Upload folder using huggingface_hub

config.json

@@ -0,0 +1,30 @@
+{
+  "architectures": [
+    "InternVisionModel"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_intern_vit.InternVisionConfig",
+    "AutoModel": "modeling_intern_vit.InternVisionModel"
+  },
+  "attention_dropout": 0.0,
+  "drop_path_rate": 0.0,
+  "dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_size": 3200,
+  "image_size": 448,
+  "initializer_factor": 0.1,
+  "initializer_range": 1e-10,
+  "intermediate_size": 12800,
+  "layer_norm_eps": 1e-06,
+  "model_type": "intern_vit_6b",
+  "num_attention_heads": 25,
+  "num_channels": 3,
+  "num_hidden_layers": 45,
+  "patch_size": 14,
+  "qk_normalization": true,
+  "qkv_bias": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.36.2",
+  "use_bfloat16": true,
+  "use_flash_attn": true
+}

configuration_intern_vit.py

@@ -0,0 +1,117 @@
+# --------------------------------------------------------
+# InternVL
+# Copyright (c) 2023 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+import os
+from typing import Union
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class InternVisionConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`InternVisionModel`]. It is used to
+    instantiate a vision encoder according to the specified arguments, defining the model architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        num_channels (`int`, *optional*, defaults to 3):
+            Number of color channels in the input images (e.g., 3 for RGB).
+        patch_size (`int`, *optional*, defaults to 14):
+            The size (resolution) of each patch.
+        image_size (`int`, *optional*, defaults to 224):
+            The size (resolution) of each image.
+        qkv_bias (`bool`, *optional*, defaults to `False`):
+            Whether to add a bias to the queries and values in the self-attention layers.
+        hidden_size (`int`, *optional*, defaults to 3200):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_attention_heads (`int`, *optional*, defaults to 25):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        intermediate_size (`int`, *optional*, defaults to 12800):
+            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        qk_normalization (`bool`, *optional*, defaults to `True`):
+            Whether to normalize the queries and keys in the self-attention layers.
+        num_hidden_layers (`int`, *optional*, defaults to 48):
+            Number of hidden layers in the Transformer encoder.
+        use_flash_attn (`bool`, *optional*, defaults to `True`):
+            Whether to use flash attention mechanism.
+        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"selu"` and `"gelu_new"` ``"gelu"` are supported.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-6):
+            The epsilon used by the layer normalization layers.
+        dropout (`float`, *optional*, defaults to 0.0):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        drop_path_rate (`float`, *optional*, defaults to 0.0):
+            Dropout rate for stochastic depth.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        initializer_factor (`float`, *optional*, defaults to 0.1):
+            A factor for layer scale.
+    """
+
+    model_type = 'intern_vit_6b'
+
+    def __init__(
+            self,
+            num_channels=3,
+            patch_size=14,
+            image_size=224,
+            qkv_bias=False,
+            hidden_size=3200,
+            num_attention_heads=25,
+            intermediate_size=12800,
+            qk_normalization=True,
+            num_hidden_layers=48,
+            use_flash_attn=True,
+            hidden_act='gelu',
+            layer_norm_eps=1e-6,
+            dropout=0.0,
+            drop_path_rate=0.0,
+            attention_dropout=0.0,
+            initializer_range=0.02,
+            initializer_factor=0.1,
+            **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.dropout = dropout
+        self.drop_path_rate = drop_path_rate
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_channels = num_channels
+        self.patch_size = patch_size
+        self.image_size = image_size
+        self.initializer_range = initializer_range
+        self.initializer_factor = initializer_factor
+        self.attention_dropout = attention_dropout
+        self.layer_norm_eps = layer_norm_eps
+        self.hidden_act = hidden_act
+        self.qkv_bias = qkv_bias
+        self.qk_normalization = qk_normalization
+        self.use_flash_attn = use_flash_attn
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> 'PretrainedConfig':
+        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
+
+        if 'vision_config' in config_dict:
+            config_dict = config_dict['vision_config']
+
+        if 'model_type' in config_dict and hasattr(cls, 'model_type') and config_dict['model_type'] != cls.model_type:
+            logger.warning(
+                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
+                f'{cls.model_type}. This is not supported for all configurations of models and can yield errors.'
+            )
+
+        return cls.from_dict(config_dict, **kwargs)

flash_attention.py

@@ -0,0 +1,75 @@
+import torch
+import torch.nn as nn
+from einops import rearrange
+
+try:  # v1
+    from flash_attn.flash_attn_interface import \
+        flash_attn_unpadded_qkvpacked_func
+except:  # v2
+    from flash_attn.flash_attn_interface import flash_attn_varlen_qkvpacked_func as flash_attn_unpadded_qkvpacked_func
+
+from flash_attn.bert_padding import pad_input, unpad_input
+
+
+class FlashAttention(nn.Module):
+    """Implement the scaled dot product attention with softmax.
+    Arguments
+    ---------
+        softmax_scale: The temperature to use for the softmax attention.
+                      (default: 1/sqrt(d_keys) where d_keys is computed at
+                      runtime)
+        attention_dropout: The dropout rate to apply to the attention
+                           (default: 0.0)
+    """
+
+    def __init__(self, softmax_scale=None, attention_dropout=0.0, device=None, dtype=None):
+        super().__init__()
+        self.softmax_scale = softmax_scale
+        self.dropout_p = attention_dropout
+
+    def forward(self, qkv, key_padding_mask=None, causal=False, cu_seqlens=None,
+                max_s=None, need_weights=False):
+        """Implements the multihead softmax attention.
+        Arguments
+        ---------
+            qkv: The tensor containing the query, key, and value. (B, S, 3, H, D) if key_padding_mask is None
+                if unpadded: (nnz, 3, h, d)
+            key_padding_mask: a bool tensor of shape (B, S)
+        """
+        assert not need_weights
+        assert qkv.dtype in [torch.float16, torch.bfloat16]
+        assert qkv.is_cuda
+
+        if cu_seqlens is None:
+            batch_size = qkv.shape[0]
+            seqlen = qkv.shape[1]
+            if key_padding_mask is None:
+                qkv = rearrange(qkv, 'b s ... -> (b s) ...')
+                max_s = seqlen
+                cu_seqlens = torch.arange(0, (batch_size + 1) * seqlen, step=seqlen, dtype=torch.int32,
+                                          device=qkv.device)
+                output = flash_attn_unpadded_qkvpacked_func(
+                    qkv, cu_seqlens, max_s, self.dropout_p if self.training else 0.0,
+                    softmax_scale=self.softmax_scale, causal=causal
+                )
+                output = rearrange(output, '(b s) ... -> b s ...', b=batch_size)
+            else:
+                nheads = qkv.shape[-2]
+                x = rearrange(qkv, 'b s three h d -> b s (three h d)')
+                x_unpad, indices, cu_seqlens, max_s = unpad_input(x, key_padding_mask)
+                x_unpad = rearrange(x_unpad, 'nnz (three h d) -> nnz three h d', three=3, h=nheads)
+                output_unpad = flash_attn_unpadded_qkvpacked_func(
+                    x_unpad, cu_seqlens, max_s, self.dropout_p if self.training else 0.0,
+                    softmax_scale=self.softmax_scale, causal=causal
+                )
+                output = rearrange(pad_input(rearrange(output_unpad, 'nnz h d -> nnz (h d)'),
+                                             indices, batch_size, seqlen),
+                                   'b s (h d) -> b s h d', h=nheads)
+        else:
+            assert max_s is not None
+            output = flash_attn_unpadded_qkvpacked_func(
+                qkv, cu_seqlens, max_s, self.dropout_p if self.training else 0.0,
+                softmax_scale=self.softmax_scale, causal=causal
+            )
+
+        return output, None

model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b69de7960bd05ff0abaeb765783a439f5fbcfe80ce643c570fe08587fda66f4
+size 4988565944

model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8434954fb6944f58fd23356265239db07abde8b6016eb5f35af0bcaa9405eaa9
+size 4937250176

model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2fa4a1cb965066571ef70ba95956ad935c0ac4adff48f777e8c567e58c5443b
+size 1147238088

model.safetensors.index.json

@@ -0,0 +1,596 @@
+{
+  "metadata": {
+    "total_size": 11072992000
+  },
+  "weight_map": {
+    "embeddings.class_embedding": "model-00001-of-00003.safetensors",
+    "embeddings.patch_embedding.bias": "model-00001-of-00003.safetensors",
+    "embeddings.patch_embedding.weight": "model-00001-of-00003.safetensors",
+    "embeddings.position_embedding": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.0.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.1.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.10.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.11.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.12.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.13.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.14.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.15.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.16.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.17.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.18.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.19.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.2.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.20.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.20.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.20.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.20.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.20.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.20.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.20.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.20.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.20.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.20.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.20.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.20.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.20.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.21.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.22.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.23.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.24.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.25.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.26.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.27.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.28.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.29.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.3.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.3.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.30.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.30.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.31.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.32.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.33.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.34.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.35.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.36.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.37.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.38.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.mlp.fc1.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.mlp.fc2.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.norm1.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.39.norm2.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.4.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.4.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.40.attn.k_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.40.attn.proj.bias": "model-00002-of-00003.safetensors",
+    "encoder.layers.40.attn.proj.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.40.attn.q_norm.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.40.attn.qkv.weight": "model-00002-of-00003.safetensors",
+    "encoder.layers.40.ls1": "model-00002-of-00003.safetensors",
+    "encoder.layers.40.ls2": "model-00002-of-00003.safetensors",
+    "encoder.layers.40.mlp.fc1.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.40.mlp.fc1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.40.mlp.fc2.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.40.mlp.fc2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.40.norm1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.40.norm2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.ls1": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.ls2": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.mlp.fc1.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.mlp.fc1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.mlp.fc2.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.mlp.fc2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.norm1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.41.norm2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.ls1": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.ls2": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.mlp.fc1.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.mlp.fc1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.mlp.fc2.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.mlp.fc2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.norm1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.42.norm2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.ls1": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.ls2": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.mlp.fc1.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.mlp.fc1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.mlp.fc2.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.mlp.fc2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.norm1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.43.norm2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.attn.k_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.attn.proj.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.attn.proj.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.attn.q_norm.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.attn.qkv.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.ls1": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.ls2": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.mlp.fc1.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.mlp.fc1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.mlp.fc2.bias": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.mlp.fc2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.norm1.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.44.norm2.weight": "model-00003-of-00003.safetensors",
+    "encoder.layers.5.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.5.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.6.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.7.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.8.norm2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.attn.k_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.attn.proj.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.attn.proj.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.attn.q_norm.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.attn.qkv.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.ls1": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.ls2": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.mlp.fc1.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.mlp.fc2.bias": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.norm1.weight": "model-00001-of-00003.safetensors",
+    "encoder.layers.9.norm2.weight": "model-00001-of-00003.safetensors"
+  }
+}

modeling_intern_vit.py

@@ -0,0 +1,343 @@
+# --------------------------------------------------------
+# InternVL
+# Copyright (c) 2023 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from einops import rearrange
+from timm.models.layers import DropPath
+from torch import nn
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import (BaseModelOutput,
+                                           BaseModelOutputWithPooling)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+
+from .configuration_intern_vit import InternVisionConfig
+
+try:
+    from .flash_attention import FlashAttention
+    has_flash_attn = True
+except:
+    print('FlashAttention is not installed.')
+    has_flash_attn = False
+
+
+logger = logging.get_logger(__name__)
+
+
+class InternRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+
+try:
+    from apex.normalization import FusedRMSNorm
+
+    InternRMSNorm = FusedRMSNorm  # noqa
+
+    logger.info('Discovered apex.normalization.FusedRMSNorm - will use it instead of InternRMSNorm')
+except ImportError:
+    # using the normal InternRMSNorm
+    pass
+except Exception:
+    logger.warning('discovered apex but it failed to load, falling back to InternRMSNorm')
+    pass
+
+
+class InternVisionEmbeddings(nn.Module):
+    def __init__(self, config: InternVisionConfig):
+        super().__init__()
+        self.config = config
+        self.embed_dim = config.hidden_size
+        self.image_size = config.image_size
+        self.patch_size = config.patch_size
+
+        self.class_embedding = nn.Parameter(
+            torch.randn(1, 1, self.embed_dim),
+        )
+
+        self.patch_embedding = nn.Conv2d(
+            in_channels=3, out_channels=self.embed_dim, kernel_size=self.patch_size, stride=self.patch_size
+        )
+
+        self.num_patches = (self.image_size // self.patch_size) ** 2
+        self.num_positions = self.num_patches + 1
+
+        self.position_embedding = nn.Parameter(torch.randn(1, self.num_positions, self.embed_dim))
+
+    def forward(self, pixel_values: torch.FloatTensor) -> torch.Tensor:
+        batch_size = pixel_values.shape[0]
+        target_dtype = self.patch_embedding.weight.dtype
+        patch_embeds = self.patch_embedding(pixel_values)  # shape = [*, width, grid, grid]
+        patch_embeds = patch_embeds.flatten(2).transpose(1, 2)
+        class_embeds = self.class_embedding.expand(batch_size, 1, -1).to(target_dtype)
+        embeddings = torch.cat([class_embeds, patch_embeds], dim=1)
+        embeddings = embeddings + self.position_embedding.to(target_dtype)
+        return embeddings
+
+
+class InternAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: InternVisionConfig):
+        super().__init__()
+        self.config = config
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.use_flash_attn = config.use_flash_attn and has_flash_attn
+        if config.use_flash_attn and not has_flash_attn:
+            print('Warning: Flash Attention is not available, use_flash_attn is set to False.')
+        self.head_dim = self.embed_dim // self.num_heads
+        if self.head_dim * self.num_heads != self.embed_dim:
+            raise ValueError(
+                f'embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:'
+                f' {self.num_heads}).'
+            )
+
+        self.scale = self.head_dim ** -0.5
+        self.qkv = nn.Linear(self.embed_dim, 3 * self.embed_dim, bias=config.qkv_bias)
+        self.attn_drop = nn.Dropout(config.attention_dropout)
+        self.proj_drop = nn.Dropout(config.dropout)
+
+        self.qk_normalization = config.qk_normalization
+
+        if self.qk_normalization:
+            self.q_norm = InternRMSNorm(self.embed_dim, eps=config.layer_norm_eps)
+            self.k_norm = InternRMSNorm(self.embed_dim, eps=config.layer_norm_eps)
+
+        if self.use_flash_attn:
+            self.inner_attn = FlashAttention(attention_dropout=config.attention_dropout)
+        self.proj = nn.Linear(self.embed_dim, self.embed_dim)
+
+    def _naive_attn(self, x):
+        B, N, C = x.shape
+        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv.unbind(0)  # make torchscript happy (cannot use tensor as tuple)
+
+        if self.qk_normalization:
+            B_, H_, N_, D_ = q.shape
+            q = self.q_norm(q.transpose(1, 2).flatten(-2, -1)).view(B_, N_, H_, D_).transpose(1, 2)
+            k = self.k_norm(k.transpose(1, 2).flatten(-2, -1)).view(B_, N_, H_, D_).transpose(1, 2)
+
+        attn = ((q * self.scale) @ k.transpose(-2, -1))
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+    def _flash_attn(self, x, key_padding_mask=None, need_weights=False):
+        qkv = self.qkv(x)
+        qkv = rearrange(qkv, 'b s (three h d) -> b s three h d', three=3, h=self.num_heads)
+
+        if self.qk_normalization:
+            q, k, v = qkv.unbind(2)
+            q = self.q_norm(q.flatten(-2, -1)).view(q.shape)
+            k = self.k_norm(k.flatten(-2, -1)).view(k.shape)
+            qkv = torch.stack([q, k, v], dim=2)
+
+        context, _ = self.inner_attn(
+            qkv, key_padding_mask=key_padding_mask, need_weights=need_weights, causal=False
+        )
+        outs = self.proj(rearrange(context, 'b s h d -> b s (h d)'))
+        outs = self.proj_drop(outs)
+        return outs
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        x = self._naive_attn(hidden_states) if not self.use_flash_attn else self._flash_attn(hidden_states)
+        return x
+
+
+class InternMLP(nn.Module):
+    def __init__(self, config: InternVisionConfig):
+        super().__init__()
+        self.config = config
+        self.act = ACT2FN[config.hidden_act]
+        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
+        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.fc1(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.fc2(hidden_states)
+        return hidden_states
+
+
+class InternVisionEncoderLayer(nn.Module):
+    def __init__(self, config: InternVisionConfig, drop_path_rate: float):
+        super().__init__()
+        self.embed_dim = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+
+        self.attn = InternAttention(config)
+        self.mlp = InternMLP(config)
+        self.norm1 = InternRMSNorm(self.embed_dim, eps=config.layer_norm_eps)
+        self.norm2 = InternRMSNorm(self.embed_dim, eps=config.layer_norm_eps)
+
+        self.ls1 = nn.Parameter(config.initializer_factor * torch.ones(self.embed_dim))
+        self.ls2 = nn.Parameter(config.initializer_factor * torch.ones(self.embed_dim))
+        self.drop_path1 = DropPath(drop_path_rate) if drop_path_rate > 0. else nn.Identity()
+        self.drop_path2 = DropPath(drop_path_rate) if drop_path_rate > 0. else nn.Identity()
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+    ) -> Tuple[torch.FloatTensor, Optional[torch.FloatTensor], Optional[Tuple[torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`Tuple[torch.FloatTensor, Optional[torch.FloatTensor]]`): input to the layer of shape `(batch, seq_len, embed_dim)`
+        """
+        hidden_states = hidden_states + self.drop_path1(self.attn(self.norm1(hidden_states)) * self.ls1)
+
+        hidden_states = hidden_states + self.drop_path2(self.mlp(self.norm2(hidden_states)) * self.ls2)
+
+        return hidden_states
+
+
+class InternVisionEncoder(nn.Module):
+    """
+    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
+    [`InternEncoderLayer`].
+
+    Args:
+        config (`InternConfig`):
+            The corresponding vision configuration for the `InternEncoder`.
+    """
+
+    def __init__(self, config: InternVisionConfig):
+        super().__init__()
+        self.config = config
+        # stochastic depth decay rule
+        dpr = [x.item() for x in torch.linspace(0, config.drop_path_rate, config.num_hidden_layers)]
+        self.layers = nn.ModuleList([
+            InternVisionEncoderLayer(config, dpr[idx]) for idx in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = True
+
+    def forward(
+            self,
+            inputs_embeds,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutput]:
+        r"""
+        Args:
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
+                Embedded representation of the inputs. Should be float, not int tokens.
+            output_hidden_states (`bool`, *optional*):
+                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
+                for more detail.
+            return_dict (`bool`, *optional*):
+                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        """
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        encoder_states = () if output_hidden_states else None
+        hidden_states = inputs_embeds
+
+        for idx, encoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                encoder_states = encoder_states + (hidden_states,)
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    encoder_layer,
+                    hidden_states)
+            else:
+                layer_outputs = encoder_layer(
+                    hidden_states,
+                )
+            hidden_states = layer_outputs
+
+        if output_hidden_states:
+            encoder_states = encoder_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, encoder_states] if v is not None)
+        return BaseModelOutput(
+            last_hidden_state=hidden_states, hidden_states=encoder_states
+        )
+
+
+class InternVisionModel(PreTrainedModel):
+    main_input_name = 'pixel_values'
+    config_class = InternVisionConfig
+    _no_split_modules = ['InternVisionEncoderLayer']
+
+    def __init__(self, config: InternVisionConfig):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = InternVisionEmbeddings(config)
+        self.encoder = InternVisionEncoder(config)
+
+    def resize_pos_embeddings(self, old_size, new_size, patch_size):
+        pos_emb = self.embeddings.position_embedding
+        _, num_positions, embed_dim = pos_emb.shape
+        cls_emb = pos_emb[:, :1, :]
+        pos_emb = pos_emb[:, 1:, :].reshape(1, old_size // patch_size, old_size // patch_size, -1).permute(0, 3, 1, 2)
+        pos_emb = F.interpolate(pos_emb.float(), size=new_size // patch_size, mode='bicubic', align_corners=False)
+        pos_emb = pos_emb.to(cls_emb.dtype).reshape(1, embed_dim, -1).permute(0, 2, 1)
+        pos_emb = torch.cat([cls_emb, pos_emb], dim=1)
+        self.embeddings.position_embedding = nn.Parameter(pos_emb)
+        logger.info('Resized position embeddings from {} to {}'.format(old_size, new_size))
+
+    def get_input_embeddings(self):
+        return self.embeddings
+
+    def forward(
+            self,
+            pixel_values: Optional[torch.FloatTensor] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            pixel_embeds: Optional[torch.FloatTensor] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPooling]:
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if pixel_values is None and pixel_embeds is None:
+            raise ValueError('You have to specify pixel_values or pixel_embeds')
+
+        if pixel_embeds is not None:
+            hidden_states = pixel_embeds
+        else:
+            if len(pixel_values.shape) == 4:
+                hidden_states = self.embeddings(pixel_values)
+            else:
+                raise ValueError(f'wrong pixel_values size: {pixel_values.shape}')
+        encoder_outputs = self.encoder(
+            inputs_embeds=hidden_states,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        last_hidden_state = encoder_outputs.last_hidden_state
+        pooled_output = last_hidden_state[:, 0, :]
+
+        if not return_dict:
+            return (last_hidden_state, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPooling(
+            last_hidden_state=last_hidden_state,
+            pooler_output=pooled_output,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+        )

preprocessor_config.json

@@ -0,0 +1,19 @@
+{
+  "crop_size": 448,
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "resample": 3,
+  "size": 448
+}