Upload folder using huggingface_hub

__init__.py

@@ -0,0 +1,87 @@
+# --------------------------------------------------------
+# InternVL
+# Copyright (c) 2023 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+
+import torch
+import torch.nn as nn
+import torchvision.transforms as T
+from torchvision.transforms import InterpolationMode
+from transformers import LlamaTokenizer
+
+from .configuration_intern_vit import InternVisionConfig
+from .configuration_internvl import InternVLConfig
+from .modeling_intern_vit import InternVisionModel
+from .modeling_internvl import InternVL_C, InternVL_G, InternVLModel
+
+__all__ = ['InternVisionConfig', 'InternVisionModel', 'InternVLConfig',
+           'InternVLModel', 'InternVL_C', 'InternVL_G']
+
+
+# Prefix the text "summarize:"
+class InternVLTokenizer(nn.Module):
+    def __init__(self, model_path):
+        super(InternVLTokenizer, self).__init__()
+        self.tokenizer = LlamaTokenizer.from_pretrained(model_path)
+        self.tokenizer.pad_token = ' '  # allow padding
+        self.tokenizer.add_eos_token = True
+
+    def forward(self, text, prefix='summarize:'):
+        if type(text) == str:
+            text = prefix + text
+        elif type(text) == list:
+            text = [prefix + item for item in text]
+        text = self.tokenizer(text, return_tensors='pt', max_length=80, truncation=True, padding='max_length').input_ids
+        return text
+
+
+def build_transform(task, image_size=224, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]):
+    if task == 'retrieval':
+        transform = T.Compose([
+            T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
+            T.Resize((image_size, image_size), interpolation=InterpolationMode.BICUBIC),
+            T.ToTensor(),
+            T.Normalize(mean=mean, std=std)])
+    else:
+        transform = T.Compose([
+            T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
+            T.Resize(image_size, interpolation=InterpolationMode.BICUBIC),
+            T.CenterCrop(image_size),
+            T.ToTensor(),
+            T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
+    return transform
+
+
+def load_internvl_c_huggingface(ckpt_path, device, task):
+    model = InternVL_C.from_pretrained(ckpt_path, torch_dtype=torch.float16).to(device)
+    if model.config.use_backbone_lora:
+        model.vision_model.merge_and_unload()
+        model.vision_model = model.vision_model.model
+    if model.config.use_qllama_lora:
+        model.qllama.merge_and_unload()
+        model.qllama = model.qllama.model
+    if model.config.force_image_size is not None:
+        image_size = model.config.force_image_size
+    else:
+        image_size = model.config.vision_config.image_size
+    transform = build_transform(task, image_size)
+    tokenizer = InternVLTokenizer(ckpt_path)
+    return model, transform, tokenizer
+
+
+def load_internvl_g_huggingface(ckpt_path, device, task):
+    model = InternVL_G.from_pretrained(ckpt_path, torch_dtype=torch.float16).to(device)
+    if model.config.use_backbone_lora:
+        model.vision_model.merge_and_unload()
+        model.vision_model = model.vision_model.model
+    if model.config.use_qllama_lora:
+        model.qllama.merge_and_unload()
+        model.qllama = model.qllama.model
+    if model.config.force_image_size is not None:
+        image_size = model.config.force_image_size
+    else:
+        image_size = model.config.vision_config.image_size
+    transform = build_transform(task, image_size)
+    tokenizer = InternVLTokenizer(ckpt_path)
+    return model, transform, tokenizer

config.json

@@ -0,0 +1,190 @@
+{
+  "_commit_hash": null,
+  "_name_or_path": "./",
+  "architectures": [
+    "InternVLModel"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_internvl.InternVLConfig",
+    "AutoModel": "modeling_internvl.InternVLModel"
+  },
+  "attn_pool_num_heads": 16,
+  "clip_embed_dim": 768,
+  "force_image_size": null,
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "label_smoothing": 0.0,
+  "max_txt_len": 32,
+  "model_type": "internvl",
+  "num_query_token": 96,
+  "qllama_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": [
+      "LlamaForCausalLM"
+    ],
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "bos_token_id": 1,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_frequency": 2,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": 2,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "silu",
+    "hidden_size": 4096,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 11008,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "max_position_embeddings": 2048,
+    "max_sequence_length": 2048,
+    "min_length": 0,
+    "model_type": "llama",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 32,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 32,
+    "num_key_value_heads": 32,
+    "num_query_token": 96,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": 0,
+    "prefix": null,
+    "pretraining_tp": 1,
+    "problem_type": null,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "rms_norm_eps": 1e-06,
+    "rope_scaling": null,
+    "sep_token_id": null,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": false,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": "float16",
+    "torchscript": false,
+    "transformers_version": "4.32.0",
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "use_cache": false,
+    "vocab_size": 49954
+  },
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": null,
+  "use_backbone_lora": 0,
+  "use_cache": false,
+  "use_decoder_only_language_model": true,
+  "use_qllama_lora": 0,
+  "vision_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "attention_dropout": 0.0,
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "drop_path_rate": 0.0,
+    "dropout": 0.0,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": null,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "gelu",
+    "hidden_size": 3200,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "image_size": 224,
+    "initializer_factor": 0.1,
+    "initializer_range": 1e-10,
+    "intermediate_size": 12800,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-06,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "min_length": 0,
+    "model_type": "intern_vit_6b",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 25,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_channels": 3,
+    "num_hidden_layers": 48,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": null,
+    "patch_size": 14,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "qk_normalization": true,
+    "qkv_bias": false,
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.32.0",
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "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)

configuration_internvl.py

@@ -0,0 +1,108 @@
+# --------------------------------------------------------
+# InternVL
+# Copyright (c) 2023 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+import copy
+
+from transformers import LlamaConfig
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+from .configuration_intern_vit import InternVisionConfig
+
+logger = logging.get_logger(__name__)
+
+
+class InternVLConfig(PretrainedConfig):
+    r"""
+    [`InternVLConfig`] is the configuration class to store the configuration of a
+    [`InternVLModel`]. It is used to instantiate a InternVLModel according to the specified
+    arguments, defining the InternViT-6B and QLLaMA configs. Instantiating a configuration with
+    the defaults will yield a similar configuration to that of the InternVL architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        vision_config (`dict`, *optional*):
+            Dictionary of configuration options used to initialize [`InternVisionConfig`].
+        qllama_config (`dict`, *optional*):
+            Dictionary of configuration options used to initialize [`LLaMAConfig`].
+        clip_embed_dim (`int`, *optional*, defaults to 768):
+            Size of the embeddings from the CLIP model.
+        attn_pool_num_heads (`int`, *optional*, defaults to 16):
+            Number of attention heads used in the attention pooling layers.
+        num_query_token (`int`, *optional*, defaults to 96):
+            Number of query tokens used in the transformer.
+        label_smoothing (`float`, *optional*, defaults to 0.0):
+            The amount of label smoothing to apply.
+        cross_attention_frequency (`int`, *optional*, defaults to 2):
+            The frequency of cross-attention layers in the model.
+        use_backbone_lora (`int`, *optional*, defaults to 0):
+            If non-zero, indicates the use of LoRA in the backbone of the model.
+        use_qllama_lora (`int`, *optional*, defaults to 0):
+            If non-zero, indicates the use of LoRA in the QLLaMA of the model.
+        force_image_size (`int` or `None`, *optional*):
+            If not None, forces the model to use this specific image size.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        kwargs (*optional*):
+            Dictionary of additional keyword arguments.
+    """
+
+    model_type = 'internvl'
+    is_composition = True
+
+    def __init__(
+            self,
+            vision_config=None,
+            qllama_config=None,
+            clip_embed_dim=768,
+            attn_pool_num_heads=16,
+            num_query_token=96,
+            label_smoothing=0.0,
+            cross_attention_frequency=2,
+            use_backbone_lora=0,
+            use_qllama_lora=0,
+            force_image_size=None,
+            initializer_range=0.02,
+            **kwargs):
+        super().__init__(**kwargs)
+
+        if vision_config is None:
+            vision_config = {}
+            logger.info('vision_config is None. initializing the InternVisionConfig with default values.')
+
+        if qllama_config is None:
+            qllama_config = {}
+            logger.info(
+                'qllama_config is None. Initializing the InternTextConfig config with default values (`LlamaConfig`).')
+
+        self.vision_config = InternVisionConfig(**vision_config)
+        self.qllama_config = LlamaConfig(**qllama_config)
+        self.qllama_config.num_query_token = num_query_token
+        self.qllama_config.cross_attention_frequency = cross_attention_frequency
+        self.hidden_size = self.qllama_config.hidden_size
+
+        self.clip_embed_dim = clip_embed_dim
+        self.attn_pool_num_heads = attn_pool_num_heads
+        self.num_query_token = num_query_token
+        self.label_smoothing = label_smoothing
+        self.use_backbone_lora = use_backbone_lora
+        self.use_qllama_lora = use_qllama_lora
+        self.force_image_size = force_image_size
+        self.initializer_range = initializer_range
+
+    def to_dict(self):
+        """
+        Serializes this instance to a Python dictionary. Override the default [`~PretrainedConfig.to_dict`].
+
+        Returns:
+            `Dict[str, any]`: Dictionary of all the attributes that make up this configuration instance,
+        """
+        output = copy.deepcopy(self.__dict__)
+        output['vision_config'] = self.vision_config.to_dict()
+        output['qllama_config'] = self.qllama_config.to_dict()
+        output['model_type'] = self.__class__.model_type
+        return output

flash_attention.py

@@ -0,0 +1,76 @@
+# https://github.com/Dao-AILab/flash-attention/blob/v0.2.8/flash_attn/flash_attention.py
+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

modeling_intern_vit.py

@@ -0,0 +1,342 @@
+# --------------------------------------------------------
+# 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
+
+    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,
+        )

modeling_internvl.py

@@ -0,0 +1,519 @@
+# --------------------------------------------------------
+# InternVL
+# Copyright (c) 2023 OpenGVLab
+# Licensed under The MIT License [see LICENSE for details]
+# --------------------------------------------------------
+from functools import partial
+from typing import Optional
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from peft import LoraConfig, get_peft_model
+from timm.models.layers import DropPath
+from torch import nn
+from transformers import GenerationConfig
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+
+from .configuration_internvl import InternVLConfig
+from .modeling_intern_vit import (InternVisionEmbeddings, InternVisionEncoder,
+                                  InternVisionModel)
+from .modeling_qllama import LlamaForCausalLM, _expand_mask, _make_causal_mask
+
+try:
+    from .flash_attention import FlashAttention  # v1/v2
+except:
+    print('FlashAttention is not installed.')
+
+logger = logging.get_logger(__name__)
+
+
+class InternVLPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = InternVLConfig
+    base_model_prefix = 'internvl'
+    supports_gradient_checkpointing = True
+    _keys_to_ignore_on_load_missing = [
+        r'position_ids',
+    ]
+    _no_split_modules = ['InternAttention', 'LlamaDecoderLayer', 'LlamaForCausalLM']
+    _skip_keys_device_placement = 'past_key_values'
+    _keep_in_fp32_modules = ['wo']
+
+    def _init_weights(self, module):
+        """Initialize the weights"""
+        factor = self.config.initializer_range
+        if isinstance(module, nn.Conv2d) or isinstance(module, nn.Embedding) or isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=factor)
+            if hasattr(module, 'bias') and module.bias is not None:
+                module.bias.data.zero_()
+        if isinstance(module, InternVisionEmbeddings):
+            if hasattr(self.config, 'vision_config'):
+                factor = self.config.vision_config.initializer_range
+            nn.init.trunc_normal_(module.position_embedding, mean=0.0, std=factor)
+            nn.init.trunc_normal_(module.class_embedding, mean=0.0, std=factor)
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+        elif isinstance(module, nn.Linear) and module.bias is not None:
+            module.bias.data.zero_()
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, InternVisionModel):
+            module.gradient_checkpointing = value
+        if isinstance(module, InternVisionEncoder):
+            module.gradient_checkpointing = value
+
+
+class CrossAttention(nn.Module):
+    def __init__(
+            self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0.,
+            proj_drop=0., attn_head_dim=None, out_dim=None):
+        super().__init__()
+        if out_dim is None:
+            out_dim = dim
+        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
+        assert all_head_dim == dim
+
+        self.q = nn.Linear(dim, all_head_dim, bias=False)
+        self.k = nn.Linear(dim, all_head_dim, bias=False)
+        self.v = nn.Linear(dim, all_head_dim, bias=False)
+
+        if qkv_bias:
+            self.q_bias = nn.Parameter(torch.zeros(all_head_dim))
+            self.k_bias = nn.Parameter(torch.zeros(all_head_dim))
+            self.v_bias = nn.Parameter(torch.zeros(all_head_dim))
+        else:
+            self.q_bias = None
+            self.k_bias = None
+            self.v_bias = None
+
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(all_head_dim, out_dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, x, k=None, v=None):
+        B, N, C = x.shape
+        N_k = k.shape[1]
+        N_v = v.shape[1]
+
+        q_bias, k_bias, v_bias = None, None, None
+        if self.q_bias is not None:
+            q_bias = self.q_bias
+            k_bias = self.k_bias
+            v_bias = self.v_bias
+
+        q = F.linear(input=x, weight=self.q.weight, bias=q_bias)
+        q = q.reshape(B, N, 1, self.num_heads, -1).permute(2, 0, 3, 1, 4).squeeze(0)  # (B, N_head, N_q, dim)
+
+        k = F.linear(input=k, weight=self.k.weight, bias=k_bias)
+        k = k.reshape(B, N_k, 1, self.num_heads, -1).permute(2, 0, 3, 1, 4).squeeze(0)
+
+        v = F.linear(input=v, weight=self.v.weight, bias=v_bias)
+        v = v.reshape(B, N_v, 1, self.num_heads, -1).permute(2, 0, 3, 1, 4).squeeze(0)
+
+        q = q * self.scale
+        attn = (q @ k.transpose(-2, -1))  # (B, N_head, N_q, N_k)
+
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B, N, -1)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+
+        return x
+
+
+class AttentiveBlock(nn.Module):
+
+    def __init__(self, dim, num_heads, qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
+                 drop_path=0., norm_layer=nn.LayerNorm, attn_head_dim=None, out_dim=None):
+        super().__init__()
+
+        self.norm1_q = norm_layer(dim)
+        self.norm1_k = norm_layer(dim)
+        self.norm1_v = norm_layer(dim)
+        self.cross_attn = CrossAttention(
+            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop,
+            proj_drop=drop, attn_head_dim=attn_head_dim, out_dim=out_dim)
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+
+    def forward(self, x_q, x_kv, pos_q, pos_k, bool_masked_pos, rel_pos_bias=None):
+        x_q = self.norm1_q(x_q + pos_q)
+        x_k = self.norm1_k(x_kv + pos_k)
+        x_v = self.norm1_v(x_kv)
+        x = self.cross_attn(x_q, k=x_k, v=x_v)
+
+        return x
+
+
+class AttentionPoolingBlock(AttentiveBlock):
+
+    def forward(self, x):
+        x_q = x.mean(1, keepdim=True)
+        x_kv, pos_q, pos_k = x, 0, 0
+        x = super().forward(x_q, x_kv, pos_q, pos_k, bool_masked_pos=None, rel_pos_bias=None)
+        x = x.squeeze(1)
+        return x
+
+
+class InternVLModel(InternVLPreTrainedModel):
+    config_class = InternVLConfig
+    main_input_name = 'pixel_values'
+
+    def __init__(self, config: InternVLConfig):
+        super().__init__(config)
+
+        text_hidden_size = config.qllama_config.hidden_size
+        vision_hidden_size = config.vision_config.hidden_size
+        clip_embed_dim = config.clip_embed_dim
+        attn_pool_num_heads = config.attn_pool_num_heads
+        config.qllama_config.num_query_token = config.num_query_token
+        self.num_query_token = config.num_query_token
+        self.label_smoothing = config.label_smoothing
+
+        self.vision_model = InternVisionModel(config.vision_config)  # frozen
+        self.qllama = LlamaForCausalLM(config.qllama_config)  # frozen
+        self.query_tokens = nn.Parameter(  # trainable
+            torch.zeros(1, config.num_query_token, text_hidden_size)
+        )
+
+        self.text_projection = nn.Parameter(torch.empty(text_hidden_size, clip_embed_dim))  # frozen
+        self.logit_scale = nn.Parameter(torch.ones([]) * np.log(1 / 0.07))  # trainable
+        self.clip_projector = AttentionPoolingBlock(  # frozen
+            dim=vision_hidden_size, num_heads=attn_pool_num_heads, qkv_bias=True, qk_scale=None,
+            drop=0., attn_drop=0., norm_layer=partial(nn.LayerNorm, eps=1e-5), out_dim=clip_embed_dim)
+        self.clip_projector2 = AttentionPoolingBlock(  # trainable
+            dim=text_hidden_size, num_heads=attn_pool_num_heads, qkv_bias=True, qk_scale=None,
+            drop=0., attn_drop=0., norm_layer=partial(nn.LayerNorm, eps=1e-5), out_dim=clip_embed_dim)
+        self.itm_head = nn.Linear(text_hidden_size, 2)  # trainable
+        self.gradient_checkpointing = True
+
+        # Initialize weights and apply final processing
+        # self.post_init()
+
+        if config.use_backbone_lora:
+            self.wrap_backbone_lora(r=config.use_backbone_lora)
+        if config.use_qllama_lora:
+            self.wrap_qllama_lora(r=config.use_qllama_lora)
+        if config.force_image_size:
+            self.vision_model.resize_pos_embeddings(
+                old_size=config.vision_config.image_size,
+                new_size=config.force_image_size,
+                patch_size=config.vision_config.patch_size
+            )
+
+    def wrap_backbone_lora(self, r=128, lora_alpha=256, lora_dropout=0.05):
+        lora_config = LoraConfig(
+            r=r,
+            target_modules=['attn.qkv', 'attn.proj', 'mlp.fc1', 'mlp.fc2'],
+            lora_alpha=lora_alpha,
+            lora_dropout=lora_dropout,
+        )
+        self.vision_model = get_peft_model(self.vision_model, lora_config)
+        self.vision_model.print_trainable_parameters()
+
+    def wrap_qllama_lora(self, r=128, lora_alpha=256, lora_dropout=0.05):
+        lora_config = LoraConfig(
+            r=r,
+            target_modules=['self_attn.q_proj', 'self_attn.k_proj', 'self_attn.v_proj', 'self_attn.o_proj',
+                            'mlp.gate_proj', 'mlp.down_proj', 'mlp.up_proj'],
+            lora_alpha=lora_alpha,
+            lora_dropout=lora_dropout,
+        )
+        self.qllama = get_peft_model(self.qllama, lora_config)
+        self.qllama.print_trainable_parameters()
+
+    def get_input_embeddings(self):
+        return self.qllama.get_input_embeddings()
+
+    def set_input_embeddings(self, value):
+        self.qllama.set_input_embeddings(value)
+
+    def set_output_embeddings(self, new_embeddings):
+        self.qllama.set_output_embeddings(new_embeddings)
+
+    def get_output_embeddings(self) -> nn.Module:
+        return self.qllama.get_output_embeddings()
+
+    @torch.no_grad()
+    def generate(
+            self,
+            pixel_values: torch.FloatTensor,
+            input_ids: torch.FloatTensor,
+            attention_mask: torch.LongTensor,
+            generation_config: Optional[GenerationConfig] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            **generate_kwargs,
+    ) -> torch.LongTensor:
+
+        vision_outputs = self.vision_model(
+            pixel_values=pixel_values,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict)
+        image_embeds = vision_outputs[0]
+
+        batch_size = image_embeds.shape[0]
+        input_embeds = self.get_input_embeddings()(input_ids)
+        query_tokens = self.query_tokens.repeat(batch_size, 1, 1)
+        input_embeds = torch.cat([query_tokens, input_embeds], dim=1)
+        image_attention_mask = torch.ones(query_tokens.size()[:-1], dtype=torch.long, device=image_embeds.device)
+        attention_mask = torch.cat([image_attention_mask, attention_mask], dim=1)
+
+        outputs = self.qllama.generate(
+            inputs_embeds=input_embeds,
+            attention_mask=attention_mask,
+            vision_hidden_states=image_embeds,
+            generation_config=generation_config,
+            use_zero_attention_mask=True,
+            **generate_kwargs,
+        )
+
+        return outputs
+
+    def get_text_features(
+            self,
+            input_ids: torch.Tensor,
+            attention_mask: torch.Tensor,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ):
+        r"""
+        Returns:
+            text_outputs (`CausalLMOutputWithPast`, or `tuple(torch.FloatTensor)` if `return_dict=False`):
+                The language model outputs. If `return_dict=True`, the output is a [`CausalLMOutputWithPast`] that
+                contains the language model logits, the past key values and the hidden states if
+                `output_hidden_states=True`.
+        ```"""
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        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
+
+        input_embeds = self.get_input_embeddings()(input_ids)
+        attention_mask = _expand_mask(attention_mask, input_embeds.dtype).to(
+            input_embeds.device)  # [bsz, 1, tgt_seq_len, src_seq_len]
+        attention_mask += _make_causal_mask(
+            (attention_mask.shape[0], attention_mask.shape[2]),
+            input_embeds.dtype,
+            device=input_embeds.device
+        )
+        if type(self.qllama.model) == LlamaForCausalLM:
+            outputs = self.qllama.model.model.forward_train(
+                inputs_embeds=input_embeds,
+                vision_hidden_states=None,
+                attention_mask=attention_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            ).last_hidden_state
+        else:
+            outputs = self.qllama.model.forward_train(
+                inputs_embeds=input_embeds,
+                vision_hidden_states=None,
+                attention_mask=attention_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            ).last_hidden_state
+        return outputs
+
+    def get_image_features(
+            self,
+            pixel_values: torch.FloatTensor,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ):
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        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
+
+        vision_outputs = self.vision_model(
+            pixel_values=pixel_values,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict)
+        image_embeds = vision_outputs[0]
+        backbone_embeds = image_embeds
+
+        batch_size = image_embeds.shape[0]
+        input_embeds = self.query_tokens.repeat(batch_size, 1, 1)
+
+        attention_mask = torch.ones(input_embeds.size()[:-1], dtype=torch.long, device=image_embeds.device)
+        attention_mask = _expand_mask(attention_mask, input_embeds.dtype).to(
+            input_embeds.device)  # [bsz, 1, tgt_seq_len, src_seq_len]
+        if type(self.qllama.model) == LlamaForCausalLM:
+            outputs = self.qllama.model.model.forward_train(
+                inputs_embeds=input_embeds,
+                vision_hidden_states=image_embeds,
+                attention_mask=attention_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            ).last_hidden_state
+        else:
+            outputs = self.qllama.model.forward_train(
+                inputs_embeds=input_embeds,
+                vision_hidden_states=image_embeds,
+                attention_mask=attention_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            ).last_hidden_state
+        return backbone_embeds, outputs
+
+    def encode_image(self, image, mode):
+        if mode == 'InternVL-C':
+            vision_outputs = self.vision_model(
+                pixel_values=image,
+                output_hidden_states=False,
+                return_dict=True)
+            image_embeds = vision_outputs[0]
+            image_embeds = self.clip_projector(image_embeds)
+        elif mode == 'InternVL-G':
+            backbone_embeds, image_embeds = self.get_image_features(
+                pixel_values=image,
+                output_hidden_states=False,
+                return_dict=True,
+            )
+            backbone_embeds = self.clip_projector(backbone_embeds)
+            image_embeds = self.clip_projector2(image_embeds)
+            # ensemble
+            backbone_embeds = backbone_embeds / backbone_embeds.norm(dim=1, keepdim=True)
+            image_embeds = image_embeds / image_embeds.norm(dim=1, keepdim=True)
+            image_embeds = image_embeds + backbone_embeds
+        else:
+            raise NotImplementedError
+        return image_embeds
+
+    def encode_text(self, text):
+        attention_mask = text > 0
+        text_embeds = self.get_text_features(
+            input_ids=text,
+            attention_mask=attention_mask,
+            output_attentions=False,
+            output_hidden_states=False,
+            return_dict=True,
+        )
+        text_embeds = text_embeds[torch.arange(text_embeds.shape[0]), attention_mask.sum(1) - 1]
+        text_embeds = text_embeds @ self.text_projection
+        return text_embeds
+
+    def forward(self, image, text, mode='InternVL-C'):
+        assert mode in ['InternVL-C', 'InternVL-G'], 'mode must be InternVL-C or InternVL-G'
+        image_features = self.encode_image(image, mode)
+        text_features = self.encode_text(text)
+
+        # normalized features
+        image_features = image_features / image_features.norm(dim=1, keepdim=True)
+        text_features = text_features / text_features.norm(dim=1, keepdim=True)
+
+        # cosine similarity as logits
+        logit_scale = self.logit_scale.exp()
+        logits_per_image = logit_scale * image_features @ text_features.t()
+        logits_per_text = logits_per_image.t()
+
+        return logits_per_image, logits_per_text
+
+
+class InternVL_C(InternVLModel):
+
+    def encode_image(self, image):
+        vision_outputs = self.vision_model(
+            pixel_values=image,
+            output_hidden_states=False,
+            return_dict=True)
+        image_embeds = vision_outputs[0]
+        image_embeds = self.clip_projector(image_embeds)
+        return image_embeds
+
+    def encode_text(self, text):
+        attention_mask = text > 0
+        text_embeds = self.get_text_features(
+            input_ids=text,
+            attention_mask=attention_mask,
+            output_attentions=False,
+            output_hidden_states=False,
+            return_dict=True,
+        )
+        text_embeds = text_embeds[torch.arange(text_embeds.shape[0]), attention_mask.sum(1) - 1]
+        text_embeds = text_embeds @ self.text_projection
+        return text_embeds
+
+    def forward(self, image, text):
+        image_features = self.encode_image(image)
+        text_features = self.encode_text(text)
+
+        # normalized features
+        image_features = image_features / image_features.norm(dim=1, keepdim=True)
+        text_features = text_features / text_features.norm(dim=1, keepdim=True)
+
+        # cosine similarity as logits
+        logit_scale = self.logit_scale.exp()
+        logits_per_image = logit_scale * image_features @ text_features.t()
+        logits_per_text = logits_per_image.t()
+
+        return logits_per_image, logits_per_text
+
+
+class InternVL_G(InternVLModel):
+
+    def encode_image(self, image):
+        backbone_embeds, image_embeds = self.get_image_features(
+            pixel_values=image,
+            output_hidden_states=False,
+            return_dict=True,
+        )
+        backbone_embeds = self.clip_projector(backbone_embeds)
+        image_embeds = self.clip_projector2(image_embeds)
+        # ensemble
+        backbone_embeds = backbone_embeds / backbone_embeds.norm(dim=1, keepdim=True)
+        image_embeds = image_embeds / image_embeds.norm(dim=1, keepdim=True)
+        image_embeds = image_embeds + backbone_embeds
+        return image_embeds
+
+    def encode_text(self, text):
+        attention_mask = text > 0
+        text_embeds = self.get_text_features(
+            input_ids=text,
+            attention_mask=attention_mask,
+            output_attentions=False,
+            output_hidden_states=False,
+            return_dict=True,
+        )
+        text_embeds = text_embeds[torch.arange(text_embeds.shape[0]), attention_mask.sum(1) - 1]
+        text_embeds = text_embeds @ self.text_projection
+        return text_embeds
+
+    def forward(self, image, text):
+        image_features = self.encode_image(image)
+        text_features = self.encode_text(text)
+
+        # normalized features
+        image_features = image_features / image_features.norm(dim=1, keepdim=True)
+        text_features = text_features / text_features.norm(dim=1, keepdim=True)
+
+        # cosine similarity as logits
+        logit_scale = self.logit_scale.exp()
+        logits_per_image = logit_scale * image_features @ text_features.t()
+        logits_per_text = logits_per_image.t()
+
+        return logits_per_image, logits_per_text

modeling_qllama.py

@@ -0,0 +1,1073 @@
+# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
+#
+# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
+# and OPT implementations in this library. It has been modified from its
+# original forms to accommodate minor architectural differences compared
+# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" PyTorch QLLaMA model."""
+import math
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+from transformers import LlamaConfig
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import (BaseModelOutputWithPast,
+                                           CausalLMOutputWithPast)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (add_start_docstrings,
+                                add_start_docstrings_to_model_forward, logging,
+                                replace_return_docstrings)
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = 'LlamaConfig'
+
+
+# Copied from transformers.models.bart.modeling_bart._make_causal_mask
+def _make_causal_mask(
+        input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
+):
+    """
+    Make causal mask used for bi-directional self-attention.
+    """
+    bsz, tgt_len = input_ids_shape
+    mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
+    mask_cond = torch.arange(mask.size(-1), device=device)
+    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
+    mask = mask.to(dtype)
+
+    if past_key_values_length > 0:
+        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
+    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
+
+
+# Copied from transformers.models.bart.modeling_bart._expand_mask
+def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+
+class LlamaRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        LlamaRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+
+        # convert into half-precision if necessary
+        if self.weight.dtype in [torch.float16, torch.bfloat16]:
+            hidden_states = hidden_states.to(self.weight.dtype)
+
+        return self.weight * hidden_states
+
+
+try:
+    from functools import partial
+
+    from apex.normalization import FusedRMSNorm
+
+    LlamaRMSNorm = partial(FusedRMSNorm, eps=1e-6)  # noqa
+    print('Discovered apex.normalization.FusedRMSNorm - will use it instead of LlamaRMSNorm')
+except ImportError:
+    # using the normal LlamaRMSNorm
+    pass
+except Exception:
+    print('discovered apex but it failed to load, falling back to LlamaRMSNorm')
+    pass
+
+
+class LlamaRotaryEmbedding(torch.nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
+        super().__init__()
+        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float().to(device) / dim))
+        self.register_buffer('inv_freq', inv_freq)
+
+        # Build here to make `torch.jit.trace` work.
+        self.max_seq_len_cached = max_position_embeddings
+        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=self.inv_freq.dtype)
+        freqs = torch.einsum('i,j->ij', t, self.inv_freq)
+        # Different from paper, but it uses a different permutation in order to obtain the same calculation
+        emb = torch.cat((freqs, freqs), dim=-1)
+        self.register_buffer('cos_cached', emb.cos()[None, None, :, :], persistent=False)
+        self.register_buffer('sin_cached', emb.sin()[None, None, :, :], persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        # This `if` block is unlikely to be run after we build sin/cos in `__init__`. Keep the logic here just in case.
+        if seq_len > self.max_seq_len_cached:
+            self.max_seq_len_cached = seq_len
+            t = torch.arange(self.max_seq_len_cached, device=x.device, dtype=self.inv_freq.dtype)
+            freqs = torch.einsum('i,j->ij', t, self.inv_freq)
+            # Different from paper, but it uses a different permutation in order to obtain the same calculation
+            emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
+            self.register_buffer('cos_cached', emb.cos()[None, None, :, :], persistent=False)
+            self.register_buffer('sin_cached', emb.sin()[None, None, :, :], persistent=False)
+        return (
+            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+        )
+
+
+class FixedLlamaRotaryEmbedding(torch.nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
+        super().__init__()
+
+        self.dim = dim
+        self.max_position_embeddings = max_position_embeddings
+        self.base = base
+        self.inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim))
+
+        # Build here to make `torch.jit.trace` work.
+        self._set_cos_sin_cache(
+            seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype()
+        )
+
+    def _set_cos_sin_cache(self, seq_len, device, dtype):
+        self.max_seq_len_cached = seq_len
+        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=torch.float32)
+
+        freqs = torch.outer(t, self.inv_freq)
+        # Different from paper, but it uses a different permutation in order to obtain the same calculation
+        emb = torch.cat((freqs, freqs), dim=-1)
+        self.register_buffer('cos_cached', emb.cos()[None, None, :, :], persistent=False)
+        self.register_buffer('sin_cached', emb.sin()[None, None, :, :], persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        if seq_len > self.max_seq_len_cached:
+            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype)
+
+        return (
+            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+        )
+
+
+LlamaRotaryEmbedding = FixedLlamaRotaryEmbedding
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2:]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
+    gather_indices = position_ids[:, None, :, None]  # [bs, 1, seq_len, 1]
+    gather_indices = gather_indices.repeat(1, cos.shape[1], 1, cos.shape[3])
+    cos = torch.gather(cos.repeat(gather_indices.shape[0], 1, 1, 1), 2, gather_indices)
+    sin = torch.gather(sin.repeat(gather_indices.shape[0], 1, 1, 1), 2, gather_indices)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class LlamaMLP(nn.Module):
+    def __init__(
+            self,
+            hidden_size: int,
+            intermediate_size: int,
+            hidden_act: str,
+    ):
+        super().__init__()
+        self.gate_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.down_proj = nn.Linear(intermediate_size, hidden_size, bias=False)
+        self.up_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.act_fn = ACT2FN[hidden_act]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+class LlamaAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: LlamaConfig):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.max_position_embeddings = config.max_position_embeddings
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f'hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}'
+                f' and `num_heads`: {self.num_heads}).'
+            )
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+        self.rotary_emb = LlamaRotaryEmbedding(self.head_dim, max_position_embeddings=self.max_position_embeddings)
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            output_attentions: bool = False,
+            use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = self.k_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        value_states = self.v_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+        # [bsz, nh, t, hd]
+
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f'Attention weights should be of size {(bsz * self.num_heads, q_len, kv_seq_len)}, but is'
+                f' {attn_weights.size()}'
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f'Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}'
+                )
+            attn_weights = attn_weights + attention_mask
+            attn_weights = torch.max(attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min))
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f'`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is'
+                f' {attn_output.size()}'
+            )
+
+        attn_output = attn_output.transpose(1, 2)
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+class LlamaCrossAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: LlamaConfig):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.vision_hidden_size = 3200
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f'hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}'
+                f' and `num_heads`: {self.num_heads}).'
+            )
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+        self.norm1 = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.k_proj = nn.Linear(self.vision_hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(self.vision_hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.norm2 = LlamaRMSNorm(self.vision_hidden_size, eps=config.rms_norm_eps)
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            vision_hidden_states: torch.Tensor,
+            repeat_time: int = 1,
+            attention_mask: Optional[torch.Tensor] = None,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            output_attentions: bool = False,
+            use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        hidden_states = self.norm1(hidden_states)
+
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+
+        vision_hidden_states = self.norm2(vision_hidden_states)
+
+        bs_v, kv_len, _ = vision_hidden_states.size()
+
+        key_states = self.k_proj(vision_hidden_states).view(
+            bs_v, kv_len, self.num_heads, self.head_dim).transpose(1, 2)
+        value_states = self.v_proj(vision_hidden_states).view(
+            bs_v, kv_len, self.num_heads, self.head_dim).transpose(1, 2)
+
+        key_states = key_states.repeat(repeat_time, 1, 1, 1)
+        value_states = value_states.repeat(repeat_time, 1, 1, 1)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f'Attention weights should be of size {(bsz * self.num_heads, q_len, kv_seq_len)}, but is'
+                f' {attn_weights.size()}'
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f'Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}'
+                )
+            attn_weights = attn_weights + attention_mask
+            attn_weights = torch.max(attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min))
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f'`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is'
+                f' {attn_output.size()}'
+            )
+
+        attn_output = attn_output.transpose(1, 2)
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+class LlamaDecoderLayer(nn.Module):
+    def __init__(self, config: LlamaConfig, use_cross_attn: bool):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = LlamaAttention(config=config)
+        self.cross_attn = LlamaCrossAttention(config=config) if use_cross_attn else None
+        self.mlp = LlamaMLP(
+            hidden_size=self.hidden_size,
+            intermediate_size=config.intermediate_size,
+            hidden_act=config.hidden_act,
+        )
+        self.num_query_token = 96
+        self.input_layernorm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            vision_hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            output_attentions: Optional[bool] = False,
+            use_cache: Optional[bool] = False,
+            repeat_time: int = 1,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        hidden_states = residual + hidden_states
+
+        # when using generate function and cache mode, the size of hidden_states is 1,
+        # so we should not use cross attention
+        if self.cross_attn is not None and hidden_states.size(1) >= self.num_query_token \
+                and vision_hidden_states is not None:
+            query_feats = hidden_states[:, :self.num_query_token, :]
+            text_feats = hidden_states[:, self.num_query_token:, :]
+            residual = query_feats
+            query_feats, _, _ = self.cross_attn(
+                hidden_states=query_feats,
+                vision_hidden_states=vision_hidden_states,
+                attention_mask=None,  # not use attention mask in cross attention
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+                repeat_time=repeat_time,
+            )
+            query_feats = residual + query_feats
+            hidden_states = torch.cat([query_feats, text_feats], dim=1)
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+LLAMA_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`LlamaConfig`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    'The bare LLaMA Model outputting raw hidden-states without any specific head on top.',
+    LLAMA_START_DOCSTRING,
+)
+class LlamaPreTrainedModel(PreTrainedModel):
+    config_class = LlamaConfig
+    base_model_prefix = 'model'
+    supports_gradient_checkpointing = True
+    _no_split_modules = ['LlamaDecoderLayer']
+    _keys_to_ignore_on_load_unexpected = [r'decoder\.version']
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, LlamaModel):
+            module.gradient_checkpointing = value
+        if isinstance(module, LlamaDecoderLayer):
+            module.gradient_checkpointing = value
+
+
+LLAMA_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        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.
+"""
+
+
+@add_start_docstrings(
+    'The bare LLaMA Model outputting raw hidden-states without any specific head on top.',
+    LLAMA_START_DOCSTRING,
+)
+class LlamaModel(LlamaPreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`LlamaDecoderLayer`]
+
+    Args:
+        config: LlamaConfig
+    """
+
+    def __init__(self, config: LlamaConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+        self.cross_attention_frequency = config.cross_attention_frequency
+        self.num_query_token = config.num_query_token
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        use_cross_attn = [idx % self.cross_attention_frequency == 0 for idx in range(config.num_hidden_layers)]
+        self.layers = nn.ModuleList(
+            [LlamaDecoderLayer(config, use_cross_attn[idx]) for idx in range(config.num_hidden_layers)])
+        self.norm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        # self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
+    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
+        # create causal mask
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        combined_attention_mask = None
+        if input_shape[-1] > 1:
+            combined_attention_mask = _make_causal_mask(
+                input_shape,
+                inputs_embeds.dtype,
+                device=inputs_embeds.device,
+                past_key_values_length=past_key_values_length,
+            )
+
+        if attention_mask is not None:
+            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
+                inputs_embeds.device
+            )
+            combined_attention_mask = (
+                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
+            )
+
+        return combined_attention_mask
+
+    @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            vision_hidden_states: Optional[torch.FloatTensor] = None,
+            repeat_time: Optional[int] = 1,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            use_zero_attention_mask: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError('You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time')
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError('You have to specify either decoder_input_ids or decoder_inputs_embeds')
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+
+        if past_key_values is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        # embed positions
+        if attention_mask is None:
+            attention_mask = torch.ones(
+                (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
+            )
+        attention_mask = self._prepare_decoder_attention_mask(
+            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
+        )
+        if use_zero_attention_mask:
+            attention_mask[:, :, :self.num_query_token, :self.num_query_token] = 0
+
+        hidden_states = inputs_embeds
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    '`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...'
+                )
+                use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+            layer_outputs = decoder_layer(
+                hidden_states,
+                vision_hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+                repeat_time=repeat_time,
+            )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+    @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+    def forward_train(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            vision_hidden_states: Optional[torch.FloatTensor] = None,
+            repeat_time: Optional[int] = 1,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError('You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time')
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError('You have to specify either decoder_input_ids or decoder_inputs_embeds')
+
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+
+        if past_key_values is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        # embed positions
+        # if attention_mask is None:
+        #     attention_mask = torch.ones(
+        #         (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
+        #     )
+        # attention_mask = self._prepare_decoder_attention_mask(
+        #     attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
+        # )
+        hidden_states = inputs_embeds
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    '`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...'
+                )
+                use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, output_attentions, None, repeat_time)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(decoder_layer),
+                    hidden_states,
+                    vision_hidden_states,
+                    attention_mask,
+                    position_ids,
+                    None,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    vision_hidden_states,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                    repeat_time=repeat_time,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+class LlamaForCausalLM(LlamaPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = LlamaModel(config)
+
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        # self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            vision_hidden_states: Optional[torch.FloatTensor] = None,
+            labels: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            use_zero_attention_mask: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, LlamaForCausalLM
+
+        >>> model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
+        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
+
+        >>> prompt = "Hey, are you consciours? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        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
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            vision_hidden_states=vision_hidden_states,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            use_zero_attention_mask=use_zero_attention_mask,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+            self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None,
+            vision_hidden_states=None, use_zero_attention_mask=None, **kwargs
+    ):
+        if past_key_values:
+            input_ids = input_ids[:, -1:]
+
+        position_ids = kwargs.get('position_ids', None)
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {'inputs_embeds': inputs_embeds}
+        else:
+            model_inputs = {'input_ids': input_ids}
+
+        model_inputs.update(
+            {
+                'position_ids': position_ids,
+                'past_key_values': past_key_values,
+                'use_cache': kwargs.get('use_cache'),
+                'attention_mask': attention_mask,
+                'vision_hidden_states': vision_hidden_states,
+                'use_zero_attention_mask': use_zero_attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past

preprocessor_config.json

@@ -0,0 +1,19 @@
+{
+  "crop_size": 224,
+  "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": 224
+}

pytorch_model-00001-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a49179b5853f94174604ee23142631020f08ce464e45a29b6238a624addb407
+size 9928107961

pytorch_model-00002-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e83b1754a4c3449848ecb7cb00ffddd80e34f2fd5cf8c4a7edea473ea9a7aac4
+size 9980526335

pytorch_model-00003-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d03c3fec7174e3543f0be009f8ab4765fcd7f1837bf74ec16976c40297aa2e63
+size 7761682141

pytorch_model.bin.index.json

@@ -0,0 +1,1055 @@
+{
+  "metadata": {
+    "total_size": 27669951494
+  },
+  "weight_map": {
+    "clip_projector.cross_attn.k.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.cross_attn.k_bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.cross_attn.proj.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.cross_attn.proj.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.cross_attn.q.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.cross_attn.q_bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.cross_attn.v.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.cross_attn.v_bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.norm1_k.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.norm1_k.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.norm1_q.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.norm1_q.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.norm1_v.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector.norm1_v.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.k.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.k_bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.proj.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.proj.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.q.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.q_bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.v.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.cross_attn.v_bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.norm1_k.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.norm1_k.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.norm1_q.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.norm1_q.weight": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.norm1_v.bias": "pytorch_model-00003-of-00003.bin",
+    "clip_projector2.norm1_v.weight": "pytorch_model-00003-of-00003.bin",
+    "itm_head.bias": "pytorch_model-00003-of-00003.bin",
+    "itm_head.weight": "pytorch_model-00003-of-00003.bin",
+    "logit_scale": "pytorch_model-00001-of-00003.bin",
+    "qllama.lm_head.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.embed_tokens.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.0.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.1.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.10.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.11.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.12.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.13.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.14.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.15.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.16.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.16.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.16.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.16.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.16.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.16.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.17.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.17.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.cross_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.cross_attn.norm1.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.cross_attn.norm2.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.cross_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.cross_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.cross_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.18.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.19.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.2.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.2.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.20.cross_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.cross_attn.norm1.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.cross_attn.norm2.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.cross_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.cross_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.cross_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.20.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.21.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.cross_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.cross_attn.norm1.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.cross_attn.norm2.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.cross_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.cross_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.cross_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.22.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.23.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.cross_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.cross_attn.norm1.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.cross_attn.norm2.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.cross_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.cross_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.cross_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.24.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.25.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.cross_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.cross_attn.norm1.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.cross_attn.norm2.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.cross_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.cross_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.cross_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.26.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.27.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.cross_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.cross_attn.norm1.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.cross_attn.norm2.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.cross_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.cross_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.cross_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.28.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.29.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.3.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.30.cross_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.cross_attn.norm1.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.cross_attn.norm2.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.cross_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.cross_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.cross_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.30.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.31.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "qllama.model.layers.4.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.4.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.5.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.6.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.7.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.cross_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.cross_attn.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.cross_attn.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.cross_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.cross_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.cross_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.8.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.layers.9.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "qllama.model.norm.weight": "pytorch_model-00003-of-00003.bin",
+    "query_tokens": "pytorch_model-00001-of-00003.bin",
+    "text_projection": "pytorch_model-00001-of-00003.bin",
+    "vision_model.embeddings.class_embedding": "pytorch_model-00001-of-00003.bin",
+    "vision_model.embeddings.patch_embedding.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.embeddings.patch_embedding.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.embeddings.position_embedding": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.0.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.1.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.10.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.11.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.12.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.13.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.14.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.15.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.16.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.17.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.18.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.19.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.2.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.20.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.21.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.22.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.23.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.24.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.25.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.26.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.27.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.28.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.29.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.3.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.30.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.31.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.32.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.33.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.34.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.35.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.36.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.37.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.38.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.39.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.4.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.40.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.40.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.40.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.40.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.40.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.40.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.attn.k_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.attn.proj.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.attn.proj.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.attn.q_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.attn.qkv.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.ls1": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.ls2": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.41.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.attn.k_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.attn.proj.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.attn.proj.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.attn.q_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.attn.qkv.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.ls1": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.ls2": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.42.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.attn.k_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.attn.proj.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.attn.proj.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.attn.q_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.attn.qkv.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.ls1": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.ls2": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.43.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.attn.k_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.attn.proj.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.attn.proj.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.attn.q_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.attn.qkv.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.ls1": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.ls2": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.44.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.attn.k_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.attn.proj.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.attn.proj.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.attn.q_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.attn.qkv.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.ls1": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.ls2": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.45.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.attn.k_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.attn.proj.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.attn.proj.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.attn.q_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.attn.qkv.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.ls1": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.ls2": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.46.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.attn.k_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.attn.proj.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.attn.proj.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.attn.q_norm.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.attn.qkv.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.ls1": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.ls2": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.mlp.fc1.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.mlp.fc1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.mlp.fc2.bias": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.mlp.fc2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.norm1.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.47.norm2.weight": "pytorch_model-00002-of-00003.bin",
+    "vision_model.encoder.layers.5.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.5.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.6.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.7.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.8.norm2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.attn.k_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.attn.proj.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.attn.proj.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.attn.q_norm.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.attn.qkv.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.ls1": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.ls2": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.mlp.fc1.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.mlp.fc1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.mlp.fc2.bias": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.mlp.fc2.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.norm1.weight": "pytorch_model-00001-of-00003.bin",
+    "vision_model.encoder.layers.9.norm2.weight": "pytorch_model-00001-of-00003.bin"
+  }
+}

special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "<s>",
+  "eos_token": "</s>",
+  "pad_token": "[PAD]",
+  "unk_token": "<unk>"
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d967e855b1213a439df6c8ce2791f869c84b4f3b6cfacf22b86440b8192a2f8
+size 757972

tokenizer_config.json

@@ -0,0 +1,37 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": true,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "legacy": null,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "use_default_system_prompt": true,
+  "use_fast": false
+}