Upload Keye-VL 1.5 8B MLX 4-bit quantized model

.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 +1,40 @@
+{
+  "</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,
+  "<|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,
+  "<|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 %}

chat_template.json

@@ -0,0 +1,3 @@
+{
+    "chat_template": "{% 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\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% 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|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}"
+}

config.json

@@ -0,0 +1,86 @@
+{
+    "_commit_hash": null,
+    "architectures": [
+        "KeyeVL1_5ForConditionalGeneration"
+    ],
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "auto_map": {
+        "AutoConfig": "configuration_keye_vl_1_5.KeyeVL1_5Config",
+        "AutoModel": "modeling_keye_vl_1_5.KeyeVL1_5ForConditionalGeneration",
+        "AutoModelForCausalLM": "modeling_keye_vl_1_5.KeyeVL1_5ForConditionalGeneration"
+    },
+    "bos_token_id": 151643,
+    "eos_token_id": [
+        151645,
+        151643
+    ],
+    "fast_video_token_id": 151678,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 4096,
+    "image_token_id": 151655,
+    "initializer_factor": 1.0,
+    "initializer_range": 0.02,
+    "intermediate_size": 12288,
+    "max_position_embeddings": 131072,
+    "max_window_layers": 36,
+    "model_type": "KeyeVL1_5",
+    "num_attention_heads": 32,
+    "num_hidden_layers": 36,
+    "num_key_value_heads": 8,
+    "quantization": {
+        "group_size": 64,
+        "bits": 4,
+        "mode": "affine"
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 4,
+        "mode": "affine"
+    },
+    "rms_norm_eps": 1e-06,
+    "rope_scaling": {
+        "type": "mrope",
+        "mrope_section": [
+            16,
+            24,
+            24
+        ]
+    },
+    "rope_theta": 8000000,
+    "sliding_window": null,
+    "tie_word_embeddings": false,
+    "transformers_version": "4.41.2",
+    "use_cache": true,
+    "use_sliding_window": false,
+    "video_token_id": 151656,
+    "vision_config": {
+        "_attn_implementation_autoset": true,
+        "add_cross_attention": false,
+        "architectures": [
+            "KeyeVL1_5VisionModel"
+        ],
+        "attention_dropout": 0.0,
+        "auto_map": {
+            "AutoConfig": "configuration_keye_vl_1_5.KeyeVL1_5VisionConfig",
+            "AutoModel": "modeling_keye_vl_1_5.KeyeVL1_5VisionModel"
+        },
+        "hidden_size": 1152,
+        "image_size": 384,
+        "intermediate_size": 4304,
+        "model_type": "keye_vl_1_5_vision_model",
+        "num_attention_heads": 16,
+        "num_hidden_layers": 27,
+        "patch_size": 14,
+        "spatial_merge_size": 2,
+        "tokens_per_second": 2,
+        "temporal_patch_size": 2,
+        "rope_theta": 10000,
+        "has_learnable_position_embedding": true
+    },
+    "vision_end_token_id": 151653,
+    "vision_start_token_id": 151652,
+    "vision_token_id": 151654,
+    "vocab_size": 151936
+}

configuration_keye_vl_1_5.py

@@ -0,0 +1,253 @@
+# coding=utf-8
+# Copyright 2025 The Kwai Keye 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 transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+
+
+
+class KeyeVL1_5VisionConfig(PretrainedConfig):
+    model_type = "keye_vl_1_5_vision_model"
+    base_config_key = "vision_config"
+
+    def __init__(
+        self,
+        hidden_size=768,
+        intermediate_size=3072,
+        num_hidden_layers=12,
+        num_attention_heads=12,
+        num_channels=3,
+        image_size=224,
+        patch_size=14,
+        hidden_act="gelu_pytorch_tanh",
+        layer_norm_eps=1e-6,
+        attention_dropout=0.0,
+        spatial_merge_size=2,
+        temporal_patch_size=2,
+        tokens_per_second=2,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_channels = num_channels
+        self.patch_size = patch_size
+        self.image_size = image_size
+        self.attention_dropout = attention_dropout
+        self.layer_norm_eps = layer_norm_eps
+        self.hidden_act = hidden_act
+        self.spatial_merge_size = spatial_merge_size
+        self.temporal_patch_size = temporal_patch_size
+        self.tokens_per_second = tokens_per_second
+
+
+class KeyeVL1_5Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`KeyeVL1_5Model`]. It is used to instantiate a
+    Keye-VL-1.5 model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of
+    Keye-VL-1.5.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 152064):
+            Vocabulary size of the KeyeVL1_5 model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`KeyeVL1_5Model`]
+        hidden_size (`int`, *optional*, defaults to 8192):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 29568):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 80):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 64):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*, defaults to 8):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `32`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 32768):
+            The maximum sequence length that this model might ever be used with.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-05):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied.
+        rope_theta (`float`, *optional*, defaults to 1000000.0):
+            The base period of the RoPE embeddings.
+        use_sliding_window (`bool`, *optional*, defaults to `False`):
+            Whether to use sliding window attention.
+        sliding_window (`int`, *optional*, defaults to 4096):
+            Sliding window attention (SWA) window size. If not specified, will default to `4096`.
+        max_window_layers (`int`, *optional*, defaults to 80):
+            The number of layers that use SWA (Sliding Window Attention). The bottom layers use SWA while the top use full attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        vision_config (`Dict`, *optional*):
+            The config for the visual encoder initialization.
+        rope_scaling (`Dict`, *optional*):
+            Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type
+            and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value
+            accordingly.
+            Expected contents:
+                `rope_type` (`str`):
+                    The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope',
+                    'llama3'], with 'default' being the original RoPE implementation.
+                `factor` (`float`, *optional*):
+                    Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In
+                    most scaling types, a `factor` of x will enable the model to handle sequences of length x *
+                    original maximum pre-trained length.
+                `original_max_position_embeddings` (`int`, *optional*):
+                    Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during
+                    pretraining.
+                `attention_factor` (`float`, *optional*):
+                    Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
+                    computation. If unspecified, it defaults to value recommended by the implementation, using the
+                    `factor` field to infer the suggested value.
+                `beta_fast` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 32.
+                `beta_slow` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 1.
+                `short_factor` (`List[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to short contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `long_factor` (`List[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to long contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `low_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE
+                `high_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE
+
+    ```python
+    >>> from transformers import KeyeVL1_5ForConditionalGeneration, KeyeVL1_5Config
+
+    >>> # Initializing a KeyeVL1_5 style configuration
+    >>> configuration = KeyeVL1_5Config()
+
+    >>> # Initializing a model from the Keye-VL-1.5-8B style configuration
+    >>> model = KeyeVL1_5ForConditionalGeneration(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "KeyeVL1_5"
+    sub_configs = {"vision_config": KeyeVL1_5VisionConfig}
+    keys_to_ignore_at_inference = ["past_key_values"]
+    # Default tensor parallel plan for base model `KeyeVL1_5`
+    base_model_tp_plan = {
+        "layers.*.self_attn.q_proj": "colwise",
+        "layers.*.self_attn.k_proj": "colwise",
+        "layers.*.self_attn.v_proj": "colwise",
+        "layers.*.self_attn.o_proj": "rowwise",
+        "layers.*.mlp.gate_proj": "colwise",
+        "layers.*.mlp.up_proj": "colwise",
+        "layers.*.mlp.down_proj": "rowwise",
+    }
+    base_model_pp_plan = {
+        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
+        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
+        "norm": (["hidden_states"], ["hidden_states"]),
+    }
+
+    def __init__(
+        self,
+        vocab_size=152064,
+        hidden_size=8192,
+        intermediate_size=29568,
+        num_hidden_layers=80,
+        num_attention_heads=64,
+        num_key_value_heads=8,
+        hidden_act="silu",
+        max_position_embeddings=131072,
+        initializer_range=0.02,
+        rms_norm_eps=1e-05,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=8000000.0,
+        use_sliding_window=False,
+        sliding_window=4096,
+        max_window_layers=80,
+        attention_dropout=0.0,
+        vision_config=None,
+        rope_scaling=None,
+        **kwargs,
+    ):
+        if isinstance(vision_config, dict):
+            self.vision_config = self.sub_configs["vision_config"](**vision_config)
+        elif vision_config is None:
+            self.vision_config = self.sub_configs["vision_config"]()
+
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.use_sliding_window = use_sliding_window
+        self.sliding_window = sliding_window
+        self.max_window_layers = max_window_layers
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_dropout = attention_dropout
+        self.rope_scaling = rope_scaling
+
+        # Validate the correctness of rotary position embeddings parameters
+        # BC: if there is a 'type' field, move it to 'rope_type'.
+        # and change type from 'mrope' to 'default' because `mrope` does default RoPE calculations
+        # one can set it to "linear"/"dynamic" etc. to have scaled RoPE
+        # TODO: @raushan update config in the hub
+        if self.rope_scaling is not None and "type" in self.rope_scaling:
+            if self.rope_scaling["type"] == "mrope":
+                self.rope_scaling["type"] = "default"
+            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
+        rope_config_validation(self, ignore_keys={"mrope_section"})
+
+        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)
+
+
+__all__ = ["KeyeVL1_5Config"]

generation_config.json

@@ -0,0 +1,13 @@
+{
+    "bos_token_id": 151643,
+    "do_sample": true,
+    "eos_token_id": [
+        151645,
+        151643
+    ],
+    "pad_token_id": 151643,
+    "temperature": 0.6,
+    "top_k": 20,
+    "top_p": 0.95,
+    "transformers_version": "4.51.0"
+}

image_processing_keye_vl_1_5.py

@@ -0,0 +1,526 @@
+# coding=utf-8
+# Copyright 2024 The Kwai Keye 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.
+"""Image processor class for Keye-VL-1.5."""
+
+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,
+    ImageInput,
+    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.video_utils import VideoInput
+from transformers.utils import TensorType, is_vision_available, logging
+
+
+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 KeyeVL1_5ImageProcessor(BaseImageProcessor):
+    r"""
+    Constructs a Keye-VL-1.5 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-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f0df70edb9727cfa83e98f7acc2823cec2dcbc714b1cbf8790302108486ad06
+size 5159314453

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d1b463f53b0469c320cff557a7168394bbb008d83f8b036f049dc9b76a7e294
+size 372658449

preprocessor_config.json

@@ -0,0 +1,22 @@
+{
+  "min_pixels": 3136,
+  "max_pixels": 16056320,
+  "patch_size": 14,
+  "temporal_patch_size": 1,
+  "merge_size": 2,
+  "image_mean": [
+    0.5,0.5,0.5
+  ],
+  "image_std": [
+    0.5,0.5,0.5
+  ],
+  "processor_class": "KeyeVL1_5Processor",
+  "auto_map": {
+    "AutoProcessor": "processing_keye_vl_1_5.KeyeVL1_5Processor",
+    "AutoImageProcessor": "image_processing_keye_vl_1_5.KeyeVL1_5ImageProcessor"
+  },
+  "size":{
+    "height": 384,
+    "width": 384
+  }
+}

processing_keye_vl_1_5.py

@@ -0,0 +1,429 @@
+# coding=utf-8
+# Copyright 2025 The Kwai Keye 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, Optional
+
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.image_utils import ImageInput
+from transformers.video_utils import VideoInput
+from transformers.processing_utils import ProcessingKwargs, ProcessorMixin, Unpack, VideosKwargs
+from transformers.tokenization_utils_base import PreTokenizedInput, TextInput
+from .image_processing_keye_vl_1_5 import KeyeVL1_5ImageProcessor
+import torch
+import torch.nn as nn
+import numpy as np
+from itertools import chain
+
+class KeyeVL1_5VideosProcessorKwargs(VideosKwargs, total=False):
+    fps: Optional[Union[List[float], float]]
+    # 准备reszie到的width（slow）
+    width: Optional[Union[List[int], int]]
+    # 准备reszie到的height（slow）
+    height: Optional[Union[List[int], int]]
+    # 准备resize到的width（fast）
+    fast_width: Optional[Union[List[int], int]]
+    # 准备resize到的height（fast）
+    fast_height: Optional[Union[List[int], int]]
+    # 用于标记每一帧的时间戳，数量和帧数相等
+    timestamps: Optional[Union[List[torch.Tensor], torch.Tensor]]
+    # 用于标记每一帧的类型是slow还是fast，slow=0, fast=1
+    frame_types: Optional[Union[List[torch.Tensor], torch.Tensor]]
+
+
+class KeyeVL1_5ProcessorKwargs(ProcessingKwargs, total=False):
+    videos_kwargs: KeyeVL1_5VideosProcessorKwargs
+    _defaults = {
+        "text_kwargs": {
+            "padding": False,
+        },
+        "videos_kwargs": {"fps": 2.0},
+    }
+
+def select_slow_fast_frames(frames: torch.Tensor, frame_types: torch.Tensor):
+    """
+    Selects frames from a tensor based on a mask list.
+
+    Args:
+        frames (torch.Tensor): A tensor of shape (nframes, c, h, w).
+        frame_types (torch.Tensor): A int tensor of shape (nframes,)
+
+    Returns:
+        tuple[torch.Tensor, torch.Tensor]: A tuple containing two tensors:
+            - slow_frames: Frames which the type is 0.
+            - fast_frames: Frames where the type is 1.
+    """
+    nframes, _, _, _ = frames.shape
+    if frame_types.shape[-1] != nframes:
+        raise ValueError("Length of mask must be equal to the number of frames.")
+
+    mask = (frame_types == 0)
+
+    slow_frames = frames[mask]
+    fast_frames = frames[~mask]
+
+    return slow_frames, fast_frames
+
+def split_thw(tensor):
+    """Split grid_thw in t dimension, the result tensor should like [[1, h, w],...]"""
+    repeats = tensor[:, 0]
+    new_thw = torch.cat([
+        torch.ones(tensor.shape[0], 1, dtype=tensor.dtype,
+            device=tensor.device),
+        tensor[:, 1:]
+    ], dim=1)
+    return torch.repeat_interleave(new_thw, repeats, dim=0)
+
+def merge_hws(hws):
+    """
+    优化版本：使用更高效的方法合并张量
+    """
+    merged = []
+    last_hw = [-1, -1]
+    
+    for hw in hws:
+        # 找到连续相同形状的张量
+        if hw[1:] == last_hw:
+            merged[-1][0] += 1
+        else:
+            merged.append(hw)
+            last_hw = hw[1:]
+    
+    return torch.tensor(merged)
+
+class KeyeVL1_5Processor(ProcessorMixin):
+    r"""
+    [`KeyeVL1_5Processor`] offers all the functionalities of [`KeyeVL1_5ImageProcessor`] and [`Qwen2TokenizerFast`]. See the
+    [`~KeyeVL1_5Processor.__call__`] and [`~KeyeVL1_5Processor.decode`] for more information.
+    Args:
+        image_processor ([`KeyeVL1_5ImageProcessor`], *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.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
+        super().__init__(image_processor, tokenizer, chat_template=chat_template)
+
+        self.slowfast = True
+
+    def __call__(
+        self,
+        text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
+        images: ImageInput = None,
+        videos: VideoInput = None,
+        **kwargs: Unpack[KeyeVL1_5ProcessorKwargs],
+    ) -> 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
+        KeyeVL1_5ImageProcessor's [`~KeyeVL1_5ImageProcessor.__call__`] if `vision_infos` is not `None`.
+
+        Args:
+            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).
+            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.
+            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(
+            KeyeVL1_5ProcessorKwargs,
+            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
+            **kwargs,
+        )
+
+        if images is not None:
+            # slow_images = images
+            image_inputs = self.image_processor(images=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
+
+        num_frames = []
+        if videos is not None:
+            batch_slow_frames = []
+            batch_fast_frames = []
+
+            videos_kwargs = output_kwargs["videos_kwargs"]
+            num_videos = len(videos)
+            batch_frame_types = videos_kwargs.get("frame_types", [None] * num_videos)
+            batch_timestamps = videos_kwargs.get("timestamps", [None] * num_videos)
+            batch_width = videos_kwargs.get("width", [None] * num_videos)
+            batch_height = videos_kwargs.get("height", [None] * num_videos)
+            batch_fast_width = videos_kwargs.get("fast_width", [None] * num_videos)
+            batch_fast_height = videos_kwargs.get("fast_height", [None] * num_videos)
+
+            for index, frames in enumerate(videos):
+                if isinstance(frames, np.ndarray):
+                    frames = torch.from_numpy(frames)
+                nframes = frames.shape[0]
+                num_frames.append(nframes)
+                assert nframes > 0, "No frames in video"
+                if batch_frame_types[index] is None:
+                    # default to all slow frames
+                    batch_frame_types[index] = torch.zeros((nframes, ), dtype=torch.long)
+                frame_types = batch_frame_types[index]
+                slow_frames, fast_frames = select_slow_fast_frames(frames, frame_types)
+                has_fast_frames = fast_frames.shape[0] > 0
+                # resize slow frames
+                resized_width = batch_width[index]
+                resized_height = batch_height[index]
+                if resized_width is not None and resized_height is not None:
+                    slow_frames = nn.functional.interpolate(
+                        slow_frames,
+                        [resized_height, resized_width],
+                        mode="bilinear",
+                        antialias=True,
+                    ).float()
+                    do_resize = False
+                else:
+                    slow_frames = slow_frames.float()
+                    do_resize = True
+                # Tensor(N, C, H, W) -> Tuple[Tensor(1, C, H, W)]
+                # slow_frames = list(slow_frames.split(1, dim=0))，不split，在模型里面做
+                slow_video_inputs = self.image_processor(
+                    images=None, videos=[slow_frames], **output_kwargs["images_kwargs"], do_resize=do_resize)
+                slow_video_grid_thw = slow_video_inputs["video_grid_thw"]
+                batch_slow_frames.append(slow_video_inputs)
+                # # 当前这个视频每一帧的token数
+                # slow_frames_patch_nums[index] = int(slow_video_inputs["pixel_values_videos"].shape[0] / \
+                #     slow_video_grid_thw.squeeze()[0])
+
+                if has_fast_frames:
+                    # TODO: shrink fast_frames
+                    fast_resized_width = batch_fast_width[index]
+                    fast_resized_height = batch_fast_height[index]
+                    if fast_resized_width is not None and fast_resized_height is not None:
+                        fast_frames = nn.functional.interpolate(
+                            fast_frames,
+                            [fast_resized_height, fast_resized_width],
+                            mode="bilinear",
+                            antialias=True,
+                        ).float()
+                        do_fast_resize = False
+                    else:
+                        fast_frames = fast_frames.float()
+                        do_fast_resize = True
+                    # Tensor(N, C, H, W) -> Tuple[Tensor(1, C, H, W)]
+                    # fast_frames = list(fast_frames.split(1, dim=0))
+                    fast_video_inputs = self.image_processor(
+                        images=None, videos=[fast_frames], **output_kwargs["images_kwargs"], do_resize=do_fast_resize)
+                    fast_video_grid_thw = fast_video_inputs["video_grid_thw"]
+                    batch_fast_frames.append(fast_video_inputs)
+                    # # 当前这个视频的所有token数
+                    # fast_frames_token_nums[index] = int(fast_video_inputs["pixel_values_videos"].shape[0] / \
+                    #     fast_video_grid_thw.squeeze()[0])
+
+            assert len(batch_slow_frames) > 0, "Slow frames should not be empty."
+            slow_pixel_values_videos_list = [
+                video["pixel_values_videos"] for video in batch_slow_frames if video is not None]
+            slow_video_grid_thw_list = [
+                video["video_grid_thw"] for video in batch_slow_frames if video is not None]
+
+            slow_pixel_values_videos = torch.concat(slow_pixel_values_videos_list, dim=0)
+            slow_video_grid_thw = torch.concat(slow_video_grid_thw_list, dim=0)
+
+            if has_fast_frames:
+                fast_pixel_values_videos_list = [
+                    video["pixel_values_videos"] for video in batch_fast_frames \
+                        if video is not None]
+                fast_video_grid_thw_list = [
+                    video["video_grid_thw"] for video in batch_fast_frames \
+                        if video is not None]
+
+                fast_pixel_values_videos = \
+                    torch.concat(fast_pixel_values_videos_list, dim=0)
+                fast_video_grid_thw = \
+                    torch.concat(fast_video_grid_thw_list, dim=0)
+            else:
+                fast_video_grid_thw = None
+        else:
+            slow_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_place_holder_tempale = "<|placeholder|>" * (
+                        image_grid_thw[index].prod() // self.image_processor.merge_size ** 2)
+                    text[i] = text[i].replace(
+                        self.image_token,
+                        image_place_holder_tempale,
+                        1,
+                    )
+                    index += 1
+                text[i] = text[i].replace("<|placeholder|>", self.image_token)
+
+        pixel_values_videos = []
+        video_grid_thw = []
+        videos_inputs = {}
+        if slow_video_grid_thw is not None:
+            slow_video_grid_thw = split_thw(slow_video_grid_thw)
+            if fast_video_grid_thw is not None:
+                fast_video_grid_thw = split_thw(fast_video_grid_thw)
+            index = 0
+            slow_index = 0
+            fast_index = 0
+            slow_pixels_index = 0
+            fast_pixels_index = 0
+            for i in range(len(text)):
+                while self.video_token in text[i]:
+                    video_place_holder_tempale = ""
+
+                    for j in range(batch_frame_types[index].shape[-1]):
+                        if batch_timestamps[index] is not None: # 如果有时间戳
+                            video_place_holder_tempale += self.frame_token + format(batch_timestamps[index][j], ".1f")
+                        else:
+                            video_place_holder_tempale += self.frame_token
+
+                        # 当前帧是slow
+                        if batch_frame_types[index][j] == 0:
+                            num_patches = int(slow_video_grid_thw[slow_index].prod())
+                            video_place_holder_tempale += "<|placeholder|>" * (
+                                num_patches // self.image_processor.merge_size ** 2)
+                            pixel_values_videos.append(
+                                slow_pixel_values_videos[slow_pixels_index:slow_pixels_index + num_patches])
+                            slow_pixels_index = slow_pixels_index + num_patches
+                            video_grid_thw.append(slow_video_grid_thw[slow_index].tolist())
+                            slow_index += 1
+
+                        # 当前帧是fast
+                        elif batch_frame_types[index][j] == 1:
+                            num_patches = int(fast_video_grid_thw[fast_index].prod()) 
+                            video_place_holder_tempale += self.fast_start + "<|placeholder|>" * (
+                                num_patches // self.image_processor.merge_size ** 2) + \
+                                self.fast_end
+                            pixel_values_videos.append(
+                                fast_pixel_values_videos[fast_pixels_index:fast_pixels_index + num_patches])
+                            fast_pixels_index = fast_pixels_index + num_patches
+                            video_grid_thw.append(fast_video_grid_thw[fast_index].tolist())
+                            fast_index += 1
+                    text[i] = text[i].replace(
+                        self.video_token,
+                        video_place_holder_tempale,
+                        1,
+                    )
+                    index += 1
+                text[i] = text[i].replace("<|placeholder|>", self.video_token)
+
+            videos_inputs["pixel_values_videos"] = torch.cat(pixel_values_videos, dim=0)
+            videos_inputs["video_grid_thw"] = merge_hws(video_grid_thw)
+            videos_inputs["num_frames"] = torch.tensor(num_frames)
+
+        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
+
+
+
+__all__ = ["KeyeVL1_5Processor"]

processor_config.json

@@ -0,0 +1,38 @@
+{
+  "auto_map": {
+    "AutoProcessor": "processing_keye_vl_1_5.KeyeVL1_5Processor"
+  },
+  "image_processor": {
+    "auto_map": {
+      "AutoImageProcessor": "image_processing_keye_vl_1_5.KeyeVL1_5ImageProcessor",
+      "AutoProcessor": "processing_keye_vl_1_5.KeyeVL1_5Processor"
+    },
+    "do_convert_rgb": true,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "KeyeVL1_5ImageProcessor",
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "max_pixels": 16056320,
+    "merge_size": 2,
+    "min_pixels": 3136,
+    "patch_size": 14,
+    "resample": 2,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "max_pixels": 16056320,
+      "min_pixels": 3136
+    },
+    "temporal_patch_size": 1
+  },
+  "processor_class": "KeyeVL1_5Processor"
+}

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 sha256:852176eb11b42e01c67097d2ea334863b3c414c78c3208c20311075339080d1f
+size 11424979

tokenizer_config.json

@@ -0,0 +1,27 @@
+{
+  "add_prefix_space": false,
+  "auto_map": {
+    "AutoProcessor": "processing_keye_vl_1_5.KeyeVL1_5Processor"
+  },
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "is_local": true,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "processor_class": "KeyeVL1_5Processor",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}