url)
+
+    if 'openaipublic' in url:
+        expected_sha256 = url.split("/")[-2]
+    elif 'mlfoundations' in url:
+        expected_sha256 = os.path.splitext(filename)[0].split("-")[-1]
+    else:
+        expected_sha256 = ''
+
+    download_target = os.path.join(cache_dir, filename)
+
+    if os.path.exists(download_target) and not os.path.isfile(download_target):
+        raise RuntimeError(f"{download_target} exists and is not a regular file")
+
+    if os.path.isfile(download_target):
+        if expected_sha256:
+            if hashlib.sha256(open(download_target, "rb").read()).hexdigest().startswith(expected_sha256):
+                return download_target
+            else:
+                warnings.warn(f"{download_target} exists, but the SHA256 checksum does not match; re-downloading the file")
+        else:
+            return download_target
+
+    with urllib.request.urlopen(url) as source, open(download_target, "wb") as output:
+        with tqdm(total=int(source.headers.get("Content-Length")), ncols=80, unit='iB', unit_scale=True) as loop:
+            while True:
+                buffer = source.read(8192)
+                if not buffer:
+                    break
+
+                output.write(buffer)
+                loop.update(len(buffer))
+
+    if expected_sha256 and not hashlib.sha256(open(download_target, "rb").read()).hexdigest().startswith(expected_sha256):
+        raise RuntimeError(f"Model has been downloaded but the SHA256 checksum does not not match")
+
+    return download_target
+
+
+def has_hf_hub(necessary=False):
+    if not _has_hf_hub and necessary:
+        # if no HF Hub module installed, and it is necessary to continue, raise error
+        raise RuntimeError(
+            'Hugging Face hub model specified but package not installed. Run `pip install huggingface_hub`.')
+    return _has_hf_hub
+
+
+def download_pretrained_from_hf(
+        model_id: str,
+        filename: str = 'open_clip_pytorch_model.bin',
+        revision=None,
+        cache_dir: Union[str, None] = None,
+):
+    has_hf_hub(True)
+    cached_file = hf_hub_download(model_id, filename, revision=revision, cache_dir=cache_dir)
+    return cached_file
+
+
+def download_pretrained(
+        cfg: Dict,
+        force_hf_hub: bool = False,
+        cache_dir: Union[str, None] = None,
+):
+    target = ''
+    if not cfg:
+        return target
+
+    download_url = cfg.get('url', '')
+    download_hf_hub = cfg.get('hf_hub', '')
+    if download_hf_hub and force_hf_hub:
+        # use HF hub even if url exists
+        download_url = ''
+
+    if download_url:
+        target = download_pretrained_from_url(download_url, cache_dir=cache_dir)
+    elif download_hf_hub:
+        has_hf_hub(True)
+        # we assume the hf_hub entries in pretrained config combine model_id + filename in
+        # 'org/model_name/filename.pt' form. To specify just the model id w/o filename and
+        # use 'open_clip_pytorch_model.bin' default, there must be a trailing slash 'org/model_name/'.
+        model_id, filename = os.path.split(download_hf_hub)
+        if filename:
+            target = download_pretrained_from_hf(model_id, filename=filename, cache_dir=cache_dir)
+        else:
+            target = download_pretrained_from_hf(model_id, cache_dir=cache_dir)
+
+    return target

consis_id/models/eva_clip/rope.py

@@ -0,0 +1,137 @@
+from math import pi
+import torch
+from torch import nn
+from einops import rearrange, repeat
+import logging
+
+def broadcat(tensors, dim = -1):
+    num_tensors = len(tensors)
+    shape_lens = set(list(map(lambda t: len(t.shape), tensors)))
+    assert len(shape_lens) == 1, 'tensors must all have the same number of dimensions'
+    shape_len = list(shape_lens)[0]
+    dim = (dim + shape_len) if dim < 0 else dim
+    dims = list(zip(*map(lambda t: list(t.shape), tensors)))
+    expandable_dims = [(i, val) for i, val in enumerate(dims) if i != dim]
+    assert all([*map(lambda t: len(set(t[1])) <= 2, expandable_dims)]), 'invalid dimensions for broadcastable concatentation'
+    max_dims = list(map(lambda t: (t[0], max(t[1])), expandable_dims))
+    expanded_dims = list(map(lambda t: (t[0], (t[1],) * num_tensors), max_dims))
+    expanded_dims.insert(dim, (dim, dims[dim]))
+    expandable_shapes = list(zip(*map(lambda t: t[1], expanded_dims)))
+    tensors = list(map(lambda t: t[0].expand(*t[1]), zip(tensors, expandable_shapes)))
+    return torch.cat(tensors, dim = dim)
+
+def rotate_half(x):
+    x = rearrange(x, '... (d r) -> ... d r', r = 2)
+    x1, x2 = x.unbind(dim = -1)
+    x = torch.stack((-x2, x1), dim = -1)
+    return rearrange(x, '... d r -> ... (d r)')
+
+
+class VisionRotaryEmbedding(nn.Module):
+    def __init__(
+        self,
+        dim,
+        pt_seq_len,
+        ft_seq_len=None,
+        custom_freqs = None,
+        freqs_for = 'lang',
+        theta = 10000,
+        max_freq = 10,
+        num_freqs = 1,
+    ):
+        super().__init__()
+        if custom_freqs:
+            freqs = custom_freqs
+        elif freqs_for == 'lang':
+            freqs = 1. / (theta ** (torch.arange(0, dim, 2)[:(dim // 2)].float() / dim))
+        elif freqs_for == 'pixel':
+            freqs = torch.linspace(1., max_freq / 2, dim // 2) * pi
+        elif freqs_for == 'constant':
+            freqs = torch.ones(num_freqs).float()
+        else:
+            raise ValueError(f'unknown modality {freqs_for}')
+
+        if ft_seq_len is None: ft_seq_len = pt_seq_len
+        t = torch.arange(ft_seq_len) / ft_seq_len * pt_seq_len
+
+        freqs_h = torch.einsum('..., f -> ... f', t, freqs)
+        freqs_h = repeat(freqs_h, '... n -> ... (n r)', r = 2)
+
+        freqs_w = torch.einsum('..., f -> ... f', t, freqs)
+        freqs_w = repeat(freqs_w, '... n -> ... (n r)', r = 2)
+
+        freqs = broadcat((freqs_h[:, None, :], freqs_w[None, :, :]), dim = -1) 
+
+        self.register_buffer("freqs_cos", freqs.cos())
+        self.register_buffer("freqs_sin", freqs.sin())
+
+        logging.info(f'Shape of rope freq: {self.freqs_cos.shape}')
+
+    def forward(self, t, start_index = 0):
+        rot_dim = self.freqs_cos.shape[-1]
+        end_index = start_index + rot_dim
+        assert rot_dim <= t.shape[-1], f'feature dimension {t.shape[-1]} is not of sufficient size to rotate in all the positions {rot_dim}'
+        t_left, t, t_right = t[..., :start_index], t[..., start_index:end_index], t[..., end_index:]
+        t = (t * self.freqs_cos) + (rotate_half(t) * self.freqs_sin)
+
+        return torch.cat((t_left, t, t_right), dim = -1)
+
+class VisionRotaryEmbeddingFast(nn.Module):
+    def __init__(
+        self,
+        dim,
+        pt_seq_len,
+        ft_seq_len=None,
+        custom_freqs = None,
+        freqs_for = 'lang',
+        theta = 10000,
+        max_freq = 10,
+        num_freqs = 1,
+        patch_dropout = 0.
+    ):
+        super().__init__()
+        if custom_freqs:
+            freqs = custom_freqs
+        elif freqs_for == 'lang':
+            freqs = 1. / (theta ** (torch.arange(0, dim, 2)[:(dim // 2)].float() / dim))
+        elif freqs_for == 'pixel':
+            freqs = torch.linspace(1., max_freq / 2, dim // 2) * pi
+        elif freqs_for == 'constant':
+            freqs = torch.ones(num_freqs).float()
+        else:
+            raise ValueError(f'unknown modality {freqs_for}')
+
+        if ft_seq_len is None: ft_seq_len = pt_seq_len
+        t = torch.arange(ft_seq_len) / ft_seq_len * pt_seq_len
+
+        freqs = torch.einsum('..., f -> ... f', t, freqs)
+        freqs = repeat(freqs, '... n -> ... (n r)', r = 2)
+        freqs = broadcat((freqs[:, None, :], freqs[None, :, :]), dim = -1)
+
+        freqs_cos = freqs.cos().view(-1, freqs.shape[-1])
+        freqs_sin = freqs.sin().view(-1, freqs.shape[-1])
+
+        self.patch_dropout = patch_dropout
+
+        self.register_buffer("freqs_cos", freqs_cos)
+        self.register_buffer("freqs_sin", freqs_sin)
+
+        logging.info(f'Shape of rope freq: {self.freqs_cos.shape}')
+
+    def forward(self, t, patch_indices_keep=None):
+        if patch_indices_keep is not None:
+            batch = t.size()[0]
+            batch_indices = torch.arange(batch)
+            batch_indices = batch_indices[..., None]
+
+            freqs_cos = repeat(self.freqs_cos, 'i j -> n i m j', n=t.shape[0], m=t.shape[1])
+            freqs_sin = repeat(self.freqs_sin, 'i j -> n i m j', n=t.shape[0], m=t.shape[1])
+
+            freqs_cos = freqs_cos[batch_indices, patch_indices_keep]
+            freqs_cos = rearrange(freqs_cos, 'n i m j -> n m i j')
+            freqs_sin = freqs_sin[batch_indices, patch_indices_keep]
+            freqs_sin = rearrange(freqs_sin, 'n i m j -> n m i j')
+
+            return  t * freqs_cos + rotate_half(t) * freqs_sin
+
+        return  t * self.freqs_cos + rotate_half(t) * self.freqs_sin

consis_id/models/eva_clip/timm_model.py

@@ -0,0 +1,122 @@
+""" timm model adapter
+
+Wraps timm (https://github.com/rwightman/pytorch-image-models) models for use as a vision tower in CLIP model.
+"""
+import logging
+from collections import OrderedDict
+
+import torch
+import torch.nn as nn
+
+try:
+    import timm
+    from timm.models.layers import Mlp, to_2tuple
+    try:
+        # old timm imports < 0.8.1
+        from timm.models.layers.attention_pool2d import RotAttentionPool2d
+        from timm.models.layers.attention_pool2d import AttentionPool2d as AbsAttentionPool2d
+    except ImportError:
+        # new timm imports >= 0.8.1
+        from timm.layers import RotAttentionPool2d
+        from timm.layers import AttentionPool2d as AbsAttentionPool2d
+except ImportError:
+    timm = None
+
+from .utils import freeze_batch_norm_2d
+
+
+class TimmModel(nn.Module):
+    """ timm model adapter
+    # FIXME this adapter is a work in progress, may change in ways that break weight compat
+    """
+
+    def __init__(
+            self,
+            model_name,
+            embed_dim,
+            image_size=224,
+            pool='avg',
+            proj='linear',
+            proj_bias=False,
+            drop=0.,
+            pretrained=False):
+        super().__init__()
+        if timm is None:
+            raise RuntimeError("Please `pip install timm` to use timm models.")
+
+        self.image_size = to_2tuple(image_size)
+        self.trunk = timm.create_model(model_name, pretrained=pretrained)
+        feat_size = self.trunk.default_cfg.get('pool_size', None)
+        feature_ndim = 1 if not feat_size else 2
+        if pool in ('abs_attn', 'rot_attn'):
+            assert feature_ndim == 2
+            # if attn pooling used, remove both classifier and default pool
+            self.trunk.reset_classifier(0, global_pool='')
+        else:
+            # reset global pool if pool config set, otherwise leave as network default
+            reset_kwargs = dict(global_pool=pool) if pool else {}
+            self.trunk.reset_classifier(0, **reset_kwargs)
+        prev_chs = self.trunk.num_features
+
+        head_layers = OrderedDict()
+        if pool == 'abs_attn':
+            head_layers['pool'] = AbsAttentionPool2d(prev_chs, feat_size=feat_size, out_features=embed_dim)
+            prev_chs = embed_dim
+        elif pool == 'rot_attn':
+            head_layers['pool'] = RotAttentionPool2d(prev_chs, out_features=embed_dim)
+            prev_chs = embed_dim
+        else:
+            assert proj, 'projection layer needed if non-attention pooling is used.'
+
+        # NOTE attention pool ends with a projection layer, so proj should usually be set to '' if such pooling is used
+        if proj == 'linear':
+            head_layers['drop'] = nn.Dropout(drop)
+            head_layers['proj'] = nn.Linear(prev_chs, embed_dim, bias=proj_bias)
+        elif proj == 'mlp':
+            head_layers['mlp'] = Mlp(prev_chs, 2 * embed_dim, embed_dim, drop=drop, bias=(True, proj_bias))
+
+        self.head = nn.Sequential(head_layers)
+
+    def lock(self, unlocked_groups=0, freeze_bn_stats=False):
+        """ lock modules
+        Args:
+            unlocked_groups (int): leave last n layer groups unlocked (default: 0)
+        """
+        if not unlocked_groups:
+            # lock full model
+            for param in self.trunk.parameters():
+                param.requires_grad = False
+            if freeze_bn_stats:
+                freeze_batch_norm_2d(self.trunk)
+        else:
+            # NOTE: partial freeze requires latest timm (master) branch and is subject to change
+            try:
+                # FIXME import here until API stable and in an official release
+                from timm.models.helpers import group_parameters, group_modules
+            except ImportError:
+                raise RuntimeError(
+                    'Please install latest timm `pip install git+https://github.com/rwightman/pytorch-image-models`')
+            matcher = self.trunk.group_matcher()
+            gparams = group_parameters(self.trunk, matcher)
+            max_layer_id = max(gparams.keys())
+            max_layer_id = max_layer_id - unlocked_groups
+            for group_idx in range(max_layer_id + 1):
+                group = gparams[group_idx]
+                for param in group:
+                    self.trunk.get_parameter(param).requires_grad = False
+            if freeze_bn_stats:
+                gmodules = group_modules(self.trunk, matcher, reverse=True)
+                gmodules = {k for k, v in gmodules.items() if v <= max_layer_id}
+                freeze_batch_norm_2d(self.trunk, gmodules)
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        try:
+            self.trunk.set_grad_checkpointing(enable)
+        except Exception as e:
+            logging.warning('grad checkpointing not supported for this timm image tower, continuing without...')
+
+    def forward(self, x):
+        x = self.trunk(x)
+        x = self.head(x)
+        return x

consis_id/models/eva_clip/tokenizer.py

@@ -0,0 +1,201 @@
+""" CLIP tokenizer
+
+Copied from https://github.com/openai/CLIP. Originally MIT License, Copyright (c) 2021 OpenAI.
+"""
+import gzip
+import html
+import os
+from functools import lru_cache
+from typing import Union, List
+
+import ftfy
+import regex as re
+import torch
+
+# https://stackoverflow.com/q/62691279
+import os
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
+
+@lru_cache()
+def default_bpe():
+    return os.path.join(os.path.dirname(os.path.abspath(__file__)), "bpe_simple_vocab_16e6.txt.gz")
+
+
+@lru_cache()
+def bytes_to_unicode():
+    """
+    Returns list of utf-8 byte and a corresponding list of unicode strings.
+    The reversible bpe codes work on unicode strings.
+    This means you need a large # of unicode characters in your vocab if you want to avoid UNKs.
+    When you're at something like a 10B token dataset you end up needing around 5K for decent coverage.
+    This is a signficant percentage of your normal, say, 32K bpe vocab.
+    To avoid that, we want lookup tables between utf-8 bytes and unicode strings.
+    And avoids mapping to whitespace/control characters the bpe code barfs on.
+    """
+    bs = list(range(ord("!"), ord("~")+1))+list(range(ord("¡"), ord("¬")+1))+list(range(ord("®"), ord("ÿ")+1))
+    cs = bs[:]
+    n = 0
+    for b in range(2**8):
+        if b not in bs:
+            bs.append(b)
+            cs.append(2**8+n)
+            n += 1
+    cs = [chr(n) for n in cs]
+    return dict(zip(bs, cs))
+
+
+def get_pairs(word):
+    """Return set of symbol pairs in a word.
+    Word is represented as tuple of symbols (symbols being variable-length strings).
+    """
+    pairs = set()
+    prev_char = word[0]
+    for char in word[1:]:
+        pairs.add((prev_char, char))
+        prev_char = char
+    return pairs
+
+
+def basic_clean(text):
+    text = ftfy.fix_text(text)
+    text = html.unescape(html.unescape(text))
+    return text.strip()
+
+
+def whitespace_clean(text):
+    text = re.sub(r'\s+', ' ', text)
+    text = text.strip()
+    return text
+
+
+class SimpleTokenizer(object):
+    def __init__(self, bpe_path: str = default_bpe(), special_tokens=None):
+        self.byte_encoder = bytes_to_unicode()
+        self.byte_decoder = {v: k for k, v in self.byte_encoder.items()}
+        merges = gzip.open(bpe_path).read().decode("utf-8").split('\n')
+        merges = merges[1:49152-256-2+1]
+        merges = [tuple(merge.split()) for merge in merges]
+        vocab = list(bytes_to_unicode().values())
+        vocab = vocab + [v+'</w>' for v in vocab]
+        for merge in merges:
+            vocab.append(''.join(merge))
+        if not special_tokens:
+            special_tokens = ['<start_of_text>', '<end_of_text>']
+        else:
+            special_tokens = ['<start_of_text>', '<end_of_text>'] + special_tokens
+        vocab.extend(special_tokens)
+        self.encoder = dict(zip(vocab, range(len(vocab))))
+        self.decoder = {v: k for k, v in self.encoder.items()}
+        self.bpe_ranks = dict(zip(merges, range(len(merges))))
+        self.cache = {t:t for t in special_tokens}
+        special = "|".join(special_tokens)
+        self.pat = re.compile(special + r"""|'s|'t|'re|'ve|'m|'ll|'d|[\p{L}]+|[\p{N}]|[^\s\p{L}\p{N}]+""", re.IGNORECASE)
+
+        self.vocab_size = len(self.encoder)
+        self.all_special_ids = [self.encoder[t] for t in special_tokens]
+
+    def bpe(self, token):
+        if token in self.cache:
+            return self.cache[token]
+        word = tuple(token[:-1]) + ( token[-1] + '</w>',)
+        pairs = get_pairs(word)
+
+        if not pairs:
+            return token+'</w>'
+
+        while True:
+            bigram = min(pairs, key = lambda pair: self.bpe_ranks.get(pair, float('inf')))
+            if bigram not in self.bpe_ranks:
+                break
+            first, second = bigram
+            new_word = []
+            i = 0
+            while i < len(word):
+                try:
+                    j = word.index(first, i)
+                    new_word.extend(word[i:j])
+                    i = j
+                except:
+                    new_word.extend(word[i:])
+                    break
+
+                if word[i] == first and i < len(word)-1 and word[i+1] == second:
+                    new_word.append(first+second)
+                    i += 2
+                else:
+                    new_word.append(word[i])
+                    i += 1
+            new_word = tuple(new_word)
+            word = new_word
+            if len(word) == 1:
+                break
+            else:
+                pairs = get_pairs(word)
+        word = ' '.join(word)
+        self.cache[token] = word
+        return word
+
+    def encode(self, text):
+        bpe_tokens = []
+        text = whitespace_clean(basic_clean(text)).lower()
+        for token in re.findall(self.pat, text):
+            token = ''.join(self.byte_encoder[b] for b in token.encode('utf-8'))
+            bpe_tokens.extend(self.encoder[bpe_token] for bpe_token in self.bpe(token).split(' '))
+        return bpe_tokens
+
+    def decode(self, tokens):
+        text = ''.join([self.decoder[token] for token in tokens])
+        text = bytearray([self.byte_decoder[c] for c in text]).decode('utf-8', errors="replace").replace('</w>', ' ')
+        return text
+
+
+_tokenizer = SimpleTokenizer()
+
+
+def tokenize(texts: Union[str, List[str]], context_length: int = 77) -> torch.LongTensor:
+    """
+    Returns the tokenized representation of given input string(s)
+
+    Parameters
+    ----------
+    texts : Union[str, List[str]]
+        An input string or a list of input strings to tokenize
+    context_length : int
+        The context length to use; all CLIP models use 77 as the context length
+
+    Returns
+    -------
+    A two-dimensional tensor containing the resulting tokens, shape = [number of input strings, context_length]
+    """
+    if isinstance(texts, str):
+        texts = [texts]
+
+    sot_token = _tokenizer.encoder["<start_of_text>"]
+    eot_token = _tokenizer.encoder["<end_of_text>"]
+    all_tokens = [[sot_token] + _tokenizer.encode(text) + [eot_token] for text in texts]
+    result = torch.zeros(len(all_tokens), context_length, dtype=torch.long)
+
+    for i, tokens in enumerate(all_tokens):
+        if len(tokens) > context_length:
+            tokens = tokens[:context_length]  # Truncate
+            tokens[-1] = eot_token
+        result[i, :len(tokens)] = torch.tensor(tokens)
+
+    return result
+
+
+class HFTokenizer:
+    "HuggingFace tokenizer wrapper"
+    def __init__(self, tokenizer_name:str):
+        from transformers import AutoTokenizer
+        self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
+
+    def __call__(self, texts:Union[str, List[str]], context_length:int=77) -> torch.Tensor:
+        # same cleaning as for default tokenizer, except lowercasing
+        # adding lower (for case-sensitive tokenizers) will make it more robust but less sensitive to nuance
+        if isinstance(texts, str):
+            texts = [texts]
+        texts = [whitespace_clean(basic_clean(text)) for text in texts]
+        input_ids = self.tokenizer(texts, return_tensors='pt', max_length=context_length, padding='max_length', truncation=True).input_ids
+        return input_ids

consis_id/models/eva_clip/transform.py

@@ -0,0 +1,103 @@
+from typing import Optional, Sequence, Tuple
+
+import torch
+import torch.nn as nn
+import torchvision.transforms.functional as F
+
+from torchvision.transforms import Normalize, Compose, RandomResizedCrop, InterpolationMode, ToTensor, Resize, \
+    CenterCrop
+
+from .constants import OPENAI_DATASET_MEAN, OPENAI_DATASET_STD
+
+
+class ResizeMaxSize(nn.Module):
+
+    def __init__(self, max_size, interpolation=InterpolationMode.BICUBIC, fn='max', fill=0):
+        super().__init__()
+        if not isinstance(max_size, int):
+            raise TypeError(f"Size should be int. Got {type(max_size)}")
+        self.max_size = max_size
+        self.interpolation = interpolation
+        self.fn = min if fn == 'min' else min
+        self.fill = fill
+
+    def forward(self, img):
+        if isinstance(img, torch.Tensor):
+            height, width = img.shape[:2]
+        else:
+            width, height = img.size
+        scale = self.max_size / float(max(height, width))
+        if scale != 1.0:
+            new_size = tuple(round(dim * scale) for dim in (height, width))
+            img = F.resize(img, new_size, self.interpolation)
+            pad_h = self.max_size - new_size[0]
+            pad_w = self.max_size - new_size[1]
+            img = F.pad(img, padding=[pad_w//2, pad_h//2, pad_w - pad_w//2, pad_h - pad_h//2], fill=self.fill)
+        return img
+
+
+def _convert_to_rgb(image):
+    return image.convert('RGB')
+
+
+# class CatGen(nn.Module):
+#     def __init__(self, num=4):
+#         self.num = num
+#     def mixgen_batch(image, text):
+#         batch_size = image.shape[0]
+#         index = np.random.permutation(batch_size)
+
+#         cat_images = []
+#         for i in range(batch_size):
+#             # image mixup
+#             image[i,:] = lam * image[i,:] + (1 - lam) * image[index[i],:]
+#             # text concat
+#             text[i] = tokenizer((str(text[i]) + " " + str(text[index[i]])))[0]
+#         text = torch.stack(text)
+#         return image, text
+
+
+def image_transform(
+        image_size: int,
+        is_train: bool,
+        mean: Optional[Tuple[float, ...]] = None,
+        std: Optional[Tuple[float, ...]] = None,
+        resize_longest_max: bool = False,
+        fill_color: int = 0,
+):
+    mean = mean or OPENAI_DATASET_MEAN
+    if not isinstance(mean, (list, tuple)):
+        mean = (mean,) * 3
+
+    std = std or OPENAI_DATASET_STD
+    if not isinstance(std, (list, tuple)):
+        std = (std,) * 3
+
+    if isinstance(image_size, (list, tuple)) and image_size[0] == image_size[1]:
+        # for square size, pass size as int so that Resize() uses aspect preserving shortest edge
+        image_size = image_size[0]
+
+    normalize = Normalize(mean=mean, std=std)
+    if is_train:
+        return Compose([
+            RandomResizedCrop(image_size, scale=(0.9, 1.0), interpolation=InterpolationMode.BICUBIC),
+            _convert_to_rgb,
+            ToTensor(),
+            normalize,
+        ])
+    else:
+        if resize_longest_max:
+            transforms = [
+                ResizeMaxSize(image_size, fill=fill_color)
+            ]
+        else:
+            transforms = [
+                Resize(image_size, interpolation=InterpolationMode.BICUBIC),
+                CenterCrop(image_size),
+            ]
+        transforms.extend([
+            _convert_to_rgb,
+            ToTensor(),
+            normalize,
+        ])
+        return Compose(transforms)

consis_id/models/eva_clip/transformer.py

@@ -0,0 +1,737 @@
+import os
+import logging
+from collections import OrderedDict
+import math
+from typing import Callable, Optional, Sequence
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+try:
+    from timm.models.layers import trunc_normal_
+except:
+    from timm.layers import trunc_normal_
+    
+from .rope import VisionRotaryEmbedding, VisionRotaryEmbeddingFast
+from .utils import to_2tuple
+
+if os.getenv('ENV_TYPE') == 'deepspeed':
+    try:
+        import deepspeed
+        from deepspeed.runtime.activation_checkpointing.checkpointing import checkpoint
+    except:
+        print("Please 'pip install deepspeed'")
+        deepspeed = None
+        from torch.utils.checkpoint import checkpoint
+else:
+    from torch.utils.checkpoint import checkpoint
+
+try:
+    import xformers.ops as xops
+except ImportError:
+    xops = None
+    print("Please 'pip install xformers'")
+
+class LayerNormFp32(nn.LayerNorm):
+    """Subclass torch's LayerNorm to handle fp16 (by casting to float32 and back)."""
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+    def forward(self, x: torch.Tensor):
+        output = F.layer_norm(
+            x.float(),
+            self.normalized_shape,
+            self.weight.float() if self.weight is not None else None,
+            self.bias.float() if self.bias is not None else None,
+            self.eps,
+        )
+        return output.type_as(x)
+
+
+class LayerNorm(nn.LayerNorm):
+    """Subclass torch's LayerNorm (with cast back to input dtype)."""
+
+    def forward(self, x: torch.Tensor):
+        orig_type = x.dtype
+        x = F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps)
+        return x.to(orig_type)
+
+class QuickGELU(nn.Module):
+    # NOTE This is slower than nn.GELU or nn.SiLU and uses more GPU memory
+    def forward(self, x: torch.Tensor):
+        return x * torch.sigmoid(1.702 * x)
+
+
+class LayerScale(nn.Module):
+    def __init__(self, dim, init_values=1e-5, inplace=False):
+        super().__init__()
+        self.inplace = inplace
+        self.gamma = nn.Parameter(init_values * torch.ones(dim))
+
+    def forward(self, x):
+        return x.mul_(self.gamma) if self.inplace else x * self.gamma
+
+class PatchDropout(nn.Module):
+    """
+    https://arxiv.org/abs/2212.00794
+    """
+
+    def __init__(self, prob, exclude_first_token=True):
+        super().__init__()
+        assert 0 <= prob < 1.
+        self.prob = prob
+        self.exclude_first_token = exclude_first_token  # exclude CLS token
+        logging.info(f"os.getenv('RoPE')={os.getenv('RoPE')}")
+
+    def forward(self, x):
+        if not self.training or self.prob == 0.:
+            return x
+
+        if self.exclude_first_token:
+            cls_tokens, x = x[:, :1], x[:, 1:]
+        else:
+            cls_tokens = torch.jit.annotate(torch.Tensor, x[:, :1])
+
+        batch = x.size()[0]
+        num_tokens = x.size()[1]
+
+        batch_indices = torch.arange(batch)
+        batch_indices = batch_indices[..., None]
+
+        keep_prob = 1 - self.prob
+        num_patches_keep = max(1, int(num_tokens * keep_prob))
+
+        rand = torch.randn(batch, num_tokens)
+        patch_indices_keep = rand.topk(num_patches_keep, dim=-1).indices
+
+        x = x[batch_indices, patch_indices_keep]
+
+        if self.exclude_first_token:
+            x = torch.cat((cls_tokens, x), dim=1)
+
+        if self.training and os.getenv('RoPE') == '1':
+            return x, patch_indices_keep
+
+        return x
+
+
+def _in_projection_packed(
+    q: torch.Tensor,
+    k: torch.Tensor,
+    v: torch.Tensor,
+    w: torch.Tensor,
+    b: Optional[torch.Tensor] = None,
+    ):
+    """
+    https://github.com/pytorch/pytorch/blob/db2a237763eb8693a20788be94f8c192e762baa8/torch/nn/functional.py#L4726
+    """
+    E = q.size(-1)
+    if k is v:
+        if q is k:
+            # self-attention
+            return F.linear(q, w, b).chunk(3, dim=-1)
+        else:
+            # encoder-decoder attention
+            w_q, w_kv = w.split([E, E * 2])
+            if b is None:
+                b_q = b_kv = None
+            else:
+                b_q, b_kv = b.split([E, E * 2])
+            return (F.linear(q, w_q, b_q),) + F.linear(k, w_kv, b_kv).chunk(2, dim=-1)
+    else:
+        w_q, w_k, w_v = w.chunk(3)
+        if b is None:
+            b_q = b_k = b_v = None
+        else:
+            b_q, b_k, b_v = b.chunk(3)
+        return F.linear(q, w_q, b_q), F.linear(k, w_k, b_k), F.linear(v, w_v, b_v)
+
+class Attention(nn.Module):
+    def __init__(
+            self,
+            dim,
+            num_heads=8,
+            qkv_bias=True,
+            scaled_cosine=False,
+            scale_heads=False,
+            logit_scale_max=math.log(1. / 0.01),
+            attn_drop=0.,
+            proj_drop=0.,
+            xattn=False,
+            rope=False
+    ):
+        super().__init__()
+        self.scaled_cosine = scaled_cosine
+        self.scale_heads = scale_heads
+        assert dim % num_heads == 0, 'dim should be divisible by num_heads'
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+        self.scale = self.head_dim ** -0.5
+        self.logit_scale_max = logit_scale_max
+
+        # keeping in_proj in this form (instead of nn.Linear) to match weight scheme of original
+        self.in_proj_weight = nn.Parameter(torch.randn((dim * 3, dim)) * self.scale)
+        if qkv_bias:
+            self.in_proj_bias = nn.Parameter(torch.zeros(dim * 3))
+        else:
+            self.in_proj_bias = None
+
+        if self.scaled_cosine:
+            self.logit_scale = nn.Parameter(torch.log(10 * torch.ones((num_heads, 1, 1))))
+        else:
+            self.logit_scale = None
+        self.attn_drop = nn.Dropout(attn_drop)
+        if self.scale_heads:
+            self.head_scale = nn.Parameter(torch.ones((num_heads, 1, 1)))
+        else:
+            self.head_scale = None
+        self.out_proj = nn.Linear(dim, dim)
+        self.out_drop = nn.Dropout(proj_drop)
+        self.xattn = xattn
+        self.xattn_drop = attn_drop
+        self.rope = rope
+
+    def forward(self, x, attn_mask: Optional[torch.Tensor] = None):
+        L, N, C = x.shape
+        q, k, v = F.linear(x, self.in_proj_weight, self.in_proj_bias).chunk(3, dim=-1)
+        if self.xattn:
+            q = q.contiguous().view(L, N, self.num_heads, -1).transpose(0, 1)
+            k = k.contiguous().view(L, N, self.num_heads, -1).transpose(0, 1)
+            v = v.contiguous().view(L, N, self.num_heads, -1).transpose(0, 1)
+
+            x = xops.memory_efficient_attention(
+                q, k, v,
+                p=self.xattn_drop,
+                scale=self.scale if self.logit_scale is None else None,
+                attn_bias=xops.LowerTriangularMask() if attn_mask is not None else None,
+                )
+        else:
+            q = q.contiguous().view(L, N * self.num_heads, -1).transpose(0, 1)
+            k = k.contiguous().view(L, N * self.num_heads, -1).transpose(0, 1)
+            v = v.contiguous().view(L, N * self.num_heads, -1).transpose(0, 1)
+
+            if self.logit_scale is not None:
+                attn = torch.bmm(F.normalize(q, dim=-1), F.normalize(k, dim=-1).transpose(-1, -2))
+                logit_scale = torch.clamp(self.logit_scale, max=self.logit_scale_max).exp()
+                attn = attn.view(N, self.num_heads, L, L) * logit_scale
+                attn = attn.view(-1, L, L)
+            else:
+                q = q * self.scale
+                attn = torch.bmm(q, k.transpose(-1, -2))
+
+            if attn_mask is not None:
+                if attn_mask.dtype == torch.bool:
+                    new_attn_mask = torch.zeros_like(attn_mask, dtype=q.dtype)
+                    new_attn_mask.masked_fill_(attn_mask, float("-inf"))
+                    attn_mask = new_attn_mask
+                attn += attn_mask
+
+            attn = attn.softmax(dim=-1)
+            attn = self.attn_drop(attn)
+
+            x = torch.bmm(attn, v)
+
+        if self.head_scale is not None:
+            x = x.view(N, self.num_heads, L, C) * self.head_scale
+            x = x.view(-1, L, C)
+        x = x.transpose(0, 1).reshape(L, N, C)
+        x = self.out_proj(x)
+        x = self.out_drop(x)
+        return x
+
+class CustomAttention(nn.Module):
+    def __init__(
+            self,
+            dim,
+            num_heads=8,
+            qkv_bias=True,
+            scaled_cosine=True,
+            scale_heads=False,
+            logit_scale_max=math.log(1. / 0.01),
+            attn_drop=0.,
+            proj_drop=0.,
+            xattn=False
+    ):
+        super().__init__()
+        self.scaled_cosine = scaled_cosine
+        self.scale_heads = scale_heads
+        assert dim % num_heads == 0, 'dim should be divisible by num_heads'
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+        self.scale = self.head_dim ** -0.5
+        self.logit_scale_max = logit_scale_max
+
+        # keeping in_proj in this form (instead of nn.Linear) to match weight scheme of original
+        self.in_proj_weight = nn.Parameter(torch.randn((dim * 3, dim)) * self.scale)
+        if qkv_bias:
+            self.in_proj_bias = nn.Parameter(torch.zeros(dim * 3))
+        else:
+            self.in_proj_bias = None
+
+        if self.scaled_cosine:
+            self.logit_scale = nn.Parameter(torch.log(10 * torch.ones((num_heads, 1, 1))))
+        else:
+            self.logit_scale = None
+        self.attn_drop = nn.Dropout(attn_drop)
+        if self.scale_heads:
+            self.head_scale = nn.Parameter(torch.ones((num_heads, 1, 1)))
+        else:
+            self.head_scale = None
+        self.out_proj = nn.Linear(dim, dim)
+        self.out_drop = nn.Dropout(proj_drop)
+        self.xattn = xattn
+        self.xattn_drop = attn_drop
+
+    def forward(self, query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
+        q, k, v = _in_projection_packed(query, key, value, self.in_proj_weight, self.in_proj_bias)
+        N_q, B_q, C_q = q.shape
+        N_k, B_k, C_k = k.shape
+        N_v, B_v, C_v = v.shape
+        if self.xattn:
+            # B, N, C -> B, N, num_heads, C
+            q = q.permute(1, 0, 2).reshape(B_q, N_q, self.num_heads, -1)
+            k = k.permute(1, 0, 2).reshape(B_k, N_k, self.num_heads, -1)
+            v = v.permute(1, 0, 2).reshape(B_v, N_v, self.num_heads, -1)
+
+            x = xops.memory_efficient_attention(
+                q, k, v,
+                p=self.xattn_drop,
+                scale=self.scale if self.logit_scale is None else None,
+                attn_bias=xops.LowerTriangularMask() if attn_mask is not None else None
+                )
+        else:
+            # B*H, L, C
+            q = q.contiguous().view(N_q, B_q * self.num_heads, -1).transpose(0, 1)
+            k = k.contiguous().view(N_k, B_k * self.num_heads, -1).transpose(0, 1)
+            v = v.contiguous().view(N_v, B_v * self.num_heads, -1).transpose(0, 1)
+
+            if self.logit_scale is not None:
+                # B*H, N_q, N_k
+                attn = torch.bmm(F.normalize(q, dim=-1), F.normalize(k, dim=-1).transpose(-1, -2))
+                logit_scale = torch.clamp(self.logit_scale, max=self.logit_scale_max).exp()
+                attn = attn.view(B_q, self.num_heads, N_q, N_k) * logit_scale
+                attn = attn.view(-1, N_q, N_k)
+            else:
+                q = q * self.scale
+                attn = torch.bmm(q, k.transpose(-1, -2))
+
+            if attn_mask is not None:
+                if attn_mask.dtype == torch.bool:
+                    new_attn_mask = torch.zeros_like(attn_mask, dtype=q.dtype)
+                    new_attn_mask.masked_fill_(attn_mask, float("-inf"))
+                    attn_mask = new_attn_mask
+                attn += attn_mask
+
+            attn = attn.softmax(dim=-1)
+            attn = self.attn_drop(attn)
+
+            x = torch.bmm(attn, v)
+            
+        if self.head_scale is not None:
+            x = x.view(B_q, self.num_heads, N_q, C_q) * self.head_scale
+            x = x.view(-1, N_q, C_q)
+        x = x.transpose(0, 1).reshape(N_q, B_q, C_q)
+        x = self.out_proj(x)
+        x = self.out_drop(x)
+        return x
+
+class CustomResidualAttentionBlock(nn.Module):
+    def __init__(
+            self,
+            d_model: int,
+            n_head: int,
+            mlp_ratio: float = 4.0,
+            ls_init_value: float = None,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = LayerNorm,
+            scale_cosine_attn: bool = False,
+            scale_heads: bool = False,
+            scale_attn: bool = False,
+            scale_fc: bool = False,
+            cross_attn: bool = False,
+            xattn: bool = False,
+    ):
+        super().__init__()
+
+        self.ln_1 = norm_layer(d_model)
+        self.ln_1_k = norm_layer(d_model) if cross_attn else self.ln_1
+        self.ln_1_v = norm_layer(d_model) if cross_attn else self.ln_1
+        self.attn = CustomAttention(
+            d_model, n_head,
+            qkv_bias=True,
+            attn_drop=0.,
+            proj_drop=0.,
+            scaled_cosine=scale_cosine_attn,
+            scale_heads=scale_heads,
+            xattn=xattn
+        )
+
+        self.ln_attn = norm_layer(d_model) if scale_attn else nn.Identity()
+        self.ls_1 = LayerScale(d_model, ls_init_value) if ls_init_value is not None else nn.Identity()
+
+        self.ln_2 = norm_layer(d_model)
+        mlp_width = int(d_model * mlp_ratio)
+        self.mlp = nn.Sequential(OrderedDict([
+            ("c_fc", nn.Linear(d_model, mlp_width)),
+            ('ln', norm_layer(mlp_width) if scale_fc else nn.Identity()),
+            ("gelu", act_layer()),
+            ("c_proj", nn.Linear(mlp_width, d_model))
+        ]))
+
+        self.ls_2 = LayerScale(d_model, ls_init_value) if ls_init_value is not None else nn.Identity()
+
+    def forward(self, q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
+        q = q + self.ls_1(self.ln_attn(self.attn(self.ln_1(q), self.ln_1_k(k), self.ln_1_v(v), attn_mask=attn_mask)))
+        q = q + self.ls_2(self.mlp(self.ln_2(q)))
+        return q
+
+class CustomTransformer(nn.Module):
+    def __init__(
+            self,
+            width: int,
+            layers: int,
+            heads: int,
+            mlp_ratio: float = 4.0,
+            ls_init_value: float = None,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = LayerNorm,
+            scale_cosine_attn: bool = True,
+            scale_heads: bool = False,
+            scale_attn: bool = False,
+            scale_fc: bool = False,
+            cross_attn: bool = False,
+            xattn: bool = False,
+    ):
+        super().__init__()
+        self.width = width
+        self.layers = layers
+        self.grad_checkpointing = False
+        self.xattn = xattn
+
+        self.resblocks = nn.ModuleList([
+            CustomResidualAttentionBlock(
+                width,
+                heads,
+                mlp_ratio,
+                ls_init_value=ls_init_value,
+                act_layer=act_layer,
+                norm_layer=norm_layer,
+                scale_cosine_attn=scale_cosine_attn,
+                scale_heads=scale_heads,
+                scale_attn=scale_attn,
+                scale_fc=scale_fc,
+                cross_attn=cross_attn,
+                xattn=xattn)
+            for _ in range(layers)
+        ])
+
+    def get_cast_dtype(self) -> torch.dtype:
+        return self.resblocks[0].mlp.c_fc.weight.dtype 
+
+    def forward(self, q: torch.Tensor, k: torch.Tensor = None, v: torch.Tensor = None, attn_mask: Optional[torch.Tensor] = None):
+        if k is None and v is None:
+            k = v = q
+        for r in self.resblocks:
+            if self.grad_checkpointing and not torch.jit.is_scripting():
+                q = checkpoint(r, q, k, v, attn_mask)
+            else:
+                q = r(q, k, v, attn_mask=attn_mask)
+        return q
+
+
+class ResidualAttentionBlock(nn.Module):
+    def __init__(
+            self,
+            d_model: int,
+            n_head: int,
+            mlp_ratio: float = 4.0,
+            ls_init_value: float = None,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = LayerNorm,
+            xattn: bool = False,
+    ):
+        super().__init__()
+
+        self.ln_1 = norm_layer(d_model)
+        if xattn:
+            self.attn = Attention(d_model, n_head, xattn=True)
+        else:
+            self.attn = nn.MultiheadAttention(d_model, n_head)
+        self.ls_1 = LayerScale(d_model, ls_init_value) if ls_init_value is not None else nn.Identity()
+
+        self.ln_2 = norm_layer(d_model)
+        mlp_width = int(d_model * mlp_ratio)
+        self.mlp = nn.Sequential(OrderedDict([
+            ("c_fc", nn.Linear(d_model, mlp_width)),
+            ("gelu", act_layer()),
+            ("c_proj", nn.Linear(mlp_width, d_model))
+        ]))
+
+        self.ls_2 = LayerScale(d_model, ls_init_value) if ls_init_value is not None else nn.Identity()
+        self.xattn = xattn
+
+    def attention(self, x: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
+        attn_mask = attn_mask.to(x.dtype) if attn_mask is not None else None
+        if self.xattn:
+            return self.attn(x, attn_mask=attn_mask)
+        return self.attn(x, x, x, need_weights=False, attn_mask=attn_mask)[0]
+
+    def forward(self, x: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
+        x = x + self.ls_1(self.attention(self.ln_1(x), attn_mask=attn_mask))
+        x = x + self.ls_2(self.mlp(self.ln_2(x)))
+        return x
+
+class Transformer(nn.Module):
+    def __init__(
+            self,
+            width: int,
+            layers: int,
+            heads: int,
+            mlp_ratio: float = 4.0,
+            ls_init_value: float = None,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = LayerNorm,
+            xattn: bool = False,
+    ):
+        super().__init__()
+        self.width = width
+        self.layers = layers
+        self.grad_checkpointing = False
+
+        self.resblocks = nn.ModuleList([
+            ResidualAttentionBlock(
+                width, heads, mlp_ratio, ls_init_value=ls_init_value, act_layer=act_layer, norm_layer=norm_layer, xattn=xattn)
+            for _ in range(layers)
+        ])
+
+    def get_cast_dtype(self) -> torch.dtype:
+        return self.resblocks[0].mlp.c_fc.weight.dtype
+
+    def forward(self, x: torch.Tensor, attn_mask: Optional[torch.Tensor] = None):
+        for r in self.resblocks:
+            if self.grad_checkpointing and not torch.jit.is_scripting():
+                x = checkpoint(r, x, attn_mask)
+            else:
+                x = r(x, attn_mask=attn_mask)
+        return x
+
+
+class VisionTransformer(nn.Module):
+    def __init__(
+            self,
+            image_size: int,
+            patch_size: int,
+            width: int,
+            layers: int,
+            heads: int,
+            mlp_ratio: float,
+            ls_init_value: float = None,
+            patch_dropout: float = 0.,
+            global_average_pool: bool = False,
+            output_dim: int = 512,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = LayerNorm,
+            xattn: bool = False,
+    ):
+        super().__init__()
+        self.image_size = to_2tuple(image_size)
+        self.patch_size = to_2tuple(patch_size)
+        self.grid_size = (self.image_size[0] // self.patch_size[0], self.image_size[1] // self.patch_size[1])
+        self.output_dim = output_dim
+        self.conv1 = nn.Conv2d(in_channels=3, out_channels=width, kernel_size=patch_size, stride=patch_size, bias=False)
+
+        scale = width ** -0.5
+        self.class_embedding = nn.Parameter(scale * torch.randn(width))
+        self.positional_embedding = nn.Parameter(scale * torch.randn(self.grid_size[0] * self.grid_size[1] + 1, width))
+
+        # setting a patch_dropout of 0. would mean it is disabled and this function would be the identity fn
+        self.patch_dropout = PatchDropout(patch_dropout) if patch_dropout > 0. else nn.Identity()
+        self.ln_pre = norm_layer(width)
+        
+        self.transformer = Transformer(
+            width,
+            layers,
+            heads,
+            mlp_ratio,
+            ls_init_value=ls_init_value,
+            act_layer=act_layer,
+            norm_layer=norm_layer,
+            xattn=xattn
+        )
+
+        self.global_average_pool = global_average_pool
+        self.ln_post = norm_layer(width)
+        self.proj = nn.Parameter(scale * torch.randn(width, output_dim))
+
+    def lock(self, unlocked_groups=0, freeze_bn_stats=False):
+        for param in self.parameters():
+            param.requires_grad = False
+        
+        if unlocked_groups != 0:
+            groups = [
+                [
+                    self.conv1,
+                    self.class_embedding,
+                    self.positional_embedding,
+                    self.ln_pre,
+                ],
+                *self.transformer.resblocks[:-1],
+                [
+                    self.transformer.resblocks[-1],
+                    self.ln_post,
+                ],
+                self.proj,
+            ]
+
+            def _unlock(x):
+                if isinstance(x, Sequence):
+                    for g in x:
+                        _unlock(g)
+                else:
+                    if isinstance(x, torch.nn.Parameter):
+                        x.requires_grad = True
+                    else:
+                        for p in x.parameters():
+                            p.requires_grad = True
+
+            _unlock(groups[-unlocked_groups:])
+
+    def get_num_layers(self):
+        return self.transformer.layers
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        self.transformer.grad_checkpointing = enable
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'positional_embedding', 'class_embedding'}
+
+    def forward(self, x: torch.Tensor, return_all_features: bool=False):
+        x = self.conv1(x)  # shape = [*, width, grid, grid]
+        x = x.reshape(x.shape[0], x.shape[1], -1)  # shape = [*, width, grid ** 2]
+        x = x.permute(0, 2, 1)  # shape = [*, grid ** 2, width]
+        x = torch.cat(
+            [self.class_embedding.to(x.dtype) + torch.zeros(x.shape[0], 1, x.shape[-1], dtype=x.dtype, device=x.device),
+             x], dim=1)  # shape = [*, grid ** 2 + 1, width]
+        x = x + self.positional_embedding.to(x.dtype)
+
+        # a patch_dropout of 0. would mean it is disabled and this function would do nothing but return what was passed in
+        x = self.patch_dropout(x)
+        x = self.ln_pre(x)
+
+        x = x.permute(1, 0, 2)  # NLD -> LND
+        x = self.transformer(x)
+        x = x.permute(1, 0, 2)  # LND -> NLD
+
+        if not return_all_features:
+            if self.global_average_pool:
+                x = x.mean(dim=1) #x = x[:,1:,:].mean(dim=1)
+            else:
+                x = x[:, 0]
+
+            x = self.ln_post(x)
+
+            if self.proj is not None:
+                x = x @ self.proj
+
+        return x
+
+
+class TextTransformer(nn.Module):
+    def __init__(
+            self,
+            context_length: int = 77,
+            vocab_size: int = 49408,
+            width: int = 512,
+            heads: int = 8,
+            layers: int = 12,
+            ls_init_value: float = None,
+            output_dim: int = 512,
+            act_layer: Callable = nn.GELU,
+            norm_layer: Callable = LayerNorm,
+            xattn: bool= False,
+            attn_mask: bool = True
+    ):
+        super().__init__()
+        self.context_length = context_length
+        self.vocab_size = vocab_size
+        self.width = width
+        self.output_dim = output_dim
+
+        self.token_embedding = nn.Embedding(vocab_size, width)
+        self.positional_embedding = nn.Parameter(torch.empty(self.context_length, width))
+        self.transformer = Transformer(
+            width=width,
+            layers=layers,
+            heads=heads,
+            ls_init_value=ls_init_value,
+            act_layer=act_layer,
+            norm_layer=norm_layer,
+            xattn=xattn
+        )
+        
+        self.xattn = xattn
+        self.ln_final = norm_layer(width)
+        self.text_projection = nn.Parameter(torch.empty(width, output_dim))
+
+        if attn_mask:
+            self.register_buffer('attn_mask', self.build_attention_mask(), persistent=False)
+        else:
+            self.attn_mask = None
+
+        self.init_parameters()
+
+    def init_parameters(self):
+        nn.init.normal_(self.token_embedding.weight, std=0.02)
+        nn.init.normal_(self.positional_embedding, std=0.01)
+
+        proj_std = (self.transformer.width ** -0.5) * ((2 * self.transformer.layers) ** -0.5)
+        attn_std = self.transformer.width ** -0.5
+        fc_std = (2 * self.transformer.width) ** -0.5
+        for block in self.transformer.resblocks:
+            nn.init.normal_(block.attn.in_proj_weight, std=attn_std)
+            nn.init.normal_(block.attn.out_proj.weight, std=proj_std)
+            nn.init.normal_(block.mlp.c_fc.weight, std=fc_std)
+            nn.init.normal_(block.mlp.c_proj.weight, std=proj_std)
+
+        if self.text_projection is not None:
+            nn.init.normal_(self.text_projection, std=self.transformer.width ** -0.5)
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        self.transformer.grad_checkpointing = enable
+    
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        # return {'positional_embedding', 'token_embedding'}
+        return {'positional_embedding'}
+
+    def get_num_layers(self):
+        return self.transformer.layers
+
+    def build_attention_mask(self):
+        # lazily create causal attention mask, with full attention between the vision tokens
+        # pytorch uses additive attention mask; fill with -inf
+        mask = torch.empty(self.context_length, self.context_length)
+        mask.fill_(float("-inf"))
+        mask.triu_(1)  # zero out the lower diagonal
+        return mask
+
+    def forward(self, text, return_all_features: bool=False):
+        cast_dtype = self.transformer.get_cast_dtype()
+        x = self.token_embedding(text).to(cast_dtype)  # [batch_size, n_ctx, d_model]
+
+        x = x + self.positional_embedding.to(cast_dtype)
+        x = x.permute(1, 0, 2)  # NLD -> LND
+        x = self.transformer(x, attn_mask=self.attn_mask)
+        # x = self.transformer(x) # no attention mask is applied
+        x = x.permute(1, 0, 2)  # LND -> NLD
+        x = self.ln_final(x)
+
+        if not return_all_features:
+            # x.shape = [batch_size, n_ctx, transformer.width]
+            # take features from the eot embedding (eot_token is the highest number in each sequence)
+            x = x[torch.arange(x.shape[0]), text.argmax(dim=-1)] @ self.text_projection
+        return x

consis_id/models/eva_clip/utils.py

@@ -0,0 +1,326 @@
+from itertools import repeat
+import collections.abc
+import logging
+import math
+import numpy as np
+
+import torch
+from torch import nn as nn
+from torchvision.ops.misc import FrozenBatchNorm2d
+import torch.nn.functional as F
+
+# open CLIP
+def resize_clip_pos_embed(state_dict, model, interpolation: str = 'bicubic', seq_dim=1):
+    # Rescale the grid of position embeddings when loading from state_dict
+    old_pos_embed = state_dict.get('visual.positional_embedding', None)
+    if old_pos_embed is None or not hasattr(model.visual, 'grid_size'):
+        return
+    grid_size = to_2tuple(model.visual.grid_size)
+    extra_tokens = 1  # FIXME detect different token configs (ie no class token, or more)
+    new_seq_len = grid_size[0] * grid_size[1] + extra_tokens
+    if new_seq_len == old_pos_embed.shape[0]:
+        return
+
+    if extra_tokens:
+        pos_emb_tok, pos_emb_img = old_pos_embed[:extra_tokens], old_pos_embed[extra_tokens:]
+    else:
+        pos_emb_tok, pos_emb_img = None, old_pos_embed
+    old_grid_size = to_2tuple(int(math.sqrt(len(pos_emb_img))))
+
+    logging.info('Resizing position embedding grid-size from %s to %s', old_grid_size, grid_size)
+    pos_emb_img = pos_emb_img.reshape(1, old_grid_size[0], old_grid_size[1], -1).permute(0, 3, 1, 2)
+    pos_emb_img = F.interpolate(
+        pos_emb_img,
+        size=grid_size,
+        mode=interpolation,
+        align_corners=True,
+    )
+    pos_emb_img = pos_emb_img.permute(0, 2, 3, 1).reshape(1, grid_size[0] * grid_size[1], -1)[0]
+    if pos_emb_tok is not None:
+        new_pos_embed = torch.cat([pos_emb_tok, pos_emb_img], dim=0)
+    else:
+        new_pos_embed = pos_emb_img
+    state_dict['visual.positional_embedding'] = new_pos_embed
+
+
+def resize_visual_pos_embed(state_dict, model, interpolation: str = 'bicubic', seq_dim=1):
+    # Rescale the grid of position embeddings when loading from state_dict
+    old_pos_embed = state_dict.get('positional_embedding', None)
+    if old_pos_embed is None or not hasattr(model.visual, 'grid_size'):
+        return
+    grid_size = to_2tuple(model.visual.grid_size)
+    extra_tokens = 1  # FIXME detect different token configs (ie no class token, or more)
+    new_seq_len = grid_size[0] * grid_size[1] + extra_tokens
+    if new_seq_len == old_pos_embed.shape[0]:
+        return
+
+    if extra_tokens:
+        pos_emb_tok, pos_emb_img = old_pos_embed[:extra_tokens], old_pos_embed[extra_tokens:]
+    else:
+        pos_emb_tok, pos_emb_img = None, old_pos_embed
+    old_grid_size = to_2tuple(int(math.sqrt(len(pos_emb_img))))
+
+    logging.info('Resizing position embedding grid-size from %s to %s', old_grid_size, grid_size)
+    pos_emb_img = pos_emb_img.reshape(1, old_grid_size[0], old_grid_size[1], -1).permute(0, 3, 1, 2)
+    pos_emb_img = F.interpolate(
+        pos_emb_img,
+        size=grid_size,
+        mode=interpolation,
+        align_corners=True,
+    )
+    pos_emb_img = pos_emb_img.permute(0, 2, 3, 1).reshape(1, grid_size[0] * grid_size[1], -1)[0]
+    if pos_emb_tok is not None:
+        new_pos_embed = torch.cat([pos_emb_tok, pos_emb_img], dim=0)
+    else:
+        new_pos_embed = pos_emb_img
+    state_dict['positional_embedding'] = new_pos_embed
+
+def resize_evaclip_pos_embed(state_dict, model, interpolation: str = 'bicubic', seq_dim=1):
+    all_keys = list(state_dict.keys())
+    # interpolate position embedding
+    if 'visual.pos_embed' in state_dict:
+        pos_embed_checkpoint = state_dict['visual.pos_embed']
+        embedding_size = pos_embed_checkpoint.shape[-1]
+        num_patches = model.visual.patch_embed.num_patches
+        num_extra_tokens = model.visual.pos_embed.shape[-2] - num_patches
+        # height (== width) for the checkpoint position embedding
+        orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
+        # height (== width) for the new position embedding
+        new_size = int(num_patches ** 0.5)
+        # class_token and dist_token are kept unchanged
+        if orig_size != new_size:
+            print("Position interpolate from %dx%d to %dx%d" % (orig_size, orig_size, new_size, new_size))
+            extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
+            # only the position tokens are interpolated
+            pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
+            pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
+            pos_tokens = torch.nn.functional.interpolate(
+                pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
+            pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
+            new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
+            state_dict['visual.pos_embed'] = new_pos_embed
+
+            patch_embed_proj = state_dict['visual.patch_embed.proj.weight']
+            patch_size = model.visual.patch_embed.patch_size
+            state_dict['visual.patch_embed.proj.weight'] = torch.nn.functional.interpolate(
+                patch_embed_proj.float(), size=patch_size, mode='bicubic', align_corners=False)
+
+
+def resize_eva_pos_embed(state_dict, model, interpolation: str = 'bicubic', seq_dim=1):
+    all_keys = list(state_dict.keys())
+    # interpolate position embedding
+    if 'pos_embed' in state_dict:
+        pos_embed_checkpoint = state_dict['pos_embed']
+        embedding_size = pos_embed_checkpoint.shape[-1]
+        num_patches = model.visual.patch_embed.num_patches
+        num_extra_tokens = model.visual.pos_embed.shape[-2] - num_patches
+        # height (== width) for the checkpoint position embedding
+        orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
+        # height (== width) for the new position embedding
+        new_size = int(num_patches ** 0.5)
+        # class_token and dist_token are kept unchanged
+        if orig_size != new_size:
+            print("Position interpolate from %dx%d to %dx%d" % (orig_size, orig_size, new_size, new_size))
+            extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
+            # only the position tokens are interpolated
+            pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
+            pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
+            pos_tokens = torch.nn.functional.interpolate(
+                pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
+            pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
+            new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
+            state_dict['pos_embed'] = new_pos_embed
+
+            patch_embed_proj = state_dict['patch_embed.proj.weight']
+            patch_size = model.visual.patch_embed.patch_size
+            state_dict['patch_embed.proj.weight'] = torch.nn.functional.interpolate(
+                patch_embed_proj.float(), size=patch_size, mode='bicubic', align_corners=False)
+                
+
+def resize_rel_pos_embed(state_dict, model, interpolation: str = 'bicubic', seq_dim=1):
+    all_keys = list(state_dict.keys())
+    for key in all_keys:
+        if "relative_position_index" in key:
+            state_dict.pop(key)
+
+        if "relative_position_bias_table" in key:
+            rel_pos_bias = state_dict[key]
+            src_num_pos, num_attn_heads = rel_pos_bias.size()
+            dst_num_pos, _ = model.visual.state_dict()[key].size()
+            dst_patch_shape = model.visual.patch_embed.patch_shape
+            if dst_patch_shape[0] != dst_patch_shape[1]:
+                raise NotImplementedError()
+            num_extra_tokens = dst_num_pos - (dst_patch_shape[0] * 2 - 1) * (dst_patch_shape[1] * 2 - 1)
+            src_size = int((src_num_pos - num_extra_tokens) ** 0.5)
+            dst_size = int((dst_num_pos - num_extra_tokens) ** 0.5)
+            if src_size != dst_size:
+                print("Position interpolate for %s from %dx%d to %dx%d" % (
+                    key, src_size, src_size, dst_size, dst_size))
+                extra_tokens = rel_pos_bias[-num_extra_tokens:, :]
+                rel_pos_bias = rel_pos_bias[:-num_extra_tokens, :]
+
+                def geometric_progression(a, r, n):
+                    return a * (1.0 - r ** n) / (1.0 - r)
+
+                left, right = 1.01, 1.5
+                while right - left > 1e-6:
+                    q = (left + right) / 2.0
+                    gp = geometric_progression(1, q, src_size // 2)
+                    if gp > dst_size // 2:
+                        right = q
+                    else:
+                        left = q
+
+                # if q > 1.090307:
+                #     q = 1.090307
+
+                dis = []
+                cur = 1
+                for i in range(src_size // 2):
+                    dis.append(cur)
+                    cur += q ** (i + 1)
+
+                r_ids = [-_ for _ in reversed(dis)]
+
+                x = r_ids + [0] + dis
+                y = r_ids + [0] + dis
+
+                t = dst_size // 2.0
+                dx = np.arange(-t, t + 0.1, 1.0)
+                dy = np.arange(-t, t + 0.1, 1.0)
+
+                print("Original positions = %s" % str(x))
+                print("Target positions = %s" % str(dx))
+
+                all_rel_pos_bias = []
+
+                for i in range(num_attn_heads):
+                    z = rel_pos_bias[:, i].view(src_size, src_size).float().numpy()
+                    f = F.interpolate.interp2d(x, y, z, kind='cubic')
+                    all_rel_pos_bias.append(
+                        torch.Tensor(f(dx, dy)).contiguous().view(-1, 1).to(rel_pos_bias.device))
+
+                rel_pos_bias = torch.cat(all_rel_pos_bias, dim=-1)
+
+                new_rel_pos_bias = torch.cat((rel_pos_bias, extra_tokens), dim=0)
+                state_dict[key] = new_rel_pos_bias
+
+    # interpolate position embedding
+    if 'pos_embed' in state_dict:
+        pos_embed_checkpoint = state_dict['pos_embed']
+        embedding_size = pos_embed_checkpoint.shape[-1]
+        num_patches = model.visual.patch_embed.num_patches
+        num_extra_tokens = model.visual.pos_embed.shape[-2] - num_patches
+        # height (== width) for the checkpoint position embedding
+        orig_size = int((pos_embed_checkpoint.shape[-2] - num_extra_tokens) ** 0.5)
+        # height (== width) for the new position embedding
+        new_size = int(num_patches ** 0.5)
+        # class_token and dist_token are kept unchanged
+        if orig_size != new_size:
+            print("Position interpolate from %dx%d to %dx%d" % (orig_size, orig_size, new_size, new_size))
+            extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
+            # only the position tokens are interpolated
+            pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
+            pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size, embedding_size).permute(0, 3, 1, 2)
+            pos_tokens = torch.nn.functional.interpolate(
+                pos_tokens, size=(new_size, new_size), mode='bicubic', align_corners=False)
+            pos_tokens = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
+            new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
+            state_dict['pos_embed'] = new_pos_embed
+
+            patch_embed_proj = state_dict['patch_embed.proj.weight']
+            patch_size = model.visual.patch_embed.patch_size
+            state_dict['patch_embed.proj.weight'] = torch.nn.functional.interpolate(
+                patch_embed_proj.float(), size=patch_size, mode='bicubic', align_corners=False)
+
+
+def freeze_batch_norm_2d(module, module_match={}, name=''):
+    """
+    Converts all `BatchNorm2d` and `SyncBatchNorm` layers of provided module into `FrozenBatchNorm2d`. If `module` is
+    itself an instance of either `BatchNorm2d` or `SyncBatchNorm`, it is converted into `FrozenBatchNorm2d` and
+    returned. Otherwise, the module is walked recursively and submodules are converted in place.
+
+    Args:
+        module (torch.nn.Module): Any PyTorch module.
+        module_match (dict): Dictionary of full module names to freeze (all if empty)
+        name (str): Full module name (prefix)
+
+    Returns:
+        torch.nn.Module: Resulting module
+
+    Inspired by https://github.com/pytorch/pytorch/blob/a5895f85be0f10212791145bfedc0261d364f103/torch/nn/modules/batchnorm.py#L762
+    """
+    res = module
+    is_match = True
+    if module_match:
+        is_match = name in module_match
+    if is_match and isinstance(module, (nn.modules.batchnorm.BatchNorm2d, nn.modules.batchnorm.SyncBatchNorm)):
+        res = FrozenBatchNorm2d(module.num_features)
+        res.num_features = module.num_features
+        res.affine = module.affine
+        if module.affine:
+            res.weight.data = module.weight.data.clone().detach()
+            res.bias.data = module.bias.data.clone().detach()
+        res.running_mean.data = module.running_mean.data
+        res.running_var.data = module.running_var.data
+        res.eps = module.eps
+    else:
+        for child_name, child in module.named_children():
+            full_child_name = '.'.join([name, child_name]) if name else child_name
+            new_child = freeze_batch_norm_2d(child, module_match, full_child_name)
+            if new_child is not child:
+                res.add_module(child_name, new_child)
+    return res
+
+
+# From PyTorch internals
+def _ntuple(n):
+    def parse(x):
+        if isinstance(x, collections.abc.Iterable):
+            return x
+        return tuple(repeat(x, n))
+    return parse
+
+
+to_1tuple = _ntuple(1)
+to_2tuple = _ntuple(2)
+to_3tuple = _ntuple(3)
+to_4tuple = _ntuple(4)
+to_ntuple = lambda n, x: _ntuple(n)(x)
+
+
+def is_logging(args):
+    def is_global_master(args):
+        return args.rank == 0
+
+    def is_local_master(args):
+        return args.local_rank == 0
+
+    def is_master(args, local=False):
+        return is_local_master(args) if local else is_global_master(args)
+    return is_master
+
+
+class AllGather(torch.autograd.Function):
+    """An autograd function that performs allgather on a tensor.
+    Performs all_gather operation on the provided tensors.
+    *** Warning ***: torch.distributed.all_gather has no gradient.
+    """
+
+    @staticmethod
+    def forward(ctx, tensor, rank, world_size):
+        tensors_gather = [torch.empty_like(tensor) for _ in range(world_size)]
+        torch.distributed.all_gather(tensors_gather, tensor)
+        ctx.rank = rank
+        ctx.batch_size = tensor.shape[0]
+        return torch.cat(tensors_gather, 0)
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        return (
+            grad_output[ctx.batch_size * ctx.rank: ctx.batch_size * (ctx.rank + 1)],
+            None,
+            None
+        )
+
+allgather = AllGather.apply

consis_id/models/eva_clip/utils_qformer.py

@@ -0,0 +1,166 @@
+import importlib
+import math
+import os
+import random
+
+import cv2
+import numpy as np
+import torch
+import torch.nn.functional as F
+from torchvision.utils import make_grid
+from transformers import PretrainedConfig
+
+
+def seed_everything(seed):
+    os.environ["PL_GLOBAL_SEED"] = str(seed)
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)
+
+
+def is_torch2_available():
+    return hasattr(F, "scaled_dot_product_attention")
+
+
+def instantiate_from_config(config):
+    if "target" not in config:
+        if config == '__is_first_stage__' or config == "__is_unconditional__":
+            return None
+        raise KeyError("Expected key `target` to instantiate.")
+    return get_obj_from_str(config["target"])(**config.get("params", {}))
+
+
+def get_obj_from_str(string, reload=False):
+    module, cls = string.rsplit(".", 1)
+    if reload:
+        module_imp = importlib.import_module(module)
+        importlib.reload(module_imp)
+    return getattr(importlib.import_module(module, package=None), cls)
+
+
+def drop_seq_token(seq, drop_rate=0.5):
+    idx = torch.randperm(seq.size(1))
+    num_keep_tokens = int(len(idx) * (1 - drop_rate))
+    idx = idx[:num_keep_tokens]
+    seq = seq[:, idx]
+    return seq
+
+
+def import_model_class_from_model_name_or_path(
+    pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
+):
+    text_encoder_config = PretrainedConfig.from_pretrained(
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
+    )
+    model_class = text_encoder_config.architectures[0]
+
+    if model_class == "CLIPTextModel":
+        from transformers import CLIPTextModel
+
+        return CLIPTextModel
+    elif model_class == "CLIPTextModelWithProjection":  # noqa RET505
+        from transformers import CLIPTextModelWithProjection
+
+        return CLIPTextModelWithProjection
+    else:
+        raise ValueError(f"{model_class} is not supported.")
+
+
+def resize_numpy_image_long(image, resize_long_edge=768):
+    h, w = image.shape[:2]
+    if max(h, w) <= resize_long_edge:
+        return image
+    k = resize_long_edge / max(h, w)
+    h = int(h * k)
+    w = int(w * k)
+    image = cv2.resize(image, (w, h), interpolation=cv2.INTER_LANCZOS4)
+    return image
+
+
+# from basicsr
+def img2tensor(imgs, bgr2rgb=True, float32=True):
+    """Numpy array to tensor.
+
+    Args:
+        imgs (list[ndarray] | ndarray): Input images.
+        bgr2rgb (bool): Whether to change bgr to rgb.
+        float32 (bool): Whether to change to float32.
+
+    Returns:
+        list[tensor] | tensor: Tensor images. If returned results only have
+            one element, just return tensor.
+    """
+
+    def _totensor(img, bgr2rgb, float32):
+        if img.shape[2] == 3 and bgr2rgb:
+            if img.dtype == 'float64':
+                img = img.astype('float32')
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+        img = torch.from_numpy(img.transpose(2, 0, 1))
+        if float32:
+            img = img.float()
+        return img
+
+    if isinstance(imgs, list):
+        return [_totensor(img, bgr2rgb, float32) for img in imgs]
+    return _totensor(imgs, bgr2rgb, float32)
+
+
+def tensor2img(tensor, rgb2bgr=True, out_type=np.uint8, min_max=(0, 1)):
+    """Convert torch Tensors into image numpy arrays.
+
+    After clamping to [min, max], values will be normalized to [0, 1].
+
+    Args:
+        tensor (Tensor or list[Tensor]): Accept shapes:
+            1) 4D mini-batch Tensor of shape (B x 3/1 x H x W);
+            2) 3D Tensor of shape (3/1 x H x W);
+            3) 2D Tensor of shape (H x W).
+            Tensor channel should be in RGB order.
+        rgb2bgr (bool): Whether to change rgb to bgr.
+        out_type (numpy type): output types. If ``np.uint8``, transform outputs
+            to uint8 type with range [0, 255]; otherwise, float type with
+            range [0, 1]. Default: ``np.uint8``.
+        min_max (tuple[int]): min and max values for clamp.
+
+    Returns:
+        (Tensor or list): 3D ndarray of shape (H x W x C) OR 2D ndarray of
+        shape (H x W). The channel order is BGR.
+    """
+    if not (torch.is_tensor(tensor) or (isinstance(tensor, list) and all(torch.is_tensor(t) for t in tensor))):
+        raise TypeError(f'tensor or list of tensors expected, got {type(tensor)}')
+
+    if torch.is_tensor(tensor):
+        tensor = [tensor]
+    result = []
+    for _tensor in tensor:
+        _tensor = _tensor.squeeze(0).float().detach().cpu().clamp_(*min_max)
+        _tensor = (_tensor - min_max[0]) / (min_max[1] - min_max[0])
+
+        n_dim = _tensor.dim()
+        if n_dim == 4:
+            img_np = make_grid(_tensor, nrow=int(math.sqrt(_tensor.size(0))), normalize=False).numpy()
+            img_np = img_np.transpose(1, 2, 0)
+            if rgb2bgr:
+                img_np = cv2.cvtColor(img_np, cv2.COLOR_RGB2BGR)
+        elif n_dim == 3:
+            img_np = _tensor.numpy()
+            img_np = img_np.transpose(1, 2, 0)
+            if img_np.shape[2] == 1:  # gray image
+                img_np = np.squeeze(img_np, axis=2)
+            else:
+                if rgb2bgr:
+                    img_np = cv2.cvtColor(img_np, cv2.COLOR_RGB2BGR)
+        elif n_dim == 2:
+            img_np = _tensor.numpy()
+        else:
+            raise TypeError(f'Only support 4D, 3D or 2D tensor. But received with dimension: {n_dim}')
+        if out_type == np.uint8:
+            # Unlike MATLAB, numpy.unit8() WILL NOT round by default.
+            img_np = (img_np * 255.0).round()
+        img_np = img_np.astype(out_type)
+        result.append(img_np)
+    if len(result) == 1:
+        result = result[0]
+    return result

consis_id/models/local_facial_extractor.py

@@ -0,0 +1,269 @@
+import math
+import torch
+import torch.nn as nn
+
+
+# FFN
+def FeedForward(dim, mult=4):
+    inner_dim = int(dim * mult)
+    return nn.Sequential(
+        nn.LayerNorm(dim),
+        nn.Linear(dim, inner_dim, bias=False),
+        nn.GELU(),
+        nn.Linear(inner_dim, dim, bias=False),
+    )
+
+
+def reshape_tensor(x, heads):
+    bs, length, width = x.shape
+    # (bs, length, width) --> (bs, length, n_heads, dim_per_head)
+    x = x.view(bs, length, heads, -1)
+    # (bs, length, n_heads, dim_per_head) --> (bs, n_heads, length, dim_per_head)
+    x = x.transpose(1, 2)
+    # (bs, n_heads, length, dim_per_head) --> (bs*n_heads, length, dim_per_head)
+    x = x.reshape(bs, heads, length, -1)
+    return x
+
+
+class PerceiverAttention(nn.Module):
+    def __init__(self, *, dim, dim_head=64, heads=8, kv_dim=None):
+        super().__init__()
+        self.scale = dim_head ** -0.5
+        self.dim_head = dim_head
+        self.heads = heads
+        inner_dim = dim_head * heads
+
+        self.norm1 = nn.LayerNorm(dim if kv_dim is None else kv_dim)
+        self.norm2 = nn.LayerNorm(dim)
+
+        self.to_q = nn.Linear(dim, inner_dim, bias=False)
+        self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
+        self.to_out = nn.Linear(inner_dim, dim, bias=False)
+
+    def forward(self, x, latents):
+        """
+        Args:
+            x (torch.Tensor): image features
+                shape (b, n1, D)
+            latent (torch.Tensor): latent features
+                shape (b, n2, D)
+        """
+        x = self.norm1(x)
+        latents = self.norm2(latents)
+
+        b, seq_len, _ = latents.shape
+
+        q = self.to_q(latents)
+        kv_input = torch.cat((x, latents), dim=-2)
+        k, v = self.to_kv(kv_input).chunk(2, dim=-1)
+
+        q = reshape_tensor(q, self.heads)
+        k = reshape_tensor(k, self.heads)
+        v = reshape_tensor(v, self.heads)
+
+        # attention
+        scale = 1 / math.sqrt(math.sqrt(self.dim_head))
+        weight = (q * scale) @ (k * scale).transpose(-2, -1)  # More stable with f16 than dividing afterwards
+        weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
+        out = weight @ v
+
+        out = out.permute(0, 2, 1, 3).reshape(b, seq_len, -1)
+
+        return self.to_out(out)
+
+
+class LocalFacialExtractor(nn.Module):
+    def __init__(
+            self,
+            dim=1024,
+            depth=10,
+            dim_head=64,
+            heads=16,
+            num_id_token=5,
+            num_queries=32,
+            output_dim=2048,
+            ff_mult=4,
+    ):
+        """
+        Initializes the LocalFacialExtractor class.
+
+        Parameters:
+        - dim (int): The dimensionality of latent features.
+        - depth (int): Total number of PerceiverAttention and FeedForward layers.
+        - dim_head (int): Dimensionality of each attention head.
+        - heads (int): Number of attention heads.
+        - num_id_token (int): Number of tokens used for identity features.
+        - num_queries (int): Number of query tokens for the latent representation.
+        - output_dim (int): Output dimension after projection.
+        - ff_mult (int): Multiplier for the feed-forward network hidden dimension.
+        """
+        super().__init__()
+
+        # Storing identity token and query information
+        self.num_id_token = num_id_token
+        self.dim = dim
+        self.num_queries = num_queries
+        assert depth % 5 == 0
+        self.depth = depth // 5
+        scale = dim ** -0.5
+
+        # Learnable latent query embeddings
+        self.latents = nn.Parameter(torch.randn(1, num_queries, dim) * scale)
+        # Projection layer to map the latent output to the desired dimension
+        self.proj_out = nn.Parameter(scale * torch.randn(dim, output_dim))
+
+        # Attention and FeedForward layer stack
+        self.layers = nn.ModuleList([])
+        for _ in range(depth):
+            self.layers.append(
+                nn.ModuleList(
+                    [
+                        PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads),  # Perceiver Attention layer
+                        FeedForward(dim=dim, mult=ff_mult),  # FeedForward layer
+                    ]
+                )
+            )
+
+        # Mappings for each of the 5 different ViT features
+        for i in range(5):
+            setattr(
+                self,
+                f'mapping_{i}',
+                nn.Sequential(
+                    nn.Linear(1024, 1024),
+                    nn.LayerNorm(1024),
+                    nn.LeakyReLU(),
+                    nn.Linear(1024, 1024),
+                    nn.LayerNorm(1024),
+                    nn.LeakyReLU(),
+                    nn.Linear(1024, dim),
+                ),
+            )
+
+        # Mapping for identity embedding vectors
+        self.id_embedding_mapping = nn.Sequential(
+            nn.Linear(1280, 1024),
+            nn.LayerNorm(1024),
+            nn.LeakyReLU(),
+            nn.Linear(1024, 1024),
+            nn.LayerNorm(1024),
+            nn.LeakyReLU(),
+            nn.Linear(1024, dim * num_id_token),
+        )
+
+    def forward(self, x, y):
+        """
+        Forward pass for LocalFacialExtractor.
+
+        Parameters:
+        - x (Tensor): The input identity embedding tensor of shape (batch_size, 1280).
+        - y (list of Tensor): A list of 5 visual feature tensors each of shape (batch_size, 1024).
+
+        Returns:
+        - Tensor: The extracted latent features of shape (batch_size, num_queries, output_dim).
+        """
+
+        # Repeat latent queries for the batch size
+        latents = self.latents.repeat(x.size(0), 1, 1)
+
+        # Map the identity embedding to tokens
+        x = self.id_embedding_mapping(x)
+        x = x.reshape(-1, self.num_id_token, self.dim)
+
+        # Concatenate identity tokens with the latent queries
+        latents = torch.cat((latents, x), dim=1)
+
+        # Process each of the 5 visual feature inputs
+        for i in range(5):
+            vit_feature = getattr(self, f'mapping_{i}')(y[i])
+            ctx_feature = torch.cat((x, vit_feature), dim=1)
+
+            # Pass through the PerceiverAttention and FeedForward layers
+            for attn, ff in self.layers[i * self.depth: (i + 1) * self.depth]:
+                latents = attn(ctx_feature, latents) + latents
+                latents = ff(latents) + latents
+
+        # Retain only the query latents
+        latents = latents[:, :self.num_queries]
+        # Project the latents to the output dimension
+        latents = latents @ self.proj_out
+        return latents
+    
+
+class PerceiverCrossAttention(nn.Module):
+    """
+    
+    Args:
+        dim (int): Dimension of the input latent and output. Default is 3072.
+        dim_head (int): Dimension of each attention head. Default is 128.
+        heads (int): Number of attention heads. Default is 16.
+        kv_dim (int): Dimension of the key/value input, allowing flexible cross-attention. Default is 2048.
+    
+    Attributes:
+        scale (float): Scaling factor used in dot-product attention for numerical stability.
+        norm1 (nn.LayerNorm): Layer normalization applied to the input image features.
+        norm2 (nn.LayerNorm): Layer normalization applied to the latent features.
+        to_q (nn.Linear): Linear layer for projecting the latent features into queries.
+        to_kv (nn.Linear): Linear layer for projecting the input features into keys and values.
+        to_out (nn.Linear): Linear layer for outputting the final result after attention.
+
+    """
+    def __init__(self, *, dim=3072, dim_head=128, heads=16, kv_dim=2048):
+        super().__init__()
+        self.scale = dim_head ** -0.5
+        self.dim_head = dim_head
+        self.heads = heads
+        inner_dim = dim_head * heads
+
+        # Layer normalization to stabilize training
+        self.norm1 = nn.LayerNorm(dim if kv_dim is None else kv_dim)
+        self.norm2 = nn.LayerNorm(dim)
+
+        # Linear transformations to produce queries, keys, and values
+        self.to_q = nn.Linear(dim, inner_dim, bias=False)
+        self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
+        self.to_out = nn.Linear(inner_dim, dim, bias=False)
+
+    def forward(self, x, latents):
+        """
+
+        Args:
+            x (torch.Tensor): Input image features with shape (batch_size, n1, D), where:
+                - batch_size (b): Number of samples in the batch.
+                - n1: Sequence length (e.g., number of patches or tokens).
+                - D: Feature dimension.
+            
+            latents (torch.Tensor): Latent feature representations with shape (batch_size, n2, D), where:
+                - n2: Number of latent elements.
+        
+        Returns:
+            torch.Tensor: Attention-modulated features with shape (batch_size, n2, D).
+        
+        """
+        # Apply layer normalization to the input image and latent features
+        x = self.norm1(x)
+        latents = self.norm2(latents)
+
+        b, seq_len, _ = latents.shape
+
+        # Compute queries, keys, and values
+        q = self.to_q(latents)
+        k, v = self.to_kv(x).chunk(2, dim=-1)
+
+        # Reshape tensors to split into attention heads
+        q = reshape_tensor(q, self.heads)
+        k = reshape_tensor(k, self.heads)
+        v = reshape_tensor(v, self.heads)
+
+        # Compute attention weights
+        scale = 1 / math.sqrt(math.sqrt(self.dim_head))
+        weight = (q * scale) @ (k * scale).transpose(-2, -1)  # More stable scaling than post-division
+        weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
+
+        # Compute the output via weighted combination of values
+        out = weight @ v
+
+        # Reshape and permute to prepare for final linear transformation
+        out = out.permute(0, 2, 1, 3).reshape(b, seq_len, -1)
+
+        return self.to_out(out)

consis_id/models/utils.py

@@ -0,0 +1,273 @@
+import cv2
+import math
+import numpy as np
+from PIL import Image
+
+import torch
+from torchvision.transforms import InterpolationMode
+from torchvision.transforms.functional import normalize, resize
+from transformers import T5EncoderModel, T5Tokenizer
+from typing import List, Optional, Tuple, Union
+from diffusers.models.embeddings import get_3d_rotary_pos_embed
+from diffusers.pipelines.cogvideo.pipeline_cogvideox import get_resize_crop_region_for_grid
+
+
+def tensor_to_pil(src_img_tensor):
+    img = src_img_tensor.clone().detach()
+    if img.dtype == torch.bfloat16:
+        img = img.to(torch.float32)
+    img = img.cpu().numpy()
+    img = np.transpose(img, (1, 2, 0))
+    img = img.astype(np.uint8)
+    pil_image = Image.fromarray(img)
+    return pil_image
+    
+
+def _get_t5_prompt_embeds(
+    tokenizer: T5Tokenizer,
+    text_encoder: T5EncoderModel,
+    prompt: Union[str, List[str]],
+    num_videos_per_prompt: int = 1,
+    max_sequence_length: int = 226,
+    device: Optional[torch.device] = None,
+    dtype: Optional[torch.dtype] = None,
+    text_input_ids=None,
+):
+    prompt = [prompt] if isinstance(prompt, str) else prompt
+    batch_size = len(prompt)
+
+    if tokenizer is not None:
+        text_inputs = tokenizer(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            add_special_tokens=True,
+            return_tensors="pt",
+        )
+        text_input_ids = text_inputs.input_ids
+    else:
+        if text_input_ids is None:
+            raise ValueError("`text_input_ids` must be provided when the tokenizer is not specified.")
+
+    prompt_embeds = text_encoder(text_input_ids.to(device))[0]
+    prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+    # duplicate text embeddings for each generation per prompt, using mps friendly method
+    _, seq_len, _ = prompt_embeds.shape
+    prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+    prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+
+    return prompt_embeds
+
+
+def encode_prompt(
+    tokenizer: T5Tokenizer,
+    text_encoder: T5EncoderModel,
+    prompt: Union[str, List[str]],
+    num_videos_per_prompt: int = 1,
+    max_sequence_length: int = 226,
+    device: Optional[torch.device] = None,
+    dtype: Optional[torch.dtype] = None,
+    text_input_ids=None,
+):
+    prompt = [prompt] if isinstance(prompt, str) else prompt
+    prompt_embeds = _get_t5_prompt_embeds(
+        tokenizer,
+        text_encoder,
+        prompt=prompt,
+        num_videos_per_prompt=num_videos_per_prompt,
+        max_sequence_length=max_sequence_length,
+        device=device,
+        dtype=dtype,
+        text_input_ids=text_input_ids,
+    )
+    return prompt_embeds
+
+
+def compute_prompt_embeddings(
+    tokenizer, text_encoder, prompt, max_sequence_length, device, dtype, requires_grad: bool = False
+):
+    if requires_grad:
+        prompt_embeds = encode_prompt(
+            tokenizer,
+            text_encoder,
+            prompt,
+            num_videos_per_prompt=1,
+            max_sequence_length=max_sequence_length,
+            device=device,
+            dtype=dtype,
+        )
+    else:
+        with torch.no_grad():
+            prompt_embeds = encode_prompt(
+                tokenizer,
+                text_encoder,
+                prompt,
+                num_videos_per_prompt=1,
+                max_sequence_length=max_sequence_length,
+                device=device,
+                dtype=dtype,
+            )
+    return prompt_embeds
+
+
+def prepare_rotary_positional_embeddings(
+    height: int,
+    width: int,
+    num_frames: int,
+    vae_scale_factor_spatial: int = 8,
+    patch_size: int = 2,
+    attention_head_dim: int = 64,
+    device: Optional[torch.device] = None,
+    base_height: int = 480,
+    base_width: int = 720,
+) -> Tuple[torch.Tensor, torch.Tensor]:
+    grid_height = height // (vae_scale_factor_spatial * patch_size)
+    grid_width = width // (vae_scale_factor_spatial * patch_size)
+    base_size_width = base_width // (vae_scale_factor_spatial * patch_size)
+    base_size_height = base_height // (vae_scale_factor_spatial * patch_size)
+
+    grid_crops_coords = get_resize_crop_region_for_grid((grid_height, grid_width), base_size_width, base_size_height)
+    freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+        embed_dim=attention_head_dim,
+        crops_coords=grid_crops_coords,
+        grid_size=(grid_height, grid_width),
+        temporal_size=num_frames,
+    )
+
+    freqs_cos = freqs_cos.to(device=device)
+    freqs_sin = freqs_sin.to(device=device)
+    return freqs_cos, freqs_sin
+
+
+def img2tensor(imgs, bgr2rgb=True, float32=True):
+    """Numpy array to tensor.
+
+    Args:
+        imgs (list[ndarray] | ndarray): Input images.
+        bgr2rgb (bool): Whether to change bgr to rgb.
+        float32 (bool): Whether to change to float32.
+
+    Returns:
+        list[tensor] | tensor: Tensor images. If returned results only have
+            one element, just return tensor.
+    """
+
+    def _totensor(img, bgr2rgb, float32):
+        if img.shape[2] == 3 and bgr2rgb:
+            if img.dtype == 'float64':
+                img = img.astype('float32')
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+        img = torch.from_numpy(img.transpose(2, 0, 1))
+        if float32:
+            img = img.float()
+        return img
+
+    if isinstance(imgs, list):
+        return [_totensor(img, bgr2rgb, float32) for img in imgs]
+    return _totensor(imgs, bgr2rgb, float32)
+
+
+def to_gray(img):
+    x = 0.299 * img[:, 0:1] + 0.587 * img[:, 1:2] + 0.114 * img[:, 2:3]
+    x = x.repeat(1, 3, 1, 1)
+    return x
+
+
+def draw_kps(image_pil, kps, color_list=[(255,0,0), (0,255,0), (0,0,255), (255,255,0), (255,0,255)]):
+    stickwidth = 4
+    limbSeq = np.array([[0, 2], [1, 2], [3, 2], [4, 2]])
+    kps = np.array(kps)
+
+    w, h = image_pil.size
+    out_img = np.zeros([h, w, 3])
+
+    for i in range(len(limbSeq)):
+        index = limbSeq[i]
+        color = color_list[index[0]]
+
+        x = kps[index][:, 0]
+        y = kps[index][:, 1]
+        length = ((x[0] - x[1]) ** 2 + (y[0] - y[1]) ** 2) ** 0.5
+        angle = math.degrees(math.atan2(y[0] - y[1], x[0] - x[1]))
+        polygon = cv2.ellipse2Poly((int(np.mean(x)), int(np.mean(y))), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
+        out_img = cv2.fillConvexPoly(out_img.copy(), polygon, color)
+    out_img = (out_img * 0.6).astype(np.uint8)
+
+    for idx_kp, kp in enumerate(kps):
+        color = color_list[idx_kp]
+        x, y = kp
+        out_img = cv2.circle(out_img.copy(), (int(x), int(y)), 10, color, -1)
+
+    out_img_pil = Image.fromarray(out_img.astype(np.uint8))
+    return out_img_pil
+
+
+def process_face_embeddings(face_helper, clip_vision_model, handler_ante, eva_transform_mean, eva_transform_std, app, device, weight_dtype, image, original_id_image=None, is_align_face=True, cal_uncond=False):
+    """
+    Args:
+        image: numpy rgb image, range [0, 255]
+    """
+    face_helper.clean_all()
+    image_bgr = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)  # (724, 502, 3)
+    # get antelopev2 embedding
+    face_info = app.get(image_bgr)
+    if len(face_info) > 0:
+        face_info = sorted(face_info, key=lambda x: (x['bbox'][2] - x['bbox'][0]) * (x['bbox'][3] - x['bbox'][1]))[
+            -1
+        ]  # only use the maximum face
+        id_ante_embedding = face_info['embedding']  # (512,)
+        face_kps = face_info['kps']
+    else:
+        id_ante_embedding = None
+        face_kps = None
+
+    # using facexlib to detect and align face
+    face_helper.read_image(image_bgr)
+    face_helper.get_face_landmarks_5(only_center_face=True)
+    if face_kps is None:
+        face_kps = face_helper.all_landmarks_5[0]
+    face_helper.align_warp_face()
+    if len(face_helper.cropped_faces) == 0:
+        raise RuntimeError('facexlib align face fail')
+    align_face = face_helper.cropped_faces[0]  # (512, 512, 3)  # RGB
+
+    # incase insightface didn't detect face
+    if id_ante_embedding is None:
+        print('fail to detect face using insightface, extract embedding on align face')
+        id_ante_embedding = handler_ante.get_feat(align_face)
+
+    id_ante_embedding = torch.from_numpy(id_ante_embedding).to(device, weight_dtype)  # torch.Size([512])
+    if id_ante_embedding.ndim == 1:
+        id_ante_embedding = id_ante_embedding.unsqueeze(0)  # torch.Size([1, 512])
+
+    # parsing
+    if is_align_face:
+        input = img2tensor(align_face, bgr2rgb=True).unsqueeze(0) / 255.0  # torch.Size([1, 3, 512, 512])
+        input = input.to(device)
+        parsing_out = face_helper.face_parse(normalize(input, [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]))[0]
+        parsing_out = parsing_out.argmax(dim=1, keepdim=True)  # torch.Size([1, 1, 512, 512])
+        bg_label = [0, 16, 18, 7, 8, 9, 14, 15]
+        bg = sum(parsing_out == i for i in bg_label).bool()
+        white_image = torch.ones_like(input)  # torch.Size([1, 3, 512, 512])
+        # only keep the face features
+        return_face_features_image = torch.where(bg, white_image, to_gray(input))  # torch.Size([1, 3, 512, 512])
+        return_face_features_image_2 = torch.where(bg, white_image, input)  # torch.Size([1, 3, 512, 512])
+    else:
+        original_image_bgr = cv2.cvtColor(original_id_image, cv2.COLOR_RGB2BGR)
+        input = img2tensor(original_image_bgr, bgr2rgb=True).unsqueeze(0) / 255.0  # torch.Size([1, 3, 512, 512])
+        input = input.to(device)
+        return_face_features_image = return_face_features_image_2 = input
+
+    # transform img before sending to eva-clip-vit
+    face_features_image = resize(return_face_features_image, clip_vision_model.image_size,
+                                 InterpolationMode.BICUBIC)  # torch.Size([1, 3, 336, 336])
+    face_features_image = normalize(face_features_image, eva_transform_mean, eva_transform_std)
+    id_cond_vit, id_vit_hidden = clip_vision_model(face_features_image.to(weight_dtype), return_all_features=False, return_hidden=True, shuffle=False)  # torch.Size([1, 768]),  list(torch.Size([1, 577, 1024]))
+    id_cond_vit_norm = torch.norm(id_cond_vit, 2, 1, True)
+    id_cond_vit = torch.div(id_cond_vit, id_cond_vit_norm)
+
+    id_cond = torch.cat([id_ante_embedding, id_cond_vit], dim=-1)  # torch.Size([1, 512]), torch.Size([1, 768])  ->  torch.Size([1, 1280])
+
+    return id_cond, id_vit_hidden, return_face_features_image_2, face_kps    # torch.Size([1, 1280]), list(torch.Size([1, 577, 1024]))

custom_cogvideox_transformer_3d.py

@@ -19,6 +19,10 @@ import torch
 from torch import nn
 import torch.nn.functional as F
 
+import os
+import json
+import glob
+
 import numpy as np
 from einops import rearrange
 
@@ -35,6 +39,8 @@ from diffusers.loaders import PeftAdapterMixin
 from diffusers.models.embeddings import apply_rotary_emb
 from .embeddings import CogVideoXPatchEmbed
 
+from .consis_id.models.local_facial_extractor import LocalFacialExtractor, PerceiverCrossAttention
+
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
@@ -408,6 +414,13 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
         use_rotary_positional_embeddings: bool = False,
         use_learned_positional_embeddings: bool = False,
         patch_bias: bool = True,
+        is_train_face: bool = False,
+        is_kps: bool = False,
+        cross_attn_interval: int = 1,
+        LFE_num_tokens: int = 32,
+        LFE_output_dim: int = 768,
+        LFE_heads: int = 12,
+        local_face_scale: float = 1.0,
     ):
         super().__init__()
         inner_dim = num_attention_heads * attention_head_dim
@@ -497,11 +510,101 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
         self.fastercache_device = "cuda"
         self.fastercache_num_blocks_to_cache = len(self.transformer_blocks)
         self.attention_mode = "sdpa"
+
+        if is_train_face:
+            self.inner_dim = inner_dim
+            self.cross_attn_interval = cross_attn_interval
+            self.num_ca = num_layers // cross_attn_interval
+            self.LFE_num_tokens = LFE_num_tokens
+            self.LFE_output_dim = LFE_output_dim
+            self.LFE_heads = LFE_heads
+            self.LFE_final_output_dim = int(self.inner_dim / 3 * 2)
+            self.local_face_scale = local_face_scale
+            self._init_face_inputs()
         
 
     def _set_gradient_checkpointing(self, module, value=False):
         self.gradient_checkpointing = value
 
+    def _init_face_inputs(self):
+        device = self.device
+        weight_dtype = next(self.transformer_blocks.parameters()).dtype
+        self.local_facial_extractor = LocalFacialExtractor()
+        self.local_facial_extractor.to(device, dtype=weight_dtype)
+        self.perceiver_cross_attention = nn.ModuleList([
+            PerceiverCrossAttention(dim=self.inner_dim, dim_head=128, heads=16, kv_dim=self.LFE_final_output_dim).to(device, dtype=weight_dtype) for _ in range(self.num_ca)
+        ])
+    @classmethod
+    def from_pretrained_cus(cls, pretrained_model_path, subfolder=None, config_path=None, transformer_additional_kwargs={}):
+        if subfolder:
+            config_path = config_path or pretrained_model_path
+            config_file = os.path.join(config_path, subfolder, 'config.json')
+            pretrained_model_path = os.path.join(pretrained_model_path, subfolder)
+        else:
+            config_file = os.path.join(config_path or pretrained_model_path, 'config.json')
+
+        print(f"Loading 3D transformer's pretrained weights from {pretrained_model_path} ...")
+
+        # Check if config file exists
+        if not os.path.isfile(config_file):
+            raise RuntimeError(f"Configuration file '{config_file}' does not exist")
+
+        # Load the configuration
+        with open(config_file, "r") as f:
+            config = json.load(f)
+
+        from diffusers.utils import WEIGHTS_NAME
+        model = cls.from_config(config, **transformer_additional_kwargs)
+        model_file = os.path.join(pretrained_model_path, WEIGHTS_NAME)
+        model_file_safetensors = model_file.replace(".bin", ".safetensors")
+        if os.path.exists(model_file):
+            state_dict = torch.load(model_file, map_location="cpu")
+        elif os.path.exists(model_file_safetensors):
+            from safetensors.torch import load_file
+            state_dict = load_file(model_file_safetensors)
+        else:
+            from safetensors.torch import load_file
+            model_files_safetensors = glob.glob(os.path.join(pretrained_model_path, "*.safetensors"))
+            state_dict = {}
+            for model_file_safetensors in model_files_safetensors:
+                _state_dict = load_file(model_file_safetensors)
+                for key in _state_dict:
+                    state_dict[key] = _state_dict[key]
+
+        if model.state_dict()['patch_embed.proj.weight'].size() != state_dict['patch_embed.proj.weight'].size():
+            new_shape   = model.state_dict()['patch_embed.proj.weight'].size()
+            if len(new_shape) == 5:
+                state_dict['patch_embed.proj.weight'] = state_dict['patch_embed.proj.weight'].unsqueeze(2).expand(new_shape).clone()
+                state_dict['patch_embed.proj.weight'][:, :, :-1] = 0
+            else:
+                if model.state_dict()['patch_embed.proj.weight'].size()[1] > state_dict['patch_embed.proj.weight'].size()[1]:
+                    model.state_dict()['patch_embed.proj.weight'][:, :state_dict['patch_embed.proj.weight'].size()[1], :, :] = state_dict['patch_embed.proj.weight']
+                    model.state_dict()['patch_embed.proj.weight'][:, state_dict['patch_embed.proj.weight'].size()[1]:, :, :] = 0
+                    state_dict['patch_embed.proj.weight'] = model.state_dict()['patch_embed.proj.weight']
+                else:
+                    model.state_dict()['patch_embed.proj.weight'][:, :, :, :] = state_dict['patch_embed.proj.weight'][:, :model.state_dict()['patch_embed.proj.weight'].size()[1], :, :]
+                    state_dict['patch_embed.proj.weight'] = model.state_dict()['patch_embed.proj.weight']
+
+        tmp_state_dict = {} 
+        for key in state_dict:
+            if key in model.state_dict().keys() and model.state_dict()[key].size() == state_dict[key].size():
+                tmp_state_dict[key] = state_dict[key]
+            else:
+                print(key, "Size don't match, skip")
+        state_dict = tmp_state_dict
+
+        m, u = model.load_state_dict(state_dict, strict=False)
+        print(f"### missing keys: {len(m)}; \n### unexpected keys: {len(u)};")
+        print(m)
+        
+        params = [p.numel() if "mamba" in n else 0 for n, p in model.named_parameters()]
+        print(f"### Mamba Parameters: {sum(params) / 1e6} M")
+
+        params = [p.numel() if "attn1." in n else 0 for n, p in model.named_parameters()]
+        print(f"### attn1 Parameters: {sum(params) / 1e6} M")
+        
+        return model
+
     @property
     # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors
     def attn_processors(self) -> Dict[str, AttentionProcessor]:
@@ -574,6 +677,7 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
         controlnet_states: torch.Tensor = None,
         controlnet_weights: Optional[Union[float, int, list, np.ndarray, torch.FloatTensor]] = 1.0,
         video_flow_features: Optional[torch.Tensor] = None,
+        consis_id: Optional[dict] = None,
         return_dict: bool = True,
     ):
         batch_size, num_frames, channels, height, width = hidden_states.shape
@@ -608,6 +712,13 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
         encoder_hidden_states = hidden_states[:, :text_seq_length]
         hidden_states = hidden_states[:, text_seq_length:]
         #print("hidden_states after split", hidden_states.shape) #1.5: torch.Size([2, 2700, 3072]) #1.0: torch.Size([2, 5400, 3072])
+
+        # ConsisID: fuse clip and insightface
+        if self.is_train_face:
+            id_cond = consis_id["id_cond"]
+            id_vit_hidden = consis_id["id_vit_hidden"]
+            assert id_cond is not None and id_vit_hidden is not None
+            valid_face_emb = self.local_facial_extractor(id_cond, id_vit_hidden)  # torch.Size([1, 1280]), list[5](torch.Size([1, 577, 1024]))  ->  torch.Size([1, 32, 2048])
       
         if self.use_fastercache:
             self.fastercache_counter+=1
@@ -684,6 +795,7 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
             recovered_uncond = rearrange(recovered_uncond.to(output.dtype), "(B T) C H W -> B T C H W", B=bb, C=cc, T=tt, H=hh, W=ww)
             output = torch.cat([output, recovered_uncond])
         else:
+            ca_idx = 0
             for i, block in enumerate(self.transformer_blocks):
                 hidden_states, encoder_hidden_states = block(
                     hidden_states=hidden_states,
@@ -712,6 +824,12 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
                     elif isinstance(controlnet_weights, (float, int)):
                         controlnet_block_weight = controlnet_weights                    
                     hidden_states = hidden_states + controlnet_states_block * controlnet_block_weight
+
+                # ConsisID
+                if self.is_train_face:
+                    if i % self.cross_attn_interval == 0 and valid_face_emb is not None:
+                        hidden_states = hidden_states + self.local_face_scale * self.perceiver_cross_attention[ca_idx](valid_face_emb, hidden_states)  # torch.Size([2, 32, 2048])  torch.Size([2, 17550, 3072])                        
+                        ca_idx += 1
                     
             if not self.config.use_rotary_positional_embeddings:
                 # CogVideoX-2B
@@ -754,4 +872,4 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
         if not return_dict:
             return (output,)
         return Transformer2DModelOutput(sample=output)
-    
+

examples/cogvideox_consis_id_test.json

@@ -0,0 +1,924 @@
+{
+  "last_node_id": 89,
+  "last_link_id": 184,
+  "nodes": [
+    {
+      "id": 78,
+      "type": "CLIPLoader",
+      "pos": [
+        504,
+        631
+      ],
+      "size": [
+        454.6266784667969,
+        82
+      ],
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "CLIP",
+          "type": "CLIP",
+          "links": [
+            164
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CLIPLoader"
+      },
+      "widgets_values": [
+        "t5\\google_t5-v1_1-xxl_encoderonly-fp8_e4m3fn.safetensors",
+        "sd3"
+      ]
+    },
+    {
+      "id": 82,
+      "type": "CogVideoImageEncode",
+      "pos": [
+        1627,
+        801
+      ],
+      "size": [
+        315,
+        194
+      ],
+      "flags": {},
+      "order": 10,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "vae",
+          "type": "VAE",
+          "link": 170
+        },
+        {
+          "name": "start_image",
+          "type": "IMAGE",
+          "link": 184
+        },
+        {
+          "name": "end_image",
+          "type": "IMAGE",
+          "link": null,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "samples",
+          "type": "LATENT",
+          "links": [
+            172
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CogVideoImageEncode"
+      },
+      "widgets_values": [
+        false,
+        0,
+        1,
+        0,
+        1
+      ]
+    },
+    {
+      "id": 68,
+      "type": "DownloadAndLoadConsisIDModel",
+      "pos": [
+        513,
+        383
+      ],
+      "size": [
+        352.79998779296875,
+        106
+      ],
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "consis_id_model",
+          "type": "CONSISIDMODEL",
+          "links": [
+            152
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "DownloadAndLoadConsisIDModel"
+      },
+      "widgets_values": [
+        "BestWishYsh/ConsisID-preview",
+        "CPU",
+        "bf16"
+      ]
+    },
+    {
+      "id": 69,
+      "type": "ConsisIDFaceEncode",
+      "pos": [
+        983,
+        224
+      ],
+      "size": [
+        393,
+        46
+      ],
+      "flags": {},
+      "order": 7,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "consis_id_model",
+          "type": "CONSISIDMODEL",
+          "link": 152
+        },
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 156
+        }
+      ],
+      "outputs": [
+        {
+          "name": "consis_id_conds",
+          "type": "CONSISID_CONDS",
+          "links": [
+            157
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "face_image",
+          "type": "IMAGE",
+          "links": [
+            174,
+            176
+          ],
+          "slot_index": 1
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ConsisIDFaceEncode"
+      },
+      "widgets_values": []
+    },
+    {
+      "id": 75,
+      "type": "DownloadAndLoadCogVideoModel",
+      "pos": [
+        1327,
+        -133
+      ],
+      "size": [
+        315,
+        218
+      ],
+      "flags": {},
+      "order": 2,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "block_edit",
+          "type": "TRANSFORMERBLOCKS",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "lora",
+          "type": "COGLORA",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "compile_args",
+          "type": "COMPILEARGS",
+          "link": null,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "model",
+          "type": "COGVIDEOMODEL",
+          "links": [
+            159
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "vae",
+          "type": "VAE",
+          "links": [
+            170,
+            181
+          ],
+          "slot_index": 1
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "DownloadAndLoadCogVideoModel"
+      },
+      "widgets_values": [
+        "BestWishYsh/ConsisID-preview",
+        "bf16",
+        "disabled",
+        false,
+        "sageattn",
+        "main_device"
+      ]
+    },
+    {
+      "id": 83,
+      "type": "ImageResizeKJ",
+      "pos": [
+        765,
+        803
+      ],
+      "size": [
+        315,
+        266
+      ],
+      "flags": {},
+      "order": 9,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 176
+        },
+        {
+          "name": "get_image_size",
+          "type": "IMAGE",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "width_input",
+          "type": "INT",
+          "link": null,
+          "widget": {
+            "name": "width_input"
+          },
+          "shape": 7
+        },
+        {
+          "name": "height_input",
+          "type": "INT",
+          "link": null,
+          "widget": {
+            "name": "height_input"
+          },
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            184
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "width",
+          "type": "INT",
+          "links": null
+        },
+        {
+          "name": "height",
+          "type": "INT",
+          "links": null
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ImageResizeKJ"
+      },
+      "widgets_values": [
+        720,
+        480,
+        "nearest-exact",
+        false,
+        2,
+        0,
+        0,
+        "disabled"
+      ]
+    },
+    {
+      "id": 81,
+      "type": "CogVideoTextEncode",
+      "pos": [
+        1125,
+        658
+      ],
+      "size": [
+        400,
+        200
+      ],
+      "flags": {},
+      "order": 6,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "clip",
+          "type": "CLIP",
+          "link": 168
+        }
+      ],
+      "outputs": [
+        {
+          "name": "conditioning",
+          "type": "CONDITIONING",
+          "links": [
+            169
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "clip",
+          "type": "CLIP",
+          "links": [],
+          "slot_index": 1
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CogVideoTextEncode"
+      },
+      "widgets_values": [
+        "bad quality, blurry",
+        1,
+        true
+      ]
+    },
+    {
+      "id": 87,
+      "type": "CogVideoDecode",
+      "pos": [
+        2207,
+        -108
+      ],
+      "size": [
+        315,
+        198
+      ],
+      "flags": {},
+      "order": 12,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "vae",
+          "type": "VAE",
+          "link": 181
+        },
+        {
+          "name": "samples",
+          "type": "LATENT",
+          "link": 182
+        }
+      ],
+      "outputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "links": [
+            183
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CogVideoDecode"
+      },
+      "widgets_values": [
+        true,
+        240,
+        360,
+        0.2,
+        0.2,
+        true
+      ]
+    },
+    {
+      "id": 72,
+      "type": "ImageResizeKJ",
+      "pos": [
+        584,
+        -223
+      ],
+      "size": [
+        315,
+        266
+      ],
+      "flags": {},
+      "order": 5,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 155
+        },
+        {
+          "name": "get_image_size",
+          "type": "IMAGE",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "width_input",
+          "type": "INT",
+          "link": null,
+          "widget": {
+            "name": "width_input"
+          },
+          "shape": 7
+        },
+        {
+          "name": "height_input",
+          "type": "INT",
+          "link": null,
+          "widget": {
+            "name": "height_input"
+          },
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            156
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "width",
+          "type": "INT",
+          "links": null
+        },
+        {
+          "name": "height",
+          "type": "INT",
+          "links": null
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ImageResizeKJ"
+      },
+      "widgets_values": [
+        512,
+        512,
+        "lanczos",
+        false,
+        2,
+        0,
+        0,
+        "center"
+      ]
+    },
+    {
+      "id": 84,
+      "type": "PreviewImage",
+      "pos": [
+        1011,
+        -100
+      ],
+      "size": [
+        210,
+        246
+      ],
+      "flags": {},
+      "order": 8,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 174
+        }
+      ],
+      "outputs": [],
+      "properties": {
+        "Node name for S&R": "PreviewImage"
+      }
+    },
+    {
+      "id": 80,
+      "type": "CogVideoTextEncode",
+      "pos": [
+        1109,
+        360
+      ],
+      "size": [
+        442.14453125,
+        234.34109497070312
+      ],
+      "flags": {},
+      "order": 4,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "clip",
+          "type": "CLIP",
+          "link": 164
+        }
+      ],
+      "outputs": [
+        {
+          "name": "conditioning",
+          "type": "CONDITIONING",
+          "links": [
+            167
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "clip",
+          "type": "CLIP",
+          "links": [
+            168
+          ],
+          "slot_index": 1
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CogVideoTextEncode"
+      },
+      "widgets_values": [
+        "A woman adorned with a delicate flower crown, is standing amidst a field of gently swaying wildflowers. Her eyes sparkle with a serene gaze, and a faint smile graces her lips, suggesting a moment of peaceful contentment. The shot is framed from the waist up, highlighting the gentle breeze lightly tousling her hair. The background reveals an expansive meadow under a bright blue sky, capturing the tranquility of a sunny afternoon.",
+        1,
+        false
+      ]
+    },
+    {
+      "id": 71,
+      "type": "LoadImage",
+      "pos": [
+        213,
+        -225
+      ],
+      "size": [
+        315,
+        314
+      ],
+      "flags": {},
+      "order": 3,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            155
+          ],
+          "slot_index": 0
+        },
+        {
+          "name": "MASK",
+          "type": "MASK",
+          "links": null
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "LoadImage"
+      },
+      "widgets_values": [
+        "2 (2).jpg",
+        "image"
+      ]
+    },
+    {
+      "id": 88,
+      "type": "VHS_VideoCombine",
+      "pos": [
+        2726,
+        -72
+      ],
+      "size": [
+        632.8847045898438,
+        732.5897827148438
+      ],
+      "flags": {},
+      "order": 13,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "images",
+          "type": "IMAGE",
+          "link": 183
+        },
+        {
+          "name": "audio",
+          "type": "AUDIO",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "meta_batch",
+          "type": "VHS_BatchManager",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "vae",
+          "type": "VAE",
+          "link": null,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "Filenames",
+          "type": "VHS_FILENAMES",
+          "links": null
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "VHS_VideoCombine"
+      },
+      "widgets_values": {
+        "frame_rate": 8,
+        "loop_count": 0,
+        "filename_prefix": "ConsisID",
+        "format": "video/h264-mp4",
+        "pix_fmt": "yuv420p",
+        "crf": 19,
+        "save_metadata": true,
+        "pingpong": false,
+        "save_output": false,
+        "videopreview": {
+          "hidden": false,
+          "paused": false,
+          "params": {
+            "filename": "ConsisID_00006.mp4",
+            "subfolder": "",
+            "type": "temp",
+            "format": "video/h264-mp4",
+            "frame_rate": 8
+          },
+          "muted": false
+        }
+      }
+    },
+    {
+      "id": 73,
+      "type": "CogVideoSampler",
+      "pos": [
+        1753,
+        -1
+      ],
+      "size": [
+        330,
+        594
+      ],
+      "flags": {},
+      "order": 11,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "model",
+          "type": "COGVIDEOMODEL",
+          "link": 159
+        },
+        {
+          "name": "positive",
+          "type": "CONDITIONING",
+          "link": 167
+        },
+        {
+          "name": "negative",
+          "type": "CONDITIONING",
+          "link": 169
+        },
+        {
+          "name": "samples",
+          "type": "LATENT",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "image_cond_latents",
+          "type": "LATENT",
+          "link": 172,
+          "shape": 7
+        },
+        {
+          "name": "context_options",
+          "type": "COGCONTEXT",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "controlnet",
+          "type": "COGVIDECONTROLNET",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "tora_trajectory",
+          "type": "TORAFEATURES",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "fastercache",
+          "type": "FASTERCACHEARGS",
+          "link": null,
+          "shape": 7
+        },
+        {
+          "name": "consis_id_cond",
+          "type": "CONSISID_CONDS",
+          "link": 157,
+          "shape": 7
+        }
+      ],
+      "outputs": [
+        {
+          "name": "samples",
+          "type": "LATENT",
+          "links": [
+            182
+          ],
+          "slot_index": 0
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "CogVideoSampler"
+      },
+      "widgets_values": [
+        49,
+        40,
+        6,
+        3,
+        "fixed",
+        "CogVideoXDPMScheduler",
+        1
+      ]
+    }
+  ],
+  "links": [
+    [
+      152,
+      68,
+      0,
+      69,
+      0,
+      "CONSISIDMODEL"
+    ],
+    [
+      155,
+      71,
+      0,
+      72,
+      0,
+      "IMAGE"
+    ],
+    [
+      156,
+      72,
+      0,
+      69,
+      1,
+      "IMAGE"
+    ],
+    [
+      157,
+      69,
+      0,
+      73,
+      9,
+      "CONSISID_CONDS"
+    ],
+    [
+      159,
+      75,
+      0,
+      73,
+      0,
+      "COGVIDEOMODEL"
+    ],
+    [
+      164,
+      78,
+      0,
+      80,
+      0,
+      "CLIP"
+    ],
+    [
+      167,
+      80,
+      0,
+      73,
+      1,
+      "CONDITIONING"
+    ],
+    [
+      168,
+      80,
+      1,
+      81,
+      0,
+      "CLIP"
+    ],
+    [
+      169,
+      81,
+      0,
+      73,
+      2,
+      "CONDITIONING"
+    ],
+    [
+      170,
+      75,
+      1,
+      82,
+      0,
+      "VAE"
+    ],
+    [
+      172,
+      82,
+      0,
+      73,
+      4,
+      "LATENT"
+    ],
+    [
+      174,
+      69,
+      1,
+      84,
+      0,
+      "IMAGE"
+    ],
+    [
+      176,
+      69,
+      1,
+      83,
+      0,
+      "IMAGE"
+    ],
+    [
+      181,
+      75,
+      1,
+      87,
+      0,
+      "VAE"
+    ],
+    [
+      182,
+      73,
+      0,
+      87,
+      1,
+      "LATENT"
+    ],
+    [
+      183,
+      87,
+      0,
+      88,
+      0,
+      "IMAGE"
+    ],
+    [
+      184,
+      83,
+      0,
+      82,
+      1,
+      "IMAGE"
+    ]
+  ],
+  "groups": [],
+  "config": {},
+  "extra": {
+    "ds": {
+      "scale": 0.7513148009015777,
+      "offset": [
+        -62.54456298631729,
+        363.8127925482066
+      ]
+    }
+  },
+  "version": 0.4
+}

model_loading.py

@@ -110,7 +110,8 @@ class DownloadAndLoadCogVideoModel:
                         "alibaba-pai/CogVideoX-Fun-V1.1-5b-Pose",
                         "alibaba-pai/CogVideoX-Fun-V1.1-5b-Control",
                         "feizhengcong/CogvideoX-Interpolation",
-                        "NimVideo/cogvideox-2b-img2vid"
+                        "NimVideo/cogvideox-2b-img2vid",
+                        "BestWishYsh/ConsisID-preview",
                     ],
                 ),
 
@@ -199,7 +200,7 @@ class DownloadAndLoadCogVideoModel:
             download_path = base_path
             repo_id = model
             subfolder = "transformer"
-            allow_patterns = ["*transformer*", "*scheduler*", "*vae*"]
+            allow_patterns = ["*transformer*", "*scheduler*", "*vae*", "*face_encoder*"]
 
         if "2b" in model:
             scheduler_path = os.path.join(script_directory, 'configs', 'scheduler_config_2b.json')
@@ -217,8 +218,24 @@ class DownloadAndLoadCogVideoModel:
                 local_dir=download_path,
                 local_dir_use_symlinks=False,
             )
-
-        transformer = CogVideoXTransformer3DModel.from_pretrained(base_path, subfolder=subfolder)
+        # transformer_additional_kwargs={}
+        if "consisid" in model.lower():
+            # transformer_additional_kwargs={
+            #     'torch_dtype': dtype,
+            #     'revision': None,
+            #     'variant': None,
+            #     'is_train_face': True,
+            #     'is_kps': False,
+            #     'LFE_num_tokens': 32,
+            #     'LFE_output_dim': 768,
+            #     'LFE_heads': 12,
+            #     'cross_attn_interval': 2, 
+            # }
+            transformer = CogVideoXTransformer3DModel.from_pretrained_cus(base_path, subfolder=subfolder)
+        else:
+            transformer = CogVideoXTransformer3DModel.from_pretrained(base_path, subfolder=subfolder)
+        
+        
         transformer = transformer.to(dtype).to(transformer_load_device)
 
         if "1.5" in model:

nodes.py

@@ -592,6 +592,7 @@ class CogVideoSampler:
                 "controlnet": ("COGVIDECONTROLNET",),
                 "tora_trajectory": ("TORAFEATURES", ),
                 "fastercache": ("FASTERCACHEARGS", ),
+                "consis_id_cond": ("CONSISID_CONDS", {"tooltip": "ConsisID conditioning"} ),
             }
         }
 
@@ -601,7 +602,7 @@ class CogVideoSampler:
     CATEGORY = "CogVideoWrapper"
 
     def process(self, model, positive, negative, steps, cfg, seed, scheduler, num_frames, samples=None,
-                denoise_strength=1.0, image_cond_latents=None, context_options=None, controlnet=None, tora_trajectory=None, fastercache=None):
+                denoise_strength=1.0, image_cond_latents=None, context_options=None, controlnet=None, tora_trajectory=None, fastercache=None, consis_id_cond=None):
         mm.unload_all_models()
         mm.soft_empty_cache()
 
@@ -610,7 +611,8 @@ class CogVideoSampler:
         "I2V" in model_name or 
         "interpolation" in model_name.lower() or 
         "fun" in model_name.lower() or
-        "img2vid" in model_name.lower()
+        "img2vid" in model_name.lower() or
+        "consisid" in model_name.lower()
         ) else False
         if "fun" in model_name.lower() and not ("pose" in model_name.lower() or "control" in model_name.lower()) and image_cond_latents is not None:
             assert image_cond_latents["mask"] is not None, "For fun inpaint models use CogVideoImageEncodeFunInP"
@@ -722,6 +724,7 @@ class CogVideoSampler:
                 tora=tora_trajectory if tora_trajectory is not None else None,
                 image_cond_start_percent=image_cond_start_percent if image_cond_latents is not None else 0.0,
                 image_cond_end_percent=image_cond_end_percent if image_cond_latents is not None else 1.0,
+                consis_id=consis_id_cond,
             )
         if not model["cpu_offloading"] and model["manual_offloading"]:
             pipe.transformer.to(offload_device)

nodes_consis_id.py

@@ -0,0 +1,148 @@
+import os
+import json
+import torch
+
+import folder_paths
+import comfy.model_management as mm
+
+class DownloadAndLoadConsisIDModel:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "model": (["BestWishYsh/ConsisID-preview",],),
+                 "onnx_device": (
+                    ['CPU', 'CUDA', 'ROCM', 'CoreML'], {
+                        "default": 'CPU'
+                    }), 
+                "precision": (["fp16", "fp32", "bf16"],
+                        {"default": "bf16", "tooltip": "official recommendation is that 2b model should be fp16, 5b model should be bf16"}),
+            },
+        }
+
+    RETURN_TYPES = ("CONSISIDMODEL", )
+    RETURN_NAMES = ("consis_id_model", )
+    FUNCTION = "loadmodel"
+    CATEGORY = "CogVideoWrapper"
+    DESCRIPTION = "Downloads and loads the selected CogVideo model from Huggingface to 'ComfyUI/models/CogVideo'"
+
+    def loadmodel(self, model, precision, onnx_device):
+        import insightface
+        from insightface.app import FaceAnalysis
+        from facexlib.parsing import init_parsing_model
+        from facexlib.utils.face_restoration_helper import FaceRestoreHelper
+        from .consis_id.models.eva_clip import create_model_and_transforms
+        from .consis_id.models.eva_clip.constants import OPENAI_DATASET_MEAN, OPENAI_DATASET_STD
+        
+        device = mm.get_torch_device()
+        offload_device = mm.unet_offload_device()
+        dtype = {"bf16": torch.bfloat16, "fp16": torch.float16, "fp32": torch.float32}[precision]
+        base_path = folder_paths.get_folder_paths("CogVideo")[0]
+        model_path = os.path.join(base_path, "ConsisID-preview")
+        face_encoder_path = os.path.join(model_path, "face_encoder")
+        
+        # 1. load face helper models
+        face_helper = FaceRestoreHelper(
+            upscale_factor=1,
+            face_size=512,
+            crop_ratio=(1, 1),
+            det_model='retinaface_resnet50',
+            save_ext='png',
+            device=device,
+            model_rootpath=model_path
+        )
+        face_helper.face_parse = None
+        face_helper.face_parse = init_parsing_model(model_name='bisenet', device=device, model_rootpath=model_path)
+        face_helper.face_det.eval()
+        face_helper.face_parse.eval()
+
+        model, _, _ = create_model_and_transforms('EVA02-CLIP-L-14-336', os.path.join(face_encoder_path, "EVA02_CLIP_L_336_psz14_s6B.pt"), force_custom_clip=True)
+        face_clip_model = model.visual
+        face_clip_model.eval()
+
+        eva_transform_mean = getattr(face_clip_model, 'image_mean', OPENAI_DATASET_MEAN)
+        eva_transform_std = getattr(face_clip_model, 'image_std', OPENAI_DATASET_STD)
+        if not isinstance(eva_transform_mean, (list, tuple)):
+            eva_transform_mean = (eva_transform_mean,) * 3
+        if not isinstance(eva_transform_std, (list, tuple)):
+            eva_transform_std = (eva_transform_std,) * 3
+        eva_transform_mean = eva_transform_mean
+        eva_transform_std = eva_transform_std
+
+        face_main_model = FaceAnalysis(name='antelopev2', root=face_encoder_path, providers=[onnx_device + 'ExecutionProvider',])
+        handler_ante = insightface.model_zoo.get_model(f'{face_encoder_path}/models/antelopev2/glintr100.onnx', providers=[onnx_device + 'ExecutionProvider',])
+        face_main_model.prepare(ctx_id=0, det_size=(640, 640))
+        handler_ante.prepare(ctx_id=0)
+            
+        face_clip_model.to(device, dtype=dtype)
+        face_helper.face_det.to(device)
+        face_helper.face_parse.to(device)
+        
+        mm.soft_empty_cache()
+
+        dtype = {"bf16": torch.bfloat16, "fp16": torch.float16, "fp32": torch.float32}[precision]
+        consis_id_model = {
+            "face_helper": face_helper,
+            "face_clip_model": face_clip_model,
+            "handler_ante": handler_ante,
+            "eva_transform_mean": eva_transform_mean,
+            "eva_transform_std": eva_transform_std,
+            "face_main_model": face_main_model,
+            "dtype": dtype,
+        }
+
+        return consis_id_model,
+
+class ConsisIDFaceEncode:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "consis_id_model": ("CONSISIDMODEL",),
+                "image": ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("CONSISID_CONDS", "IMAGE",)
+    RETURN_NAMES = ("consis_id_conds", "face_image", )
+    FUNCTION = "faceencode"
+    CATEGORY = "CogVideoWrapper"
+    DESCRIPTION = "Downloads and loads the selected CogVideo model from Huggingface to 'ComfyUI/models/CogVideo'"
+
+    def faceencode(self, image, consis_id_model):
+        from .consis_id.models.utils import process_face_embeddings
+
+        device = mm.get_torch_device()
+        dtype = consis_id_model["dtype"]
+        
+        id_image = image[0].cpu().numpy() * 255
+
+        face_helper = consis_id_model["face_helper"]
+        face_clip_model = consis_id_model["face_clip_model"]
+        handler_ante = consis_id_model["handler_ante"]
+        eva_transform_mean = consis_id_model["eva_transform_mean"]
+        eva_transform_std = consis_id_model["eva_transform_std"]
+        face_main_model = consis_id_model["face_main_model"]
+        id_cond, id_vit_hidden, align_crop_face_image, face_kps = process_face_embeddings(face_helper, face_clip_model, handler_ante, 
+                                                                                eva_transform_mean, eva_transform_std, 
+                                                                                face_main_model, device, dtype, id_image, 
+                                                                                original_id_image=id_image, is_align_face=True, 
+                                                                                cal_uncond=False)
+        consis_id_conds = {
+            "id_cond": id_cond,
+            "id_vit_hidden": id_vit_hidden,
+            #"align_crop_face_image": align_crop_face_image,
+            #"face_kps": face_kps
+        }
+        print(align_crop_face_image.shape)
+        align_crop_face_image = align_crop_face_image.permute(0, 2, 3, 1).float().cpu()
+        return consis_id_conds, align_crop_face_image,
+    
+NODE_CLASS_MAPPINGS = {
+    "DownloadAndLoadConsisIDModel": DownloadAndLoadConsisIDModel,
+    "ConsisIDFaceEncode": ConsisIDFaceEncode,
+}
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "DownloadAndLoadConsisIDModel": "DownloadAndLoadConsisIDModel",
+    "ConsisIDFaceEncode": "ConsisID FaceEncode",
+    }

pipeline_cogvideox.py

@@ -351,6 +351,7 @@ class CogVideoXPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
         tora: Optional[dict] = None,
         image_cond_start_percent: float = 0.0,
         image_cond_end_percent: float = 1.0,
+        consis_id: Optional[dict] = None,
         
     ):
         """
@@ -762,6 +763,7 @@ class CogVideoXPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
                         controlnet_states=controlnet_states,
                         controlnet_weights=control_weights,
                         video_flow_features=video_flow_features if (tora is not None and tora["start_percent"] <= current_step_percentage <= tora["end_percent"]) else None,
+                        consis_id=consis_id,
                     )[0]
                     noise_pred = noise_pred.float()
                     if isinstance(self.scheduler, CogVideoXDPMScheduler):
@@ -798,5 +800,5 @@ class CogVideoXPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
 
         # Offload all models
         self.maybe_free_model_hooks()
-
+        print("latents shape", latents.shape)
         return latents