Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

added_tokens.json

@@ -0,0 +1,43 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|clip_time_end|>": 151674,
+  "<|clip_time_start|>": 151673,
+  "<|endoftext|>": 151643,
+  "<|fast_end|>": 151680,
+  "<|fast_start|>": 151679,
+  "<|fast_video_pad|>": 151678,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|frame|>": 151677,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|mm_pos_end|>": 151683,
+  "<|mm_pos_start|>": 151682,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|ocr_text_end|>": 151672,
+  "<|ocr_text_start|>": 151671,
+  "<|point_end|>": 151670,
+  "<|point_start|>": 151669,
+  "<|polygon_end|>": 151676,
+  "<|polygon_start|>": 151675,
+  "<|q_eos|>": 151681,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

chat_template.jinja

@@ -0,0 +1,7 @@
+{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system
+You are a helpful assistant.<|im_end|>
+{% endif %}<|im_start|>{{ message['role'] }}
+{% if message['content'] is string %}{{ message['content'] }}<|im_end|>
+{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>
+{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant
+{% endif %}

config.json

@@ -0,0 +1,102 @@
+{
+  "model_class": "KeyeARModel",
+  "qwen_config": {
+    "model_class": "UnifiedQwen3Model",
+    "vocab_size": 151936,
+    "embed_dim": 4096,
+    "num_layers": 36,
+    "tie_word_embeddings": false,
+    "hidden_act": "silu",
+    "num_heads": 32,
+    "num_kv_heads": 8,
+    "head_dim": 128,
+    "attn_dropout": 0.0,
+    "attention_function": "flash_attention_2",
+    "q_proj_bias": false,
+    "k_proj_bias": false,
+    "v_proj_bias": false,
+    "attention_bias": false,
+    "intermediate_dim": 12288,
+    "max_seq_len": 40960,
+    "rope_base": 1000000.0,
+    "rope_impl": "llama",
+    "rope_theta": 10000.0,
+    "rope_scaling": null,
+    "use_sliding_window": false,
+    "sliding_window": null,
+    "norm_eps": 1e-06,
+    "rms_norm_eps": 1e-06,
+    "q_norm": true,
+    "k_norm": true,
+    "eos_token_id": 151645,
+    "use_multimodal_rope": false,
+    "mrope_section": null,
+    "pre_embedding_size": null,
+    "pre_embedding_tokens": null,
+    "image_token_id": 151655,
+    "pad_token_id": null,
+    "q_eos_token": 151681,
+    "vision_start_token_id": 151652,
+    "vision_end_token_id": 151653,
+    "output_last_hidden_states_only": false,
+    "token_decoder_with_teacher_forcing": true,
+    "codebook_size": 65536,
+    "n_q_tokens": 8,
+    "skip_output_layer": false
+  },
+  "tokenizer_config": {
+    "model_class": "KeyeImageTokenizer",
+    "vision_config": {
+      "model_class": "KeyeVL1_5VisionModel",
+      "image_size": 384,
+      "patch_size": 14,
+      "num_channels": 3,
+      "hidden_size": 1152,
+      "num_hidden_layers": 27,
+      "num_attention_heads": 16,
+      "intermediate_size": 4304,
+      "hidden_act": "gelu_pytorch_tanh",
+      "max_seq_len": 4096,
+      "layer_norm_eps": 1e-06,
+      "attention_dropout": 0.0,
+      "has_learnable_position_embedding": true,
+      "use_qk_norm": false,
+      "qk_norm_eps": 1e-06,
+      "rope_theta": 10000.0,
+      "attention_function": "flash_attention_2",
+      "output_attentions": false,
+      "output_hidden_states": false
+    },
+    "codebook_size": 65536,
+    "embedding_dim": 128,
+    "init_embedding_dim": 4096,
+    "llm_hidden_size": 4096,
+    "n_q_tokens": 8,
+    "split_dim": false,
+    "split_voc": 8,
+    "add_voc_reducer": false,
+    "vq_sampling_mode": "argmin",
+    "vq_temperature": 1.0,
+    "vq_temperature_decay": 0.999,
+    "vq_min_temperature": 0.1,
+    "pre_llm_align": false,
+    "output_dim": 1024,
+    "fusion_type": "sum"
+  },
+  "token_decoder_config": {
+    "model_class": "UnifiedTokenDecoder",
+    "vocab_size": 65536,
+    "max_pos_length": 65537,
+    "max_length": 9,
+    "d_model": 512,
+    "eos_token": 151681,
+    "nhead": 4,
+    "num_layers": 1,
+    "dim_feedforward": 2048,
+    "use_gradient_checkpointing": true,
+    "input_dim": 4096,
+    "reduce": true,
+    "attention_function": "eager"
+  },
+  "__class__": "KeyeARConfig"
+}

image_processing_keye.py

@@ -0,0 +1,538 @@
+# coding=utf-8
+# Copyright 2024 The Qwen team, Alibaba Group 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.
+"""Image processor class for Qwen2-VL."""
+
+import math
+from typing import Dict, List, Optional, Union
+from PIL import Image
+
+import numpy as np
+import torch
+from transformers.image_processing_utils import BaseImageProcessor, BatchFeature
+from torchvision.transforms import functional as TF
+from transformers.image_transforms import (
+    convert_to_rgb,
+    resize,
+    to_channel_dimension_format,
+)
+from transformers.image_utils import (
+    OPENAI_CLIP_MEAN,
+    OPENAI_CLIP_STD,
+    ChannelDimension,
+    PILImageResampling,
+    get_image_size,
+    infer_channel_dimension_format,
+    is_scaled_image,
+    is_valid_image,
+    make_list_of_images,
+    to_numpy_array,
+    valid_images,
+    validate_preprocess_arguments,
+)
+from transformers.utils import TensorType, is_vision_available, logging
+
+
+ImageInput = Union[
+    "PIL.Image.Image", np.ndarray, "torch.Tensor", list["PIL.Image.Image"], list[np.ndarray], list["torch.Tensor"]
+]  # noqa
+
+VideoInput = Union[
+    list["PIL.Image.Image"],
+    "np.ndarray",
+    "torch.Tensor",
+    list["np.ndarray"],
+    list["torch.Tensor"],
+    list[list["PIL.Image.Image"]],
+    list[list["np.ndarrray"]],
+    list[list["torch.Tensor"]],
+]  # noqa
+logger = logging.get_logger(__name__)
+
+
+if is_vision_available():
+    from PIL import Image
+
+
+def make_batched_images(images) -> List[List[ImageInput]]:
+    """
+    Accepts images in list or nested list format, and makes a list of images for preprocessing.
+
+    Args:
+        images (`Union[List[List[ImageInput]], List[ImageInput], ImageInput]`):
+            The input image.
+
+    Returns:
+        list: A list of images.
+    """
+    if isinstance(images, (list, tuple)) and isinstance(images[0], (list, tuple)) and is_valid_image(images[0][0]):
+        return [img for img_list in images for img in img_list]
+
+    elif isinstance(images, (list, tuple)) and is_valid_image(images[0]):
+        return images
+
+    elif is_valid_image(images):
+        return [images]
+
+    raise ValueError(f"Could not make batched images from {images}")
+
+
+def adjust_size(size, patch_size):
+    num_patches = size // patch_size
+    if num_patches % 2 != 0:  # 如果是奇数，减1
+        num_patches -= 1
+    return num_patches * patch_size
+
+
+def make_batched_videos(videos) -> List[VideoInput]:
+    if isinstance(videos, (list, tuple)) and isinstance(videos[0], (list, tuple)) and is_valid_image(videos[0][0]):
+        return videos
+
+    elif isinstance(videos, (list, tuple)) and is_valid_image(videos[0]):
+        if isinstance(videos[0], Image.Image):
+            return [videos]
+        elif len(videos[0].shape) == 4:
+            return [list(video) for video in videos]
+
+    elif is_valid_image(videos) and len(videos.shape) == 4:
+        return [list(videos)]
+
+    raise ValueError(f"Could not make batched video from {videos}")
+
+
+def smart_resize(
+    height: int, width: int, factor: int = 28, min_pixels: int = 56 * 56, max_pixels: int = 14 * 14 * 4096
+):
+    """Rescales the image so that the following conditions are met:
+
+    1. Both dimensions (height and width) are divisible by 'factor'.
+
+    2. The total number of pixels is within the range ['min_pixels', 'max_pixels'].
+
+    3. The aspect ratio of the image is maintained as closely as possible.
+
+    """
+    #if height < factor or width < factor:
+    #    raise ValueError(f"height:{height} or width:{width} must be larger than factor:{factor}")
+    # if int(height < factor//4) + int(width < factor//4):
+    #     raise ValueError(f"height:{height} or width:{width} must be larger than factor:{factor//4}")
+
+    if height < factor:
+        print(f"smart_resize: height={height} < factor={factor}, reset height=factor")
+        width = round((width * factor) / height) 
+        height = factor
+
+    if width < factor:
+        print(f"smart_resize: width={width} < factor={factor}, reset width=factor")
+        height = round((height * factor) / width)
+        width = factor
+
+    if max(height, width) / min(height, width) > 200:
+        raise ValueError(
+            f"absolute aspect ratio must be smaller than 200, got {max(height, width) / min(height, width)}"
+        )
+    h_bar = round(height / factor) * factor
+    w_bar = round(width / factor) * factor
+    if h_bar * w_bar > max_pixels:
+        beta = math.sqrt((height * width) / max_pixels)
+        h_bar = math.floor(height / beta / factor) * factor
+        w_bar = math.floor(width / beta / factor) * factor
+    elif h_bar * w_bar < min_pixels:
+        beta = math.sqrt(min_pixels / (height * width))
+        h_bar = math.ceil(height * beta / factor) * factor
+        w_bar = math.ceil(width * beta / factor) * factor
+    return h_bar, w_bar
+
+
+class SiglipImageProcessor(BaseImageProcessor):
+    r"""
+    Constructs a Qwen2-VL image processor that dynamically resizes images based on the original images.
+
+    Args:
+        do_resize (`bool`, *optional*, defaults to `True`):
+            Whether to resize the image's (height, width) dimensions.
+        resample (`PILImageResampling`, *optional*, defaults to `Resampling.BICUBIC`):
+            Resampling filter to use when resizing the image.
+        do_rescale (`bool`, *optional*, defaults to `True`):
+            Whether to rescale the image by the specified scale `rescale_factor`.
+        rescale_factor (`int` or `float`, *optional*, defaults to `1/255`):
+            Scale factor to use if rescaling the image.
+        do_normalize (`bool`, *optional*, defaults to `True`):
+            Whether to normalize the image.
+        image_mean (`float` or `List[float]`, *optional*, defaults to `[0.48145466, 0.4578275, 0.40821073]`):
+            Mean to use if normalizing the image. This is a float or list of floats for each channel in the image.
+        image_std (`float` or `List[float]`, *optional*, defaults to `[0.26862954, 0.26130258, 0.27577711]`):
+            Standard deviation to use if normalizing the image. This is a float or list of floats for each channel in the image.
+        do_convert_rgb (`bool`, *optional*, defaults to `True`):
+            Whether to convert the image to RGB.
+        min_pixels (`int`, *optional*, defaults to `56 * 56`):
+            The min pixels of the image to resize the image.
+        max_pixels (`int`, *optional*, defaults to `28 * 28 * 1280`):
+            The max pixels of the image to resize the image.
+        patch_size (`int`, *optional*, defaults to 14):
+            The spacial patch size of the vision encoder.
+        temporal_patch_size (`int`, *optional*, defaults to 2):
+            The temporal patch size of the vision encoder.
+        merge_size (`int`, *optional*, defaults to 2):
+            The merge size of the vision encoder to llm encoder.
+    """
+
+    model_input_names = ["pixel_values", "image_grid_thw", "pixel_values_videos", "video_grid_thw"]
+
+    def __init__(
+        self,
+        do_resize: bool = True,
+        resample: PILImageResampling = PILImageResampling.BILINEAR,
+        do_rescale: bool = True,
+        rescale_factor: Union[int, float] = 1 / 255,
+        do_normalize: bool = True,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        do_convert_rgb: bool = True,
+        min_pixels: int = 56 * 56,
+        max_pixels: int = 28 * 28 * 1280,
+        patch_size: int = 14,
+        temporal_patch_size: int = 1,
+        merge_size: int = 2,
+        **kwargs,
+    ) -> None:
+        super().__init__(**kwargs)
+        self.do_resize = do_resize
+        self.resample = resample
+        self.do_rescale = do_rescale
+        self.rescale_factor = rescale_factor
+        self.do_normalize = do_normalize
+        self.image_mean = image_mean if image_mean is not None else OPENAI_CLIP_MEAN
+        self.image_std = image_std if image_std is not None else OPENAI_CLIP_STD
+        self.min_pixels = min_pixels
+        self.max_pixels = max_pixels
+        self.patch_size = patch_size
+        self.temporal_patch_size = temporal_patch_size
+        self.merge_size = merge_size
+        self.size = {"min_pixels": min_pixels, "max_pixels": max_pixels}
+        self.do_convert_rgb = do_convert_rgb
+        
+    def mvit_rescale(
+        self, image: Image.Image, merge_size: int = 2
+    ) -> Image.Image:
+        try:
+            w, h = image.size
+        except:
+            raise ValueError(str((type(image), image)))
+        patch_size = self.patch_size
+
+        if (w // patch_size) * (h // patch_size) > self.in_token_limit:
+            scale = math.sqrt(self.in_token_limit / ((w // patch_size) * (h // patch_size)))
+            new_w, new_h = int(w * scale), int(h * scale)
+            
+            image = image.resize((new_w, new_h), Image.Resampling.BILINEAR)
+        if self.pad_input:
+            new_w, new_h = image.size
+            pad_size_h = merge_size * patch_size
+            pad_size_w = merge_size * patch_size
+
+            pad_h = (pad_size_h - new_h % pad_size_h) % pad_size_h
+            pad_w = (pad_size_w - new_w % pad_size_w) % pad_size_w
+
+            image = TF.pad(image, (0, 0, pad_w, pad_h))
+        else:
+            new_w, new_h = image.size
+            new_w = new_w - new_w % patch_size
+            new_h = new_h - new_h % patch_size
+
+            new_w = adjust_size(new_w, patch_size)
+            new_h = adjust_size(new_h, patch_size)
+
+            image = TF.center_crop(image, (new_h, new_w))
+
+        w, h = image.size
+        if w // patch_size >= 512 or h // patch_size >= 512:
+            new_h = min(patch_size * 510, h)
+            new_w = min(patch_size * 510, w)
+            image = TF.center_crop(image, (new_h, new_w))
+            #raise ValueError("Exceed pos emb")
+        return image
+    def _preprocess(
+        self,
+        images: Union[ImageInput, VideoInput],
+        do_resize: bool = None,
+        size: Dict[str, int] = None,
+        resample: PILImageResampling = None,
+        do_rescale: bool = None,
+        rescale_factor: float = None,
+        do_normalize: bool = None,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        do_convert_rgb: bool = None,
+        data_format: Optional[ChannelDimension] = ChannelDimension.FIRST,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+    ):
+        """
+        Preprocess an image or batch of images. Copy of the `preprocess` method from `CLIPImageProcessor`.
+
+        Args:
+            images (`ImageInput`):
+                Image or batch of images to preprocess. Expects pixel values ranging from 0 to 255. If pixel values range from 0 to 1, set `do_rescale=False`.
+            vision_info (`List[Dict]`, *optional*):
+                Optional list of dictionaries containing additional information about vision inputs.
+            do_resize (`bool`, *optional*, defaults to `self.do_resize`):
+                Whether to resize the image.
+            resample (`PILImageResampling`, *optional*, defaults to `self.resample`):
+                Resampling filter to use if resizing the image. This can be one of the `PILImageResampling` enums.
+            do_rescale (`bool`, *optional*, defaults to `self.do_rescale`):
+                Whether to rescale the image.
+            rescale_factor (`float`, *optional*, defaults to `self.rescale_factor`):
+                Scale factor to use if rescaling the image.
+            do_normalize (`bool`, *optional*, defaults to `self.do_normalize`):
+                Whether to normalize the image.
+            image_mean (`float` or `List[float]`, *optional*, defaults to `self.image_mean`):
+                Mean to use if normalizing the image. Can be a float or a list of floats corresponding to the number of channels in the image.
+            image_std (`float` or `List[float]`, *optional*, defaults to `self.image_std`):
+                Standard deviation to use if normalizing the image. Can be a float or a list of floats corresponding to the number of channels in the image.
+            do_convert_rgb (`bool`, *optional*, defaults to `self.do_convert_rgb`):
+                Whether to convert the image to RGB.
+            data_format (`ChannelDimension`, *optional*, defaults to `ChannelDimension.FIRST`):
+                The channel dimension format for the output image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+                - Unset: Use the channel dimension format of the input image.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format for the input image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+                - `"none"` or `ChannelDimension.NONE`: image in (height, width) format.   - `"none"` or `ChannelDimension.NONE`: image in (height, width) format.
+        """
+        images = make_list_of_images(images)
+
+        if do_convert_rgb:
+            images = [convert_to_rgb(image) for image in images]
+
+        # All transformations expect numpy arrays.
+        images = [to_numpy_array(image) for image in images]
+
+        if is_scaled_image(images[0]) and do_rescale:
+            logger.warning_once(
+                "It looks like you are trying to rescale already rescaled images. If the input"
+                " images have pixel values between 0 and 1, set `do_rescale=False` to avoid rescaling them again."
+            )
+        if input_data_format is None:
+            # We assume that all images have the same channel dimension format.
+            input_data_format = infer_channel_dimension_format(images[0])
+
+        height, width = get_image_size(images[0], channel_dim=input_data_format)
+        resized_height, resized_width = height, width
+        processed_images = []
+        for image in images:
+            # image = self.mvit_rescale(image, merge_size=self.merge_size)
+            if do_resize:
+                if size is not None and "height" in size.keys():
+                    resized_height, resized_width = size["height"], size["width"]
+                else:
+                    resized_height, resized_width = smart_resize(
+                        height,
+                        width,
+                        factor=self.patch_size * self.merge_size,
+                        min_pixels=self.min_pixels,
+                        max_pixels=self.max_pixels,
+                    )
+                image = resize(
+                    image, size=(resized_height, resized_width), resample=resample, input_data_format=input_data_format
+                )
+
+            if do_rescale:
+                image = self.rescale(image, scale=rescale_factor, input_data_format=input_data_format)
+
+            if do_normalize:
+                image = self.normalize(
+                    image=image, mean=image_mean, std=image_std, input_data_format=input_data_format
+                )
+
+            image = to_channel_dimension_format(image, data_format, input_channel_dim=input_data_format)
+            processed_images.append(image)
+
+        patches = np.array(processed_images)
+        if data_format == ChannelDimension.LAST:
+            patches = patches.transpose(0, 3, 1, 2)
+        if patches.shape[0] == 1:
+            patches = np.tile(patches, (self.temporal_patch_size, 1, 1, 1))
+        init_patches = patches
+        channel = patches.shape[1]
+        grid_t = patches.shape[0] // self.temporal_patch_size
+        grid_h, grid_w = resized_height // self.patch_size, resized_width // self.patch_size
+        patches = patches.reshape(
+            grid_t,
+            self.temporal_patch_size,
+            channel,
+            grid_h,
+            self.patch_size,
+            grid_w,
+            self.patch_size,
+        )
+        patches = patches.transpose(0, 3, 5, 2, 1, 4, 6)
+        assert self.temporal_patch_size == 1
+        flatten_patches = patches.reshape(
+            grid_t * grid_h * grid_w, channel, self.patch_size, self.patch_size
+        )
+        return flatten_patches, (grid_t, grid_h, grid_w)
+
+    def preprocess(
+        self,
+        images: ImageInput,
+        videos: VideoInput = None,
+        do_resize: bool = None,
+        size: Dict[str, int] = None,
+        resample: PILImageResampling = None,
+        do_rescale: bool = None,
+        rescale_factor: float = None,
+        do_normalize: bool = None,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        do_convert_rgb: bool = None,
+        return_tensors: Optional[Union[str, TensorType]] = None,
+        data_format: Optional[ChannelDimension] = ChannelDimension.FIRST,
+        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+    ):
+        """
+        Args:
+            images (`ImageInput`):
+                Image to preprocess. Expects a single or batch of images with pixel values ranging from 0 to 255. If
+                passing in images with pixel values between 0 and 1, set `do_rescale=False`.
+            videos (`VideoInput`):
+                Video to preprocess. Expects a single or batch of videos with pixel values ranging from 0 to 255. If
+                passing in videos with pixel values between 0 and 1, set `do_rescale=False`.
+            do_resize (`bool`, *optional*, defaults to `self.do_resize`):
+                Whether to resize the image.
+            size (`Dict[str, int]`, *optional*, defaults to `self.size`):
+                Size of the image after resizing. Shortest edge of the image is resized to size["shortest_edge"], with
+                the longest edge resized to keep the input aspect ratio.
+            resample (`int`, *optional*, defaults to `self.resample`):
+                Resampling filter to use if resizing the image. This can be one of the enum `PILImageResampling`. Only
+                has an effect if `do_resize` is set to `True`.
+            do_rescale (`bool`, *optional*, defaults to `self.do_rescale`):
+                Whether to rescale the image.
+            rescale_factor (`float`, *optional*, defaults to `self.rescale_factor`):
+                Rescale factor to rescale the image by if `do_rescale` is set to `True`.
+            do_normalize (`bool`, *optional*, defaults to `self.do_normalize`):
+                Whether to normalize the image.
+            image_mean (`float` or `List[float]`, *optional*, defaults to `self.image_mean`):
+                Image mean to use for normalization. Only has an effect if `do_normalize` is set to `True`.
+            image_std (`float` or `List[float]`, *optional*, defaults to `self.image_std`):
+                Image standard deviation to use for normalization. Only has an effect if `do_normalize` is set to
+                `True`.
+            do_convert_rgb (`bool`, *optional*, defaults to `self.do_convert_rgb`):
+                Whether to convert the image to RGB.
+            return_tensors (`str` or `TensorType`, *optional*):
+                The type of tensors to return. Can be one of:
+                - Unset: Return a list of `np.ndarray`.
+                - `TensorType.TENSORFLOW` or `'tf'`: Return a batch of type `tf.Tensor`.
+                - `TensorType.PYTORCH` or `'pt'`: Return a batch of type `torch.Tensor`.
+                - `TensorType.NUMPY` or `'np'`: Return a batch of type `np.ndarray`.
+                - `TensorType.JAX` or `'jax'`: Return a batch of type `jax.numpy.ndarray`.
+            data_format (`ChannelDimension` or `str`, *optional*, defaults to `ChannelDimension.FIRST`):
+                The channel dimension format for the output image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+                - Unset: Use the channel dimension format of the input image.
+            input_data_format (`ChannelDimension` or `str`, *optional*):
+                The channel dimension format for the input image. If unset, the channel dimension format is inferred
+                from the input image. Can be one of:
+                - `"channels_first"` or `ChannelDimension.FIRST`: image in (num_channels, height, width) format.
+                - `"channels_last"` or `ChannelDimension.LAST`: image in (height, width, num_channels) format.
+                - `"none"` or `ChannelDimension.NONE`: image in (height, width) format.
+
+        """
+        do_resize = do_resize if do_resize is not None else self.do_resize
+        size = size if size is not None else self.size
+        resample = resample if resample is not None else self.resample
+        do_rescale = do_rescale if do_rescale is not None else self.do_rescale
+        rescale_factor = rescale_factor if rescale_factor is not None else self.rescale_factor
+        do_normalize = do_normalize if do_normalize is not None else self.do_normalize
+        image_mean = image_mean if image_mean is not None else self.image_mean
+        image_std = image_std if image_std is not None else self.image_std
+        do_convert_rgb = do_convert_rgb if do_convert_rgb is not None else self.do_convert_rgb
+
+        if images is not None:
+            images = make_batched_images(images)
+        if videos is not None:
+            videos = make_batched_videos(videos)
+
+        if images is not None and not valid_images(images):
+            raise ValueError(
+                "Invalid image type. Must be of type PIL.Image.Image, numpy.ndarray, "
+                "torch.Tensor, tf.Tensor or jax.ndarray."
+            )
+
+        validate_preprocess_arguments(
+            rescale_factor=rescale_factor,
+            do_normalize=do_normalize,
+            image_mean=image_mean,
+            image_std=image_std,
+            do_resize=do_resize,
+            size=size,
+            resample=resample,
+        )
+
+        if images is not None:
+            pixel_values, vision_grid_thws = [], []
+            for image in images:
+                patches, image_grid_thw = self._preprocess(
+                    image,
+                    do_resize=do_resize,
+                    size = size,
+                    resample=resample,
+                    do_rescale=do_rescale,
+                    rescale_factor=rescale_factor,
+                    do_normalize=do_normalize,
+                    image_mean=image_mean,
+                    image_std=image_std,
+                    data_format=data_format,
+                    do_convert_rgb=do_convert_rgb,
+                    input_data_format=input_data_format,
+                )
+                pixel_values.extend(patches)
+                vision_grid_thws.append(image_grid_thw)
+            pixel_values = np.array(pixel_values)
+            vision_grid_thws = np.array(vision_grid_thws)
+            data = {"pixel_values": pixel_values, "image_grid_thw": vision_grid_thws}
+
+        if videos is not None:
+            pixel_values, vision_grid_thws = [], []
+            for images in videos:
+                patches, video_grid_thw = self._preprocess(
+                    images,
+                    do_resize=do_resize,
+                    size = size,
+                    resample=resample,
+                    do_rescale=do_rescale,
+                    rescale_factor=rescale_factor,
+                    do_normalize=do_normalize,
+                    image_mean=image_mean,
+                    image_std=image_std,
+                    data_format=data_format,
+                    do_convert_rgb=do_convert_rgb,
+                    input_data_format=input_data_format,
+                )
+                pixel_values.extend(patches)
+                vision_grid_thws.append(video_grid_thw)
+            pixel_values = np.array(pixel_values)
+            vision_grid_thws = np.array(vision_grid_thws)
+            data = {"pixel_values_videos": pixel_values, "video_grid_thw": vision_grid_thws}
+
+        return BatchFeature(data=data, tensor_type=return_tensors)
+

model-00001-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb83224153d4490234c7399b45634c0718a9cb8b88cb19d249ff295c98a7f06c
+size 5338465192

model-00002-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2ed50bc53f931390ad69a8c40ce1e93223e43f75d05ae3639bbdef55638739c
+size 5360574528

model-00003-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20a87e78a2e4216952b97b799d9d11b2e1366556d620f689d28f0359a77a02a4
+size 4974687872

model-00004-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f35d39f9d2e13d4fde84d15048038c73d476b6acab4da45fc013e42575e8261d
+size 5349379136

model-00005-of-00005.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e6652ec6dd3138d5a8a5deb9098053699bf1a891da54ec216c18294dc8c5bc4
+size 3593236272

model.safetensors.index.json

@@ -0,0 +1,939 @@
+{
+  "metadata": {
+    "total_size": 24616227552
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.0.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.1.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.10.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.11.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.12.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.13.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.14.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.15.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.mlp.w1.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.mlp.w2.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.mlp.w3.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.mlp_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.16.sa_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.17.attn.k_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.17.attn.k_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.17.attn.output_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.17.attn.q_norm.scale": "model-00001-of-00005.safetensors",
+    "model.model.layers.17.attn.q_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.17.attn.v_proj.weight": "model-00001-of-00005.safetensors",
+    "model.model.layers.17.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.17.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.17.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.17.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.17.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.18.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.19.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.2.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.20.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.21.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.22.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.23.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.24.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.25.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.26.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.27.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.28.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.attn.q_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.attn.v_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.mlp.w1.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.mlp.w2.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.mlp.w3.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.mlp_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.29.sa_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.3.attn.k_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.3.attn.k_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.3.attn.output_proj.weight": "model-00002-of-00005.safetensors",
+    "model.model.layers.3.attn.q_norm.scale": "model-00002-of-00005.safetensors",
+    "model.model.layers.3.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.3.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.3.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.3.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.3.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.3.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.3.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.30.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.31.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.32.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.33.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.34.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.35.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.4.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.5.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.6.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.7.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.8.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.attn.k_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.attn.k_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.attn.output_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.attn.q_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.attn.q_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.attn.v_proj.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.mlp.w1.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.mlp.w2.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.mlp.w3.weight": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.mlp_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.layers.9.sa_norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.norm.scale": "model-00003-of-00005.safetensors",
+    "model.model.output.weight": "model-00004-of-00005.safetensors",
+    "model.model.tok_embeddings.embed_tokens.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.final_norm.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.final_norm.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.input_linear.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.input_linear.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.output_linear.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.output_linear.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.position_embedding.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.k_proj.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.k_proj.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.output_proj.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.output_proj.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.q_proj.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.q_proj.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.v_proj.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.attn.v_proj.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.mlp.w1.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.mlp.w2.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.mlp.w3.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.mlp_norm.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.mlp_norm.weight": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.sa_norm.bias": "model-00004-of-00005.safetensors",
+    "model.model.token_head.transformer.layers.0.sa_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.encoder.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.encoder.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.mlp_AR.linear_1.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.mlp_AR.linear_1.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.mlp_AR.linear_2.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.mlp_AR.linear_2.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.mlp_AR.pre_norm.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.mlp_AR.pre_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.0.embedding.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.0.embedding_proj.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.0.embedding_proj.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.0.q_norm.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.0.q_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.0.z_norm.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.0.z_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.1.embedding.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.1.embedding_proj.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.1.embedding_proj.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.1.q_norm.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.1.q_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.1.z_norm.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.1.z_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.2.embedding.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.2.embedding_proj.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.2.embedding_proj.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.2.q_norm.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.2.q_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.2.z_norm.bias": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.2.z_norm.weight": "model-00004-of-00005.safetensors",
+    "visual_tokenizer.quantizer.3.embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.3.embedding_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.3.embedding_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.3.q_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.3.q_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.3.z_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.3.z_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.4.embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.4.embedding_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.4.embedding_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.4.q_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.4.q_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.4.z_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.4.z_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.5.embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.5.embedding_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.5.embedding_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.5.q_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.5.q_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.5.z_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.5.z_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.6.embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.6.embedding_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.6.embedding_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.6.q_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.6.q_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.6.z_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.6.z_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.7.embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.7.embedding_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.7.embedding_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.7.q_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.7.q_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.7.z_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.quantizer.7.z_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.0.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.3.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.4.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.5.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.6.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.up_projectors.7.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.embeddings.packing_position_embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.embeddings.patch_embedding.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.embeddings.patch_embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.embeddings.position_embedding.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.0.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.1.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.10.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.11.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.12.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.13.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.14.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.15.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.16.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.17.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.18.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.19.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.2.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.20.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.21.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.22.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.23.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.24.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.25.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.26.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.3.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.4.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.5.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.6.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.7.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.8.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.k_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.k_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.output_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.output_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.q_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.q_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.v_proj.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.attn.v_proj.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.mlp.w1.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.mlp.w1.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.mlp.w2.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.mlp.w2.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.mlp_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.mlp_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.sa_norm.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.encoder.layers.9.sa_norm.weight": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.ln_post.bias": "model-00005-of-00005.safetensors",
+    "visual_tokenizer.visual.ln_post.weight": "model-00005-of-00005.safetensors"
+  }
+}

preprocessor_config.json

@@ -0,0 +1,33 @@
+{
+  "auto_map": {
+    "AutoImageProcessor": "image_processing_keye.SiglipImageProcessor",
+    "AutoProcessor": "processing_keye.KeyeProcessor"
+  },
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "image_processor_type": "SiglipImageProcessor",
+  "image_std": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "max_pixels": 16056320,
+  "merge_size": 2,
+  "min_pixels": 3136,
+  "patch_size": 14,
+  "processor_class": "KeyeProcessor",
+  "resample": 2,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "max_pixels": 16056320,
+    "min_pixels": 3136
+  },
+  "temporal_patch_size": 1
+}

processing_keye.py

@@ -0,0 +1,391 @@
+
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/Keye/modular_Keye.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_Keye.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# coding=utf-8
+# Copyright 2025 The Qwen Team 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.
+from typing import List, Union
+import numpy as np
+from transformers.feature_extraction_utils import BatchFeature
+#from transformers.image_utils import ImageInput, VideoInput
+from transformers.processing_utils import ProcessingKwargs, ProcessorMixin, Unpack, VideosKwargs
+from transformers.tokenization_utils_base import PreTokenizedInput, TextInput
+from .image_processing_keye import SiglipImageProcessor
+import torch
+from itertools import chain
+
+
+ImageInput = Union[
+    "PIL.Image.Image", np.ndarray, "torch.Tensor", list["PIL.Image.Image"], list[np.ndarray], list["torch.Tensor"]
+]  # noqa
+
+VideoInput = Union[
+    list["PIL.Image.Image"],
+    "np.ndarray",
+    "torch.Tensor",
+    list["np.ndarray"],
+    list["torch.Tensor"],
+    list[list["PIL.Image.Image"]],
+    list[list["np.ndarrray"]],
+    list[list["torch.Tensor"]],
+]  # noqa
+
+
+class KeyeVideosProcessorKwargs(VideosKwargs, total=False):
+    fps: Union[List[float], float]
+
+
+class KeyeProcessorKwargs(ProcessingKwargs, total=False):
+    videos_kwargs: KeyeVideosProcessorKwargs
+    _defaults = {
+        "text_kwargs": {
+            "padding": False,
+        },
+        "videos_kwargs": {"fps": 2.0},
+    }
+
+
+class KeyeProcessor(ProcessorMixin):
+    r"""
+    [`KeyeProcessor`] offers all the functionalities of [`SiglipImageProcessor`] and [`Qwen2TokenizerFast`]. See the
+    [`~KeyeProcessor.__call__`] and [`~KeyeProcessor.decode`] for more information.
+    Args:
+        image_processor ([`SiglipImageProcessor`], *optional*):
+            The image processor is a required input.
+        tokenizer ([`Qwen2TokenizerFast`], *optional*):
+            The tokenizer is a required input.
+        chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages
+            in a chat into a tokenizable string.
+    """
+
+    attributes = ["image_processor", "tokenizer"]
+    valid_kwargs = ["chat_template","image_std", "min_pixels", "image_mean", "merge_size", "image_processor_type", "temporal_patch_size", "patch_size", "max_pixels"]
+
+    image_processor_class = "AutoImageProcessor"
+    tokenizer_class = ("Qwen2Tokenizer", "Qwen2TokenizerFast")
+
+    def __init__(self, image_processor=None, tokenizer=None, chat_template=None, **kwargs):
+        self.image_token = "<|image_pad|>" if not hasattr(tokenizer, "image_token") else tokenizer.image_token
+        self.vision_start_token = "<|vision_start|>" if not hasattr(tokenizer, "vision_start_token") else tokenizer.vision_start_token
+        self.video_token = "<|video_pad|>" if not hasattr(tokenizer, "video_token") else tokenizer.video_token
+        self.frame_token = "<|frame|>" if not hasattr(tokenizer, "frame_token") else tokenizer.frame_token
+        self.fast_video_token = "<|fast_video_pad|>" if not hasattr(tokenizer, "fast_video_token") else tokenizer.fast_video_token
+        self.fast_start = "<|fast_start|>" if not hasattr(tokenizer, "fast_start") else tokenizer.fast_start
+        self.fast_end = "<|fast_end|>" if not hasattr(tokenizer, "fast_end") else tokenizer.fast_end
+        self.image_info_tag = "<|image_info|>" if not hasattr(tokenizer, "image_info_tag") else tokenizer.image_info_tag
+        super().__init__(image_processor, tokenizer, chat_template=chat_template)
+
+        # self.fast_patch_size = 16
+        # self.fast_image_processor = SiglipImageProcessor(patch_size=self.fast_patch_size)
+        self.slowfast = True
+
+    def __call__(
+        self,
+        images: ImageInput = None,
+        text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
+        videos: VideoInput = None,
+        get_resolution_from_cropped_image_size = None,
+        **kwargs: Unpack[KeyeProcessorKwargs],
+    ) -> BatchFeature:
+        """
+        Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
+        and `kwargs` arguments to Qwen2TokenizerFast's [`~Qwen2TokenizerFast.__call__`] if `text` is not `None` to encode
+        the text. To prepare the vision inputs, this method forwards the `vision_infos` and `kwrags` arguments to
+        SiglipImageProcessor's [`~SiglipImageProcessor.__call__`] if `vision_infos` is not `None`.
+
+        Args:
+            images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`):
+                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
+                tensor. Both channels-first and channels-last formats are supported.
+            text (`str`, `List[str]`, `List[List[str]]`):
+                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
+                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
+                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
+            videos (`np.ndarray`, `torch.Tensor`, `List[np.ndarray]`, `List[torch.Tensor]`):
+                The image or batch of videos to be prepared. Each video can be a 4D NumPy array or PyTorch
+                tensor, or a nested list of 3D frames. Both channels-first and channels-last formats are supported.
+            return_tensors (`str` or [`~utils.TensorType`], *optional*):
+                If set, will return tensors of a particular framework. Acceptable values are:
+                - `'tf'`: Return TensorFlow `tf.constant` objects.
+                - `'pt'`: Return PyTorch `torch.Tensor` objects.
+                - `'np'`: Return NumPy `np.ndarray` objects.
+                - `'jax'`: Return JAX `jnp.ndarray` objects.
+
+        Returns:
+            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
+
+            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
+            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
+              `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
+              `None`).
+            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
+            - **pixel_values_videos** -- Pixel values of videos to be fed to a model. Returned when `videos` is not `None`.
+            - **image_grid_thw** -- List of image 3D grid in LLM. Returned when `images` is not `None`.
+            - **video_grid_thw** -- List of video 3D grid in LLM. Returned when `videos` is not `None`.
+            - **second_per_grid_ts** -- List of video seconds per time grid. Returned when `videos` is not `None`.
+        """
+        output_kwargs = self._merge_kwargs(
+            KeyeProcessorKwargs,
+            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
+            **kwargs,
+        )
+        if images is not None:
+            slow_images = images
+            image_inputs = self.image_processor(images=slow_images, return_tensors="pt")
+            image_inputs['pixel_values'] = image_inputs['pixel_values']
+            image_grid_thw = image_inputs["image_grid_thw"]
+        else:
+            image_inputs = {}
+            image_grid_thw = None
+
+
+        if videos is not None:
+            #TODO: add video processing
+            all_slow_videos = []
+            all_fast_videos = []
+            # 这个是因为视频会划分为多张图片，需要在这个地方提前统计好token量，后面就不清楚界限在哪了
+            slow_videos_token_nums = [[] for i in range(len(videos))]
+            fast_videos_token_nums = [[] for i in range(len(videos))]
+            all_position = []
+
+            for current_index, current_video in enumerate(videos):
+                if len(current_video) == 4: # slow_frames, fast_frames, time_position, slow_fast_order, 这里需要注意的是fast_frames，有可能和slow的长度不等，需要靠slow_fast_order来进行识别
+                    slow_frames, fast_frames, time_position, slow_fast_order = current_video[0], current_video[1], current_video[2], current_video[3]
+                    all_position.append((time_position, slow_fast_order))
+                    ####### slow part #########
+                    if slow_frames is not None:
+                        slow_videos_inputs = self.image_processor(images=None, videos=slow_frames, **output_kwargs["images_kwargs"])
+                        slow_video_grid_thw = slow_videos_inputs["video_grid_thw"]
+                        all_slow_videos.append(slow_videos_inputs)
+                        slow_videos_token_nums[current_index] = slow_video_grid_thw.prod(dim=1).tolist() # 当前这个视频的所有token数
+                    else:
+                        all_slow_videos.append(None) # 这样的话，slow_fast_order都是1了，这里应该不会用到的
+                        slow_videos_token_nums[current_index] = None # 如果全为fast？但目前不存在这种情况
+                    ###########################
+
+                    ####### fast part #########
+                    if self.slowfast:
+                        if fast_frames is not None:
+                            fast_videos_inputs = self.image_processor(images=None, videos=fast_frames, **output_kwargs["images_kwargs"])
+                            fast_video_grid_thw = fast_videos_inputs["video_grid_thw"]
+                            all_fast_videos.append(fast_videos_inputs)
+                            fast_videos_token_nums[current_index] = fast_video_grid_thw.prod(dim=1).tolist() # 当前这个视频的fast的所有token数
+                        else:
+                            all_fast_videos.append(None) # 如果全为slow
+                            fast_videos_token_nums[current_index] = None
+                    ###########################
+                else:
+                    slow_frames, fast_frames, slow_fast_order = current_video[0], current_video[1], current_video[2]
+                    if kwargs.get("image_video_pad", False):
+                        fast_frames = slow_frames
+                        slow_fast_order += [1]
+
+                    all_position.append((None, slow_fast_order))
+                    ####### slow part #########
+                    if slow_frames is not None:
+                        for each_image in slow_frames:
+                            if kwargs.get("image_video_pad", False):
+                                slow_videos_inputs = self.image_processor.preprocess(images=None, videos=[each_image], size = {"height": 28, "width": 28}, **output_kwargs["images_kwargs"])
+                            else:
+                                slow_videos_inputs = self.image_processor(images=None, videos=[each_image], **output_kwargs["images_kwargs"])
+                            slow_video_grid_thw = slow_videos_inputs["video_grid_thw"]
+
+                            all_slow_videos.append(slow_videos_inputs)
+                            slow_videos_token_nums[current_index].append(slow_video_grid_thw.prod(dim=1).item()) # 这里因为没在前面split开，所以要这么写
+                        ###########################
+                    else:
+                        all_slow_videos.append(None) # 这样的话，slow_fast_order都是1了，这里应该不会用到的
+                        slow_videos_token_nums[current_index] = None # 如果全为fast？但目前不存在这种情况
+
+                    ####### fast part #########
+                    if self.slowfast:
+                        if fast_frames is not None:
+                            for each_image in fast_frames:
+                                if kwargs.get("image_video_pad", False):
+                                    fast_videos_inputs = self.image_processor.preprocess(images=None, videos=[each_image], size = {"height": 28, "width": 28}, **output_kwargs["images_kwargs"])
+                                else:
+                                    fast_videos_inputs = self.image_processor.preprocess(images=None, videos=[each_image], **output_kwargs["images_kwargs"])
+                                fast_video_grid_thw = fast_videos_inputs["video_grid_thw"]
+
+                                all_fast_videos.append(fast_videos_inputs)
+                                fast_videos_token_nums[current_index].append(fast_video_grid_thw.prod(dim=1).item())
+                        else:
+                            all_fast_videos.append(None)
+                            fast_videos_token_nums[current_index] = None
+                    ###########################
+
+
+            # todo: zdj debug 多次concat速度会慢很多
+            slow_pixel_values_videos_list = [single_slow_video["pixel_values_videos"] for single_slow_video in all_slow_videos if single_slow_video is not None]
+            slow_video_grid_thw_list = [single_slow_video["video_grid_thw"] for single_slow_video in all_slow_videos if single_slow_video is not None]
+
+            total_slow_pixel_values_videos = torch.concat(slow_pixel_values_videos_list, dim=0)
+            total_slow_video_grid_thw = torch.concat(slow_video_grid_thw_list, dim=0)
+            # todo: zdj debug end
+
+            if len(total_slow_pixel_values_videos):
+                videos_inputs = {
+                    "pixel_values_videos": total_slow_pixel_values_videos,
+                    "video_grid_thw": total_slow_video_grid_thw,
+                }
+                video_grid_thw = videos_inputs["video_grid_thw"]
+            else:
+                videos_inputs = {}
+                video_grid_thw = None
+
+            if self.slowfast:
+                # todo: zdj debug 多次concat速度会慢很多
+                fast_pixel_values_videos_list = [single_fast_video["pixel_values_videos"] for single_fast_video in all_fast_videos if single_fast_video is not None]
+                fast_video_grid_thw_list = [single_fast_video["video_grid_thw"] for single_fast_video in all_fast_videos if single_fast_video is not None]
+                # fast_second_per_grid_ts = torch.tensor(list(chain(*fast_second_per_grid_ts_list)))
+
+                if len(fast_pixel_values_videos_list):
+                    videos_inputs["fast_pixel_values_videos"] = torch.concat(fast_pixel_values_videos_list, dim=0)
+                    videos_inputs["fast_video_grid_thw"] = torch.concat(fast_video_grid_thw_list, dim=0)
+                    fast_video_grid_thw = videos_inputs["fast_video_grid_thw"]
+                else:
+                    fast_video_grid_thw = None
+
+                # todo: zdj debug end
+        else:
+            videos_inputs = {}
+            video_grid_thw = None
+            fast_video_grid_thw = None
+
+
+        if not isinstance(text, list):
+            text = [text]
+
+        if image_grid_thw is not None:
+            index = 0
+            for i in range(len(text)):
+                while self.image_token in text[i]:
+                    image_downsample_ratio = self.image_processor.merge_size * self.image_processor.patch_size
+                    
+                    _, h_merged, w_merged = image_grid_thw[index]// self.image_processor.merge_size
+                    image_place_holder_tempale = f"{image_downsample_ratio*h_merged.item()},{image_downsample_ratio*w_merged.item()}"
+                    if get_resolution_from_cropped_image_size is not None:
+                        raise NotImplementedError
+
+                    image_place_holder_tempale = ""
+
+                    for i_h in range(h_merged.item()):
+                        image_place_holder_tempale += "<|mm_pos_start|>" + f"{i_h},{w_merged}" + "<|mm_pos_end|>" + "<|placeholder|>" * w_merged
+                    
+                    
+                    text[i] = text[i].replace(
+                        self.image_token,
+                        image_place_holder_tempale,
+                        1,
+                    )
+                    index += 1
+                text[i] = text[i].replace("<|placeholder|>", self.image_token)
+            # text[0].count("<|placeholder|>")
+
+        if video_grid_thw is not None or fast_video_grid_thw is not None:
+            index = 0
+            for i in range(len(text)):
+                while self.video_token in text[i]:
+                    video_place_holder_tempale = ""
+                    slow_index = 0
+                    fast_index = 0
+                    for j in range(len(all_position[index][1])):
+                        if all_position[index][0] is not None: # 如果有时间戳
+                            video_place_holder_tempale += self.frame_token + format(all_position[index][0][j], ".1f")
+                        else:
+                            video_place_holder_tempale += self.frame_token
+                        
+                        if all_position[index][1][j] == 0: # 当前帧是slow？
+                            video_place_holder_tempale += "<|placeholder|>" * (slow_videos_token_nums[index][slow_index]//self.image_processor.merge_size//self.image_processor.merge_size)
+                            slow_index += 1
+                        elif all_position[index][1][j] == 1: # 当前帧是fast？
+                            video_place_holder_tempale += self.fast_start + "<|fast_placeholder|>" * (fast_videos_token_nums[index][fast_index]//self.image_processor.merge_size//self.image_processor.merge_size) + self.fast_end
+                            fast_index += 1
+                    text[i] = text[i].replace(
+                        self.video_token,
+                        video_place_holder_tempale,
+                        1,
+                    )
+                    index += 1
+                #  self.tokenizer.decode(191678)
+                # self.tokenizer.encode("<|fast_video_pad|>")
+                text[i] = text[i].replace("<|placeholder|>", self.video_token)
+                text[i] = text[i].replace("<|fast_placeholder|>", self.fast_video_token)
+            # text[0].count(self.video_token)
+        text_inputs = self.tokenizer(text, **output_kwargs["text_kwargs"])
+
+        return BatchFeature(data={**text_inputs, **image_inputs, **videos_inputs})
+
+    def batch_decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to Qwen2TokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
+        refer to the docstring of this method for more information.
+        """
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    def decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to Qwen2TokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
+        the docstring of this method for more information.
+        """
+        return self.tokenizer.decode(*args, **kwargs)
+
+    def post_process_image_text_to_text(
+        self, generated_outputs, skip_special_tokens=True, clean_up_tokenization_spaces=False, **kwargs
+    ):
+        """
+        Post-process the output of the model to decode the text.
+
+        Args:
+            generated_outputs (`torch.Tensor` or `np.ndarray`):
+                The output of the model `generate` function. The output is expected to be a tensor of shape `(batch_size, sequence_length)`
+                or `(sequence_length,)`.
+            skip_special_tokens (`bool`, *optional*, defaults to `True`):
+                Whether or not to remove special tokens in the output. Argument passed to the tokenizer's `batch_decode` method.
+            Clean_up_tokenization_spaces (`bool`, *optional*, defaults to `False`):
+                Whether or not to clean up the tokenization spaces. Argument passed to the tokenizer's `batch_decode` method.
+            **kwargs:
+                Additional arguments to be passed to the tokenizer's `batch_decode method`.
+
+        Returns:
+            `List[str]`: The decoded text.
+        """
+        return self.tokenizer.batch_decode(
+            generated_outputs,
+            skip_special_tokens=skip_special_tokens,
+            clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+            **kwargs,
+        )
+
+    @property
+    def model_input_names(self):
+        tokenizer_input_names = self.tokenizer.model_input_names
+        image_processor_input_names = self.image_processor.model_input_names
+        names_from_processor = list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
+        return names_from_processor + ["second_per_grid_ts"]
+
+
+
+__all__ = ["KeyeProcessor", "KeyeProcessor_moonvit", "KeyeProcessor"]
+
+

processor_config.json

@@ -0,0 +1,6 @@
+{
+  "auto_map": {
+    "AutoProcessor": "processing_keye.KeyeProcessor"
+  },
+  "processor_class": "KeyeProcessor"
+}

special_tokens_map.json

@@ -0,0 +1,25 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03fa0e33a395d9d2f682ab62d137f0332be852db8bbddf188ef233bb9d8e7f59
+size 11425552

tokenizer_config.json

@@ -0,0 +1,357 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151669": {
+      "content": "<|point_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151670": {
+      "content": "<|point_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151671": {
+      "content": "<|ocr_text_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151672": {
+      "content": "<|ocr_text_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151673": {
+      "content": "<|clip_time_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151674": {
+      "content": "<|clip_time_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151675": {
+      "content": "<|polygon_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151676": {
+      "content": "<|polygon_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151677": {
+      "content": "<|frame|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151678": {
+      "content": "<|fast_video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151679": {
+      "content": "<|fast_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151680": {
+      "content": "<|fast_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151681": {
+      "content": "<|q_eos|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151682": {
+      "content": "<|mm_pos_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151683": {
+      "content": "<|mm_pos_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "auto_map": {
+    "AutoProcessor": "processing_keye.KeyeProcessor"
+  },
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "processor_class": "KeyeProcessor",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}