Upload EMOVAForConditionalGeneration

README.md

@@ -0,0 +1,199 @@
+---
+library_name: transformers
+tags: []
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]

config.json

@@ -0,0 +1,51 @@
+{
+  "_name_or_path": "/8tssd1/kchenbf/cache/emova-qwen-2-5-7b-hf/",
+  "architectures": [
+    "EMOVAForConditionalGeneration"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_emova.EMOVAConfig",
+    "AutoModel": "modeling_emova.EMOVAForConditionalGeneration",
+    "AutoModelForCausalLM": "modeling_emova.EMOVAForConditionalGeneration"
+  },
+  "default_system_prompt": "You are a helpful assistant.",
+  "ignore_index": -100,
+  "image_token_index": 155761,
+  "mm_projector_config": {
+    "mlp_depth": 2,
+    "trainable": true,
+    "type": "MLPProjector"
+  },
+  "model_type": "emova",
+  "text_config": {
+    "architectures": [
+      "Qwen2ForCausalLM"
+    ],
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "hidden_size": 3584,
+    "intermediate_size": 18944,
+    "max_position_embeddings": 32768,
+    "max_window_layers": 28,
+    "model_type": "qwen2",
+    "num_attention_heads": 28,
+    "num_hidden_layers": 28,
+    "num_key_value_heads": 4,
+    "rope_theta": 1000000.0,
+    "sliding_window": null,
+    "torch_dtype": "float32",
+    "use_sliding_window": false,
+    "vocab_size": 155776
+  },
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.44.2",
+  "vision_config": {
+    "max_pixels": 3211264,
+    "model_type": "qwen2_vl",
+    "pretrained_model_name_or_path": "Emova-ollm/qwen2vit600m",
+    "trainable": true,
+    "type": "Qwen2VisionTower",
+    "unfreeze_mm_vision_tower": true
+  }
+}

configuration_emova.py

@@ -0,0 +1,160 @@
+# coding=utf-8
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+# 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.
+"""EMOVA model configuration"""
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+from transformers.models.auto import CONFIG_MAPPING
+
+from .configuration_qwen2vit import Qwen2VLVisionConfig
+
+logger = logging.get_logger(__name__)
+
+
+class EMOVAConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`EMOVAForConditionalGeneration`]. It is used to instantiate an
+    EMOVA model according to the specified arguments, defining the model architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        vision_config (`Union[AutoConfig, dict]`,  *optional*, defaults to `CLIPVisionConfig`):
+            The config object or dictionary of the vision backbone.
+        text_config (`Union[AutoConfig, dict]`, *optional*, defaults to `LlamaConfig`):
+            The config object or dictionary of the text backbone.
+        ignore_index (`int`, *optional*, defaults to -100):
+            The ignore index for the loss function.
+        image_token_index (`int`, *optional*, defaults to 32000):
+            The image token index to encode the image prompt.
+        projector_hidden_act (`str`, *optional*, defaults to `"gelu"`):
+            The activation function used by the multimodal projector.
+        vision_feature_select_strategy (`str`, *optional*, defaults to `"default"`):
+            The feature selection strategy used to select the vision feature from the vision backbone.
+            Can be one of `"default"` or `"full"`. If `"default"`, the CLS token is removed from the vision features.
+            If `"full"`, the full vision features are used.
+        vision_feature_layer (`int`, *optional*, defaults to -2):
+            The index of the layer to select the vision feature.
+        image_grid_pinpoints (`List`, *optional*, defaults to `[[336, 672], [672, 336], [672, 672], [1008, 336], [336, 1008]]`):
+            A list of possible resolutions to use for processing high resolution images. Each item in the list should be a tuple or list
+            of the form `(height, width)`.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied.
+
+    Example:
+
+    ```python
+    >>> from transformers import EMOVAForConditionalGeneration, EMOVAConfig, CLIPVisionConfig, LlamaConfig
+
+    >>> # Initializing a CLIP-vision config
+    >>> vision_config = CLIPVisionConfig()
+
+    >>> # Initializing a Llama config
+    >>> text_config = LlamaConfig()
+
+    >>> # Initializing a EMOVA style configuration
+    >>> configuration = EMOVAConfig(vision_config, text_config)
+
+    >>> # Initializing a model from the style configuration
+    >>> model = EMOVAForConditionalGeneration(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "emova"
+    is_composition = False
+
+    def __init__(
+            self,
+            vision_config=None,
+            mm_projector_config=None,
+            text_config=None,
+            ignore_index=-100,
+            image_token_index=32000,
+            vision_feature_select_strategy="default",
+            vision_feature_layer=-2,
+            is_native_resolution=True,
+            image_grid_pinpoints=None,
+            use_image_newline_parameter=False,
+            tie_word_embeddings=False,
+            **kwargs,
+    ):
+        self.ignore_index = ignore_index
+        self.image_token_index = image_token_index
+
+        if vision_feature_select_strategy not in ["default", "full"]:
+            raise ValueError(
+                "vision_feature_select_strategy should be one of 'default', 'full'."
+                f"Got: {vision_feature_select_strategy}"
+            )
+
+        # self.vision_feature_select_strategy = vision_feature_select_strategy
+        # self.vision_feature_layer = vision_feature_layer
+        # image_grid_pinpoints = (
+        #     image_grid_pinpoints
+        #     if image_grid_pinpoints is not None
+        #     else [[336, 672], [672, 336], [672, 672], [1008, 336], [336, 1008]]
+        # )
+        # self.is_native_resolution = is_native_resolution
+        # self.image_grid_pinpoints = image_grid_pinpoints if not self.is_native_resolution else None
+        # self.use_image_newline_parameter = use_image_newline_parameter
+
+        if isinstance(vision_config, dict):
+            vision_config = Qwen2VLVisionConfig(**vision_config)
+        elif vision_config is None:
+            vision_config = Qwen2VLVisionConfig(
+                depth=32,
+                embed_dim=1280,
+                hidden_act="quick_gelu",
+                hidden_size=3584,
+                in_channels=3,
+                in_chans=3,
+                mlp_ratio=4,
+                model_type="qwen2_vl",
+                num_heads=16,
+                patch_size=14,
+                spatial_merge_size=2,
+                spatial_patch_size=14,
+                temporal_patch_size=2,
+                initializer_range=0.02,
+            )
+
+        # if isinstance(vision_config, dict):
+        #     vision_config["model_type"] = (
+        #         vision_config["model_type"] if "model_type" in vision_config else "clip_vision_model"
+        #     )
+        #     vision_config = CONFIG_MAPPING[vision_config["model_type"]](**vision_config)
+        # elif vision_config is None:
+        #     vision_config = CONFIG_MAPPING["clip_vision_model"](
+        #         intermediate_size=4096,
+        #         hidden_size=1024,
+        #         patch_size=14,
+        #         image_size=336,
+        #         num_hidden_layers=24,
+        #         num_attention_heads=16,
+        #         vocab_size=32000,
+        #         projection_dim=768,
+        #     )
+
+        self.vision_config = vision_config
+        self.mm_projector_config = mm_projector_config
+        if isinstance(text_config, dict):
+            text_config["model_type"] = text_config["model_type"] if "model_type" in text_config else "llama"
+            text_config = CONFIG_MAPPING[text_config["model_type"]](**text_config)
+        elif text_config is None:
+            text_config = CONFIG_MAPPING["llama"]()
+
+        self.text_config = text_config
+        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)

configuration_qwen2vit.py

@@ -0,0 +1,71 @@
+# coding=utf-8
+# Copyright 2024 The Qwen team, Alibaba Group and the HuggingFace Inc. team. All rights reserved.
+#
+# 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.
+"""Qwen2VL model configuration"""
+
+import os
+from typing import Union
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class Qwen2VLVisionConfig(PretrainedConfig):
+    model_type = "qwen2_vl"
+
+    def __init__(
+            self,
+            depth=32,
+            embed_dim=1280,
+            hidden_size=3584,
+            hidden_act="quick_gelu",
+            mlp_ratio=4,
+            num_heads=16,
+            in_channels=3,
+            patch_size=14,
+            spatial_merge_size=2,
+            temporal_patch_size=2,
+            **kwargs,
+    ):
+        super().__init__(**kwargs)
+
+        self.depth = depth
+        self.embed_dim = embed_dim
+        self.hidden_size = hidden_size
+        self.hidden_act = hidden_act
+        self.mlp_ratio = mlp_ratio
+        self.num_heads = num_heads
+        self.in_channels = in_channels
+        self.patch_size = patch_size
+        self.spatial_merge_size = spatial_merge_size
+        self.temporal_patch_size = temporal_patch_size
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
+        cls._set_token_in_kwargs(kwargs)
+
+        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
+
+        # if config_dict.get("model_type") == "qwen2_vl":
+        #     config_dict = config_dict["vision_config"]
+
+        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
+            logger.warning(
+                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
+                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
+            )
+
+        return cls.from_dict(config_dict, **kwargs)

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "transformers_version": "4.44.2"
+}

model-00001-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc3a4159d2383c434613cd0273005331a59a33a3188ac70b3660e202285ec8f7
+size 4908875200

model-00002-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e657ab12a61acc374acdbd087ad5b4640f64387bf868184f9124ffbcb77e6a7c
+size 4932752808

model-00003-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3f05e160738389c89a2c48790d2f1befb39118830957b67930afc8cc4c7dcef
+size 4991497848

model-00004-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:849a82141a4b722c193ca108432d873a7231a9e029081d479e86038b931ac5be
+size 1854322976

model.safetensors.index.json

@@ -0,0 +1,741 @@
+{
+  "metadata": {
+    "total_size": 16687361024
+  },
+  "weight_map": {
+    "language_model.lm_head.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.embed_tokens.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.0.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.1.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.10.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.10.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.11.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.12.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.13.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.14.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.15.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.15.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.15.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.15.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.16.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.16.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.17.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.18.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.19.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.2.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.2.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.20.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.20.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.21.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.22.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.23.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.24.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.input_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.mlp.down_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.mlp.up_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.25.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.input_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.26.mlp.down_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.26.mlp.gate_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.mlp.up_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.26.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.26.self_attn.k_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.self_attn.k_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.self_attn.o_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.self_attn.q_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.self_attn.q_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.self_attn.v_proj.bias": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.26.self_attn.v_proj.weight": "model-00003-of-00004.safetensors",
+    "language_model.model.layers.27.input_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.mlp.down_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.mlp.gate_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.mlp.up_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.self_attn.k_proj.bias": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.self_attn.k_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.self_attn.o_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.self_attn.q_proj.bias": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.self_attn.q_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.self_attn.v_proj.bias": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.27.self_attn.v_proj.weight": "model-00004-of-00004.safetensors",
+    "language_model.model.layers.3.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.3.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.input_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.mlp.down_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.mlp.gate_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.mlp.up_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.4.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.5.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.5.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.5.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.5.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.5.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.5.self_attn.k_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.5.self_attn.k_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.5.self_attn.o_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.5.self_attn.q_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.5.self_attn.q_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.5.self_attn.v_proj.bias": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.5.self_attn.v_proj.weight": "model-00001-of-00004.safetensors",
+    "language_model.model.layers.6.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.6.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.7.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.8.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.input_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.mlp.down_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.mlp.gate_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.mlp.up_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.self_attn.k_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.self_attn.k_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.self_attn.o_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.self_attn.q_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.self_attn.q_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.self_attn.v_proj.bias": "model-00002-of-00004.safetensors",
+    "language_model.model.layers.9.self_attn.v_proj.weight": "model-00002-of-00004.safetensors",
+    "language_model.model.norm.weight": "model-00004-of-00004.safetensors",
+    "multi_modal_projector.0.bias": "model-00001-of-00004.safetensors",
+    "multi_modal_projector.0.weight": "model-00001-of-00004.safetensors",
+    "multi_modal_projector.2.bias": "model-00001-of-00004.safetensors",
+    "multi_modal_projector.2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.0.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.1.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.10.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.11.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.12.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.13.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.14.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.15.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.16.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.17.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.18.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.19.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.2.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.20.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.21.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.22.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.23.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.24.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.25.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.26.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.27.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.28.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.29.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.3.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.30.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.31.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.4.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.5.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.6.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.7.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.8.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.attn.proj.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.attn.proj.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.attn.qkv.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.attn.qkv.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.mlp.fc1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.mlp.fc1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.mlp.fc2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.mlp.fc2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.norm1.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.norm1.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.norm2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.blocks.9.norm2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.merger.ln_q.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.merger.ln_q.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.merger.mlp.0.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.merger.mlp.0.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.merger.mlp.2.bias": "model-00001-of-00004.safetensors",
+    "vision_tower.merger.mlp.2.weight": "model-00001-of-00004.safetensors",
+    "vision_tower.patch_embed.proj.weight": "model-00001-of-00004.safetensors"
+  }
+}

modeling_emova.py

@@ -0,0 +1,816 @@
+# coding=utf-8
+# Copyright 2024 the HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""PyTorch EMOVA model."""
+
+import math
+from dataclasses import dataclass
+from functools import partial
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+import torch.utils.checkpoint
+from torch import nn
+
+from transformers import PreTrainedModel
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache
+from transformers.image_processing_utils import select_best_resolution
+from transformers.modeling_outputs import ModelOutput
+from transformers.utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+from transformers.models.auto import AutoModel, AutoModelForCausalLM
+
+from .configuration_emova import EMOVAConfig
+from .modeling_qwen2vit import Qwen2VisionTower
+
+from timm.models.regnet import RegStage
+
+try:
+    from timm.layers import LayerNorm2d
+except:
+    from timm.models.layers import LayerNorm2d
+from einops import rearrange
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "EMOVAConfig"
+
+
+@dataclass
+class EMOVACausalLMOutputWithPast(ModelOutput):
+    """
+    Base class for EMOVA causal language model (or autoregressive) outputs.
+
+    Args:
+        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        image_hidden_states (`tuple(torch.FloatTensor)`, *optional*):
+            Tuple of `torch.FloatTensor` (one for the output of the image embeddings, `(batch_size, num_images,
+            sequence_length, hidden_size)`.
+
+            image_hidden_states of the model produced by the vision encoder, and optionally by the perceiver
+    """
+
+    loss: Optional[torch.FloatTensor] = None
+    logits: torch.FloatTensor = None
+    past_key_values: Optional[List[torch.FloatTensor]] = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+    image_hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+
+
+class EMOVAMultiModalProjector(nn.Sequential):
+    # CAbstractor
+    def __init__(self, config):
+        super(EMOVAMultiModalProjector, self).__init__()
+        hidden_size = config.text_config.hidden_size
+        mm_hidden_size = config.vision_config.hidden_size
+        mlp_depth = config.mm_projector_config['mlp_depth']
+
+        modules = [nn.Linear(mm_hidden_size, hidden_size)]
+        for _ in range(1, mlp_depth):
+            modules.append(nn.GELU())
+            modules.append(nn.Linear(hidden_size, hidden_size))
+        super(EMOVAMultiModalProjector, self).__init__(*modules)
+
+
+EMOVA_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`EMOVAConfig`] or [`EMOVAVisionConfig`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare LLaMA Model outputting raw hidden-states without any specific head on top.",
+    EMOVA_START_DOCSTRING,
+)
+class EMOVAPreTrainedModel(PreTrainedModel):
+    config_class = EMOVAConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["EMOVAVisionAttention"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = True
+    _supports_cache_class = True
+
+    def _init_weights(self, module):
+        std = (
+            self.config.initializer_range
+            if hasattr(self.config, "initializer_range")
+            else self.config.text_config.initializer_range
+        )
+
+        if hasattr(module, "class_embedding"):
+            module.class_embedding.data.normal_(mean=0.0, std=std)
+
+        if isinstance(module, (nn.Linear, nn.Conv2d)):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    @property
+    def _supports_sdpa(self):
+        """
+        Retrieve language_model's attribute to check whether the model supports
+        SDPA or not.
+        """
+        return self.language_model._supports_sdpa
+
+
+EMOVA_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, image_size, image_size)):
+            The tensors corresponding to the input images. Pixel values can be obtained using
+            [`AutoImageProcessor`]. See [`EMOVAImageProcessor.__call__`] for details. [`EMOVAProcessor`] uses
+            [`EMOVAImageProcessor`] for processing images.
+        image_sizes (`torch.LongTensor` of shape `(batch_size, 2)`, *optional*):
+            The sizes of the images in the batch, being (height, width) for each image.
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`. [What are position IDs?](../glossary#position-ids)
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        vision_feature_layer (`int`, *optional*, defaults to -2):
+            The index of the layer to select the vision feature.
+        vision_feature_select_strategy (`str`, *optional*, defaults to `"default"`):
+            The feature selection strategy used to select the vision feature from the vision backbone.
+            Can be one of `"default"` or `"full"`. If `"default"`, the CLS token is removed from the vision features.
+            If `"full"`, the full vision features are used.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    """The EMOVA model which consists of a vision backbone and a language model.""",
+    EMOVA_START_DOCSTRING,
+)
+class EMOVAForConditionalGeneration(EMOVAPreTrainedModel):
+    def __init__(self, config: EMOVAConfig, **kwargs):
+        super().__init__(config)
+        self.vision_tower = Qwen2VisionTower(config.vision_config)
+        self.multi_modal_projector = EMOVAMultiModalProjector(config)
+
+        self.vocab_size = config.text_config.vocab_size
+        self.language_model = AutoModelForCausalLM.from_config(
+            config.text_config, attn_implementation=config._attn_implementation
+        )
+        self.pad_token_id = self.config.pad_token_id if self.config.pad_token_id is not None else -1
+        self._padding_side = "left"  # set it to left by default, user can use setter to change padding_sides
+        self.post_init()
+
+    @property
+    def padding_side(self):
+        return self._padding_side
+
+    @padding_side.setter
+    def padding_side(self, padding_side: str):
+        if padding_side not in ["left", "right"]:
+            raise ValueError(f"{padding_side} is not `left` or `right`.")
+        self._padding_side = padding_side
+
+    def get_input_embeddings(self):
+        return self.language_model.get_input_embeddings()
+
+    def set_input_embeddings(self, value):
+        self.language_model.set_input_embeddings(value)
+
+    def get_output_embeddings(self):
+        return self.language_model.get_output_embeddings()
+
+    def set_output_embeddings(self, new_embeddings):
+        self.language_model.set_output_embeddings(new_embeddings)
+
+    def set_decoder(self, decoder):
+        self.language_model.set_decoder(decoder)
+
+    def get_decoder(self):
+        return self.language_model.get_decoder()
+
+    def tie_weights(self):
+        return self.language_model.tie_weights()
+
+    def resize_token_embeddings(self, new_num_tokens: Optional[int] = None, pad_to_multiple_of=None) -> nn.Embedding:
+        model_embeds = self.language_model.resize_token_embeddings(new_num_tokens, pad_to_multiple_of)
+        # update vocab size
+        self.config.text_config.vocab_size = model_embeds.num_embeddings
+        self.vocab_size = model_embeds.num_embeddings
+        return model_embeds
+
+    def _merge_input_ids_with_image_features(
+            self,
+            image_features,
+            feature_lens,
+            inputs_embeds,
+            input_ids,
+            attention_mask,
+            position_ids=None,
+            labels=None,
+            image_token_index=None,
+            ignore_index=-100,
+    ):
+        """
+        Merge input_ids with with image features into final embeddings
+
+        Args:
+            image_features (`torch.Tensor` of shape `(all_feature_lens, embed_dim)`):
+                All vision vectors of all images in the batch
+            feature_lens (`torch.LongTensor` of shape `(num_images)`):
+                The length of visual embeddings of each image as stacked in `image_features`
+            inputs_embeds (`torch.Tensor` of shape `(batch_size, sequence_length, embed_dim)`):
+                Token embeddings before merging with visual embeddings
+            input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Input_ids of tokens, possibly filled with image token
+            attention_mask (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Mask to avoid performing attention on padding token indices.
+            position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+                config.n_positions - 1]`.
+            labels (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*)
+                :abels need to be recalculated to support training (if provided)
+            image_token_index (`int`, *optional*)
+                Token id used to indicate the special "image" token. Defaults to `config.image_token_index`
+            ignore_index (`int`, *optional*)
+                Value that is used to pad `labels` and will be ignored when calculated loss. Default: -100.
+        Returns:
+            final_embedding, final_attention_mask, position_ids, final_labels
+
+        Explanation:
+            each image has variable length embeddings, with length specified by feature_lens
+            image_features is concatenation of all visual embed vectors
+            task: fill each <image> with the correct number of visual embeddings
+            Example:
+                X (5 patches), Y (3 patches), Z (8)
+                X, Y are in the same sequence (in-context learning)
+            if right padding
+                input_ids: [
+                    a b c d e f X g h i j k Y l m
+                    o p q r Z s t u v _ _ _ _ _ _
+                ]
+                input_ids should be: [
+                    a b c d e f X X X X X g h i j k Y Y Y l m
+                    o p q r Z Z Z Z Z Z Z Z s t u v _ _ _ _ _
+                ]
+                labels should be: [
+                    a b c d e f _ _ _ _ _ g h i j k _ _ _ l m
+                    o p q r _ _ _ _ _ _ _ _ s t u v _ _ _ _ _
+                ]
+            elif left padding
+                input_ids: [
+                    a b c d e f X g h i j k Y l m
+                    _ _ _ _ _ _ o p q r Z s t u v
+                ]
+                input_ids should be: [
+                    a b c d e f X X X X X g h i j k Y Y Y l m
+                    _ _ _ _ _ o p q r Z Z Z Z Z Z Z Z s t u v
+                ]
+                labels should be: [
+                    a b c d e f _ _ _ _ _ g h i j k _ _ _ l m
+                    _ _ _ _ _ o p q r _ _ _ _ _ _ _ _ s t u v
+                ]
+            Edge cases:
+                * If tokens are same but image token sizes are different, then cannot infer left or right padding
+                ```python
+                cat_img = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+                chart_img = Image.open(requests.get("https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true", stream=True).raw)
+                prompts = [
+                    "[INST] <image>\nWhat is shown in this image? [/INST]",
+                    "[INST] <image>\nWhat is shown in this image? [/INST]",
+                ]
+                inputs = processor(prompts, [chart_img, cat_img], return_tensors='pt', padding=True).to("cuda")
+                    chart_img has 2634 tokens, while cat_img has 2340 tokens
+                ```
+
+                input_ids: [
+                    a b c d X g h
+                    i j Y k l m n
+                ]
+                where X is 3 tokens while Y is 5, this mean after merge
+                if left-padding (batched generation)
+                    input_ids should be: [
+                        _ _ a b c d X X X g h
+                        i j Y Y Y Y Y k l m n
+                    ]
+                elif (right padding) (training)
+                    input_ids should be: [
+                        a b c d X X X g h _ _
+                        i j Y Y Y Y Y k l m n
+                    ]
+        """
+        image_token_index = image_token_index if image_token_index is not None else self.config.image_token_index
+        ignore_index = ignore_index if ignore_index is not None else self.config.ignore_index
+
+        with torch.no_grad():
+            num_images = feature_lens.size(0)
+            num_image_features, embed_dim = image_features.shape
+            if feature_lens.sum() != num_image_features:
+                raise ValueError(f"{feature_lens=} / {feature_lens.sum()} != {image_features.shape=}")
+            batch_size = input_ids.shape[0]
+            _left_padding = torch.any(attention_mask[:, 0] == 0)
+            _right_padding = torch.any(attention_mask[:, -1] == 0)
+
+            left_padding = True if not self.training else False
+            if batch_size > 1 and not self.training:
+                if _left_padding and not _right_padding:
+                    left_padding = True
+                elif not _left_padding and _right_padding:
+                    left_padding = False
+                elif not _left_padding and not _right_padding:
+                    # both side is 1, so cannot tell
+                    left_padding = self.padding_side == "left"
+                else:
+                    # invalid attention_mask
+                    raise ValueError(f"both side of attention_mask has zero, invalid. {attention_mask}")
+
+            # Whether to turn off right padding
+            # 1. Create a mask to know where special image tokens are
+            special_image_token_mask = input_ids == image_token_index
+            # special_image_token_mask: [bsz, seqlen]
+            num_special_image_tokens = torch.sum(special_image_token_mask, dim=-1)
+            # num_special_image_tokens: [bsz]
+            # Reserve for padding of num_images
+            total_num_special_image_tokens = torch.sum(special_image_token_mask)
+            if total_num_special_image_tokens != num_images:
+                raise ValueError(
+                    f"Number of image tokens in input_ids ({total_num_special_image_tokens}) different from num_images ({num_images})."
+                )
+            # Compute the maximum embed dimension
+            # max_image_feature_lens is max_feature_lens per batch
+            feature_lens = feature_lens.to(input_ids.device)
+            feature_lens_batch = feature_lens.split(num_special_image_tokens.tolist(), dim=0)
+            feature_lens_batch_sum = torch.tensor([x.sum() for x in feature_lens_batch], device=input_ids.device)
+            embed_sequence_lengths = (
+                    (attention_mask == 1).long().sum(-1) - num_special_image_tokens + feature_lens_batch_sum
+            )
+            max_embed_dim = embed_sequence_lengths.max()
+
+            batch_indices, non_image_indices = torch.where((input_ids != image_token_index) & (attention_mask == 1))
+            # 2. Compute the positions where text should be written
+            # Calculate new positions for text tokens in merged image-text sequence.
+            # `special_image_token_mask` identifies image tokens. Each image token will be replaced by `nb_text_tokens_per_images` text tokens.
+            # `torch.cumsum` computes how each image token shifts subsequent text token positions.
+            # - 1 to adjust for zero-based indexing, as `cumsum` inherently increases indices by one.
+            # ! instead of special_image_token_mask * (num_image_patches - 1)
+            #   special_image_token_mask * (num_feature_len - 1)
+            special_image_token_mask = special_image_token_mask.long()
+            special_image_token_mask[special_image_token_mask == 1] = feature_lens - 1
+            new_token_positions = torch.cumsum((special_image_token_mask + 1), -1) - 1
+            if left_padding:
+                # shift right token positions so that they are ending at the same number
+                # the below here was incorrect? new_token_positions += new_token_positions[:, -1].max() - new_token_positions[:, -1:]
+                new_token_positions += max_embed_dim - 1 - new_token_positions[:, -1:]
+
+            text_to_overwrite = new_token_positions[batch_indices, non_image_indices]
+
+        # 3. Create the full embedding, already padded to the maximum position
+        final_embedding = torch.zeros(
+            batch_size, max_embed_dim, embed_dim, dtype=inputs_embeds.dtype, device=inputs_embeds.device
+        )
+        final_attention_mask = torch.zeros(
+            batch_size, max_embed_dim, dtype=attention_mask.dtype, device=inputs_embeds.device
+        )
+        final_input_ids = torch.full(
+            (batch_size, max_embed_dim), self.pad_token_id, dtype=input_ids.dtype, device=inputs_embeds.device
+        )
+        # In case the Vision model or the Language model has been offloaded to CPU, we need to manually
+        # set the corresponding tensors into their correct target device.
+        target_device = inputs_embeds.device
+        batch_indices, non_image_indices, text_to_overwrite = (
+            batch_indices.to(target_device),
+            non_image_indices.to(target_device),
+            text_to_overwrite.to(target_device),
+        )
+        attention_mask = attention_mask.to(target_device)
+        input_ids = input_ids.to(target_device)
+
+        # 4. Fill the embeddings based on the mask. If we have ["hey" "<image>", "how", "are"]
+        # we need to index copy on [0, 577, 578, 579] for the text and [1:576] for the image features
+        final_embedding[batch_indices, text_to_overwrite] = inputs_embeds[batch_indices, non_image_indices]
+        final_attention_mask[batch_indices, text_to_overwrite] = attention_mask[batch_indices, non_image_indices]
+        final_input_ids[batch_indices, text_to_overwrite] = input_ids[batch_indices, non_image_indices]
+        final_labels = None
+        if labels is not None:
+            labels = labels.to(target_device)
+            final_labels = torch.full_like(final_attention_mask, ignore_index).to(torch.long)
+            final_labels[batch_indices, text_to_overwrite] = labels[batch_indices, non_image_indices]
+
+        # 5. Fill the embeddings corresponding to the images. Anything that is not `text_positions` needs filling (#29835)
+        with torch.no_grad():
+            image_to_overwrite = torch.full(
+                (batch_size, max_embed_dim), True, dtype=torch.bool, device=inputs_embeds.device
+            )
+            image_to_overwrite[batch_indices, text_to_overwrite] = False
+            embed_indices = torch.arange(max_embed_dim).unsqueeze(0).to(target_device)
+            embed_indices = embed_indices.expand(batch_size, max_embed_dim)
+            embed_seq_lens = embed_sequence_lengths[:, None].to(target_device)
+
+            if left_padding:
+                # exclude padding on the left
+                max_embed_dim = max_embed_dim.to(target_device)
+                val = (max_embed_dim - embed_indices) <= embed_seq_lens
+            else:
+                # exclude padding on the right
+                val = embed_indices < embed_seq_lens
+            image_to_overwrite &= val
+
+            if image_to_overwrite.sum() != num_image_features:
+                raise ValueError(
+                    f"{image_to_overwrite.sum()=} != {num_image_features=} The input provided to the model are wrong. "
+                    f"The number of image tokens is {torch.sum(special_image_token_mask)} while"
+                    f" the number of image given to the model is {num_images}. "
+                    f"This prevents correct indexing and breaks batch generation."
+                )
+        final_embedding[image_to_overwrite] = image_features.contiguous().reshape(-1, embed_dim).to(target_device)
+        final_attention_mask |= image_to_overwrite
+        position_ids = (final_attention_mask.cumsum(-1) - 1).masked_fill_((final_attention_mask == 0), 1)
+
+        return final_embedding, final_attention_mask, position_ids, final_labels, final_input_ids
+
+    @add_start_docstrings_to_model_forward(EMOVA_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=EMOVACausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            pixel_values: torch.FloatTensor = None,
+            image_sizes: Optional[torch.LongTensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            vision_feature_layer: Optional[int] = None,
+            vision_feature_select_strategy: Optional[str] = None,
+            labels: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, EMOVACausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from PIL import Image
+        >>> import requests
+        >>> from transformers import AutoProcessor, EMOVAForConditionalGeneration
+
+        >>> model = EMOVAForConditionalGeneration.from_pretrained("Emova-ollm/emova-qwen-2-5-7b-hf")
+        >>> processor = AutoProcessor.from_pretrained("Emova-ollm/emova-qwen-2-5-7b-hf")
+
+        >>> prompt = "<image>\nWhat is shown in this image?"
+        >>> url = "https://www.ilankelman.org/stopsigns/australia.jpg"
+        >>> image = Image.open(requests.get(url, stream=True).raw)
+
+        >>> inputs = processor(text=prompt, images=image, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(**inputs, max_length=30)
+        >>> processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "\nWhat is shown in this image? The image appears to be a radar chart, which is a type of multi-dimensional plot (...)"
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if inputs_embeds is None:
+            # 1. Extract the input embeddings
+            # In case image_token_index is not in the embeddings (extra token but embedding don't have it)
+            for_inputs_embeds_ids = input_ids.clone()
+            for_inputs_embeds_ids[(input_ids == self.config.image_token_index)] = 0
+            inputs_embeds = self.get_input_embeddings()(for_inputs_embeds_ids)
+
+            # 2. Merge text and images
+            if pixel_values is not None and input_ids.shape[1] != 1 and pixel_values.size(0) > 0:
+                # ! infer image_num_patches from image_sizes
+
+                image_features = self.vision_tower(pixel_values.to(self.dtype), image_sizes)
+                image_features = self.multi_modal_projector(image_features)
+
+                spatial_merge_size = self.vision_tower.spatial_merge_size
+                feature_lens = torch.as_tensor(
+                    [t * h * w // (self.vision_tower.spatial_merge_size ** 2) for t, h, w in image_sizes])
+                image_num_patches = sum(feature_lens)
+
+                # NOTE we only support multimodal_patch_merge_type == "spatial_unpad"
+                inputs_embeds = inputs_embeds.to(image_features.dtype)
+                inputs_embeds, attention_mask, position_ids, labels, _ = self._merge_input_ids_with_image_features(
+                    image_features,
+                    feature_lens,
+                    inputs_embeds,
+                    input_ids,
+                    attention_mask,
+                    position_ids,
+                    labels=labels,
+                )
+
+            # pixel_values is not None but is empty ---> text only cases
+            elif pixel_values is not None and input_ids.shape[1] != 1 and pixel_values.size(0) == 0:
+                # there are no images
+                pass
+
+            # In case input_ids.shape[1] == 1 & pixel_values==None & past_key_values != None, we are in the case of
+            # generation with cache
+            elif past_key_values is not None and pixel_values is not None and input_ids.shape[1] == 1:
+                # Retrieve the first layer to inspect the logits and mask out the hidden states
+                # that are set to 0
+                first_layer_past_key_value = past_key_values[0][0][:, :, :, 0]
+
+                # Sum all dimensions of head_dim (-2) to avoid random errors such as: https://github.com/huggingface/transformers/pull/28032#issuecomment-1863691941
+                batch_index, non_attended_tokens = torch.where(first_layer_past_key_value.float().sum(-2) == 0)
+
+                # Get the target length
+                target_length = input_ids.shape[1]
+                past_length = first_layer_past_key_value.shape[-1]
+
+                extended_attention_mask = torch.ones(
+                    (attention_mask.shape[0], past_length),
+                    dtype=attention_mask.dtype,
+                    device=attention_mask.device,
+                )
+
+                # Filter out only the tokens that can be un-attended, this can happen
+                # if one uses EMOVA + Fused modules where the cache on the
+                # first iteration is already big enough, or if one passes custom cache
+                valid_indices = non_attended_tokens < extended_attention_mask.size(-1)
+                new_batch_index = batch_index[valid_indices]
+                new_non_attended_tokens = non_attended_tokens[valid_indices]
+
+                # Zero-out the places where we don't need to attend
+                extended_attention_mask[new_batch_index, new_non_attended_tokens] = 0
+
+                attention_mask = torch.cat((extended_attention_mask, attention_mask[:, -target_length:]), dim=1)
+
+                position_ids = torch.sum(attention_mask, dim=1).unsqueeze(-1) - 1
+
+        outputs = self.language_model(
+            attention_mask=attention_mask.to(inputs_embeds.device),
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        logits = outputs[0]
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            if attention_mask is not None:
+                shift_attention_mask = attention_mask[..., 1:]
+                shift_logits = logits[..., :-1, :][shift_attention_mask.to(logits.device) != 0].contiguous()
+                shift_labels = labels[..., 1:][shift_attention_mask.to(labels.device) != 0].contiguous()
+            else:
+                shift_logits = logits[..., :-1, :].contiguous()
+                shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(
+                shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1).to(shift_logits.device)
+            )
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return EMOVACausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids,
+            past_key_values=None,
+            inputs_embeds=None,
+            pixel_values=None,
+            image_sizes=None,
+            attention_mask=None,
+            **kwargs,
+    ):
+        if past_key_values is not None:
+            if isinstance(past_key_values, Cache):
+                cache_length = past_key_values.get_seq_length()
+                past_length = past_key_values.seen_tokens
+            else:
+                cache_length = past_length = past_key_values[0][0].shape[2]
+
+            # Keep only the unprocessed tokens:
+            # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
+            # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as
+            # input)
+            if attention_mask is not None and attention_mask.shape[1] > input_ids.shape[1]:
+                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length):]
+            # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
+            # input_ids based on the past_length.
+            elif past_length < input_ids.shape[1]:
+                input_ids = input_ids[:, past_length:]
+            # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.
+            elif self.config.image_token_index in input_ids:
+                input_ids = input_ids[:, input_ids.shape[1] - 1:]
+            # If the cache has seen more tokens than it can hold, then the cache has a size limit. Let's discard the
+            # older attention values, as their corresponding values are not part of the input.
+            if cache_length < past_length and attention_mask is not None:
+                attention_mask = attention_mask[:, -(cache_length + input_ids.shape[1]):]
+
+        position_ids = kwargs.get("position_ids", None)
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -input_ids.shape[1]:]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+                "pixel_values": pixel_values,
+                "image_sizes": image_sizes,
+            }
+        )
+        return model_inputs
+
+    def chat(self, tokenizer, pixel_values, question, generation_config, history=None, return_history=False,
+             num_patches_list=None, IMG_START_TOKEN='<img>', IMG_END_TOKEN='</img>', IMG_CONTEXT_TOKEN='<IMG_CONTEXT>',
+             verbose=False):
+        raise RuntimeError("!!!")
+
+        if history is None and pixel_values is not None and '<image>' not in question:
+            question = '<image>\n' + question
+
+        if num_patches_list is None:
+            num_patches_list = [pixel_values.shape[0]] if pixel_values is not None else []
+        assert pixel_values is None or len(pixel_values) == sum(num_patches_list)
+
+        img_context_token_id = tokenizer.convert_tokens_to_ids(IMG_CONTEXT_TOKEN)
+        self.img_context_token_id = img_context_token_id
+
+        template = get_conv_template(self.template)
+        template.system_message = self.system_message
+        eos_token_id = tokenizer.convert_tokens_to_ids(template.sep)
+
+        history = [] if history is None else history
+        for (old_question, old_answer) in history:
+            template.append_message(template.roles[0], old_question)
+            template.append_message(template.roles[1], old_answer)
+        template.append_message(template.roles[0], question)
+        template.append_message(template.roles[1], None)
+        query = template.get_prompt()
+
+        if verbose and pixel_values is not None:
+            image_bs = pixel_values.shape[0]
+            print(f'dynamic ViT batch size: {image_bs}')
+
+        for num_patches in num_patches_list:
+            image_tokens = IMG_START_TOKEN + IMG_CONTEXT_TOKEN * self.num_image_token * num_patches + IMG_END_TOKEN
+            query = query.replace('<image>', image_tokens, 1)
+
+        model_inputs = tokenizer(query, return_tensors='pt')
+        input_ids = model_inputs['input_ids'].cuda()
+        attention_mask = model_inputs['attention_mask'].cuda()
+        generation_config['eos_token_id'] = eos_token_id
+        generation_output = self.generate(
+            pixel_values=pixel_values,
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            **generation_config
+        )
+        response = tokenizer.batch_decode(generation_output, skip_special_tokens=True)[0]
+        response = response.split(template.sep)[0].strip()
+        history.append((question, response))
+        if return_history:
+            return response, history
+        else:
+            query_to_print = query.replace(IMG_CONTEXT_TOKEN, '')
+            query_to_print = query_to_print.replace(f'{IMG_START_TOKEN}{IMG_END_TOKEN}', '<image>')
+            if verbose:
+                print(query_to_print, response)
+            return response
+
+    def _reorder_cache(self, *args, **kwargs):
+        return self.language_model._reorder_cache(*args, **kwargs)

modeling_qwen2vit.py

@@ -0,0 +1,335 @@
+# 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.
+"""PyTorch Qwen2-VL model."""
+
+import math
+from dataclasses import dataclass
+from typing import Any, Dict, List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch.nn import CrossEntropyLoss, LayerNorm
+
+from transformers.activations import ACT2FN
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    is_flash_attn_2_available,
+    logging,
+)
+from .configuration_qwen2vit import Qwen2VLVisionConfig
+
+if is_flash_attn_2_available():
+    from flash_attn import flash_attn_varlen_func
+
+    from transformers.modeling_flash_attention_utils import _flash_attention_forward
+else:
+    flash_attn_varlen_func = None
+
+logger = logging.get_logger(__name__)
+
+
+# Copied from transformers.models.llama.modeling_llama.rotate_half
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2:]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb_vision(tensor: torch.Tensor, freqs: torch.Tensor) -> torch.Tensor:
+    orig_dtype = tensor.dtype
+    tensor = tensor.float()
+    cos = freqs.cos()
+    sin = freqs.sin()
+    cos = cos.unsqueeze(1).repeat(1, 1, 2).unsqueeze(0).float()
+    sin = sin.unsqueeze(1).repeat(1, 1, 2).unsqueeze(0).float()
+    output = (tensor * cos) + (rotate_half(tensor) * sin)
+    output = output.to(orig_dtype)
+    return output
+
+
+class VisionRotaryEmbedding(nn.Module):
+    def __init__(self, dim: int, theta: float = 10000.0) -> None:
+        super().__init__()
+        inv_freq = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=torch.float) / dim))
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+    def forward(self, seqlen: int) -> torch.Tensor:
+        seq = torch.arange(seqlen, device=self.inv_freq.device, dtype=self.inv_freq.dtype)
+        freqs = torch.outer(seq, self.inv_freq)
+        return freqs
+
+
+class PatchEmbed(nn.Module):
+    def __init__(
+            self,
+            patch_size: int = 14,
+            temporal_patch_size: int = 2,
+            in_channels: int = 3,
+            embed_dim: int = 1152,
+    ) -> None:
+        super().__init__()
+        self.patch_size = patch_size
+        self.temporal_patch_size = temporal_patch_size
+        self.in_channels = in_channels
+        self.embed_dim = embed_dim
+
+        kernel_size = [temporal_patch_size, patch_size, patch_size]
+        self.proj = nn.Conv3d(in_channels, embed_dim, kernel_size=kernel_size, stride=kernel_size, bias=False)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        target_dtype = self.proj.weight.dtype
+        hidden_states = hidden_states.view(
+            -1, self.in_channels, self.temporal_patch_size, self.patch_size, self.patch_size
+        )
+        hidden_states = self.proj(hidden_states.to(dtype=target_dtype)).view(-1, self.embed_dim)
+        return hidden_states
+
+
+class PatchMerger(nn.Module):
+    def __init__(self, dim: int, context_dim: int, spatial_merge_size: int = 2) -> None:
+        super().__init__()
+        self.hidden_size = context_dim * (spatial_merge_size ** 2)
+        self.ln_q = LayerNorm(context_dim, eps=1e-6)
+        self.mlp = nn.Sequential(
+            nn.Linear(self.hidden_size, self.hidden_size),
+            nn.GELU(),
+            nn.Linear(self.hidden_size, dim),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.mlp(self.ln_q(x).view(-1, self.hidden_size))
+        return x
+
+
+class VisionMlp(nn.Module):
+    def __init__(self, dim: int, hidden_dim: int, hidden_act: str) -> None:
+        super().__init__()
+        self.fc1 = nn.Linear(dim, hidden_dim)
+        self.act = ACT2FN[hidden_act]
+        self.fc2 = nn.Linear(hidden_dim, dim)
+
+    def forward(self, x) -> torch.Tensor:
+        return self.fc2(self.act(self.fc1(x)))
+
+
+class VisionAttention(nn.Module):
+    def __init__(self, dim: int, num_heads: int = 16) -> None:
+        super().__init__()
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+        self.qkv = nn.Linear(dim, dim * 3, bias=True)
+        self.proj = nn.Linear(dim, dim)
+
+    def forward(
+            self, hidden_states: torch.Tensor, cu_seqlens: torch.Tensor, rotary_pos_emb: torch.Tensor = None
+    ) -> torch.Tensor:
+        seq_length = hidden_states.shape[0]
+        q, k, v = self.qkv(hidden_states).reshape(seq_length, 3, self.num_heads, -1).permute(1, 0, 2, 3).unbind(0)
+        q = apply_rotary_pos_emb_vision(q.unsqueeze(0), rotary_pos_emb).squeeze(0)
+        k = apply_rotary_pos_emb_vision(k.unsqueeze(0), rotary_pos_emb).squeeze(0)
+
+        attention_mask = torch.full(
+            [1, seq_length, seq_length], torch.finfo(q.dtype).min, device=q.device, dtype=q.dtype
+        )
+        for i in range(1, len(cu_seqlens)):
+            attention_mask[..., cu_seqlens[i - 1]: cu_seqlens[i], cu_seqlens[i - 1]: cu_seqlens[i]] = 0
+
+        q = q.transpose(0, 1)
+        k = k.transpose(0, 1)
+        v = v.transpose(0, 1)
+        attn_weights = torch.matmul(q, k.transpose(1, 2)) / math.sqrt(self.head_dim)
+        attn_weights = attn_weights + attention_mask
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(q.dtype)
+        attn_output = torch.matmul(attn_weights, v)
+        attn_output = attn_output.transpose(0, 1)
+        attn_output = attn_output.reshape(seq_length, -1)
+        attn_output = self.proj(attn_output)
+        return attn_output
+
+
+class VisionFlashAttention2(nn.Module):
+    def __init__(self, dim: int, num_heads: int = 16) -> None:
+        super().__init__()
+        self.num_heads = num_heads
+        self.qkv = nn.Linear(dim, dim * 3, bias=True)
+        self.proj = nn.Linear(dim, dim)
+
+    def forward(
+            self, hidden_states: torch.Tensor, cu_seqlens: torch.Tensor, rotary_pos_emb: torch.Tensor = None
+    ) -> torch.Tensor:
+        seq_length = hidden_states.shape[0]
+        q, k, v = self.qkv(hidden_states).reshape(seq_length, 3, self.num_heads, -1).permute(1, 0, 2, 3).unbind(0)
+        q = apply_rotary_pos_emb_vision(q.unsqueeze(0), rotary_pos_emb).squeeze(0)
+        k = apply_rotary_pos_emb_vision(k.unsqueeze(0), rotary_pos_emb).squeeze(0)
+
+        max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max().item()
+        attn_output = flash_attn_varlen_func(q, k, v, cu_seqlens, cu_seqlens, max_seqlen, max_seqlen).reshape(
+            seq_length, -1
+        )
+        attn_output = self.proj(attn_output)
+        return attn_output
+
+
+class VisionSdpaAttention(nn.Module):
+    def __init__(self, dim: int, num_heads: int = 16) -> None:
+        super().__init__()
+        self.num_heads = num_heads
+        self.qkv = nn.Linear(dim, dim * 3, bias=True)
+        self.proj = nn.Linear(dim, dim)
+
+    def forward(
+            self, hidden_states: torch.Tensor, cu_seqlens: torch.Tensor, rotary_pos_emb: torch.Tensor = None
+    ) -> torch.Tensor:
+        seq_length = hidden_states.shape[0]
+        q, k, v = self.qkv(hidden_states).reshape(seq_length, 3, self.num_heads, -1).permute(1, 0, 2, 3).unbind(0)
+        q = apply_rotary_pos_emb_vision(q.unsqueeze(0), rotary_pos_emb).squeeze(0)
+        k = apply_rotary_pos_emb_vision(k.unsqueeze(0), rotary_pos_emb).squeeze(0)
+
+        attention_mask = torch.zeros([1, seq_length, seq_length], device=q.device, dtype=torch.bool)
+        for i in range(1, len(cu_seqlens)):
+            attention_mask[..., cu_seqlens[i - 1]: cu_seqlens[i], cu_seqlens[i - 1]: cu_seqlens[i]] = True
+        q = q.transpose(0, 1)
+        k = k.transpose(0, 1)
+        v = v.transpose(0, 1)
+        attn_output = F.scaled_dot_product_attention(q, k, v, attention_mask, dropout_p=0.0)
+        attn_output = attn_output.transpose(0, 1)
+        attn_output = attn_output.reshape(seq_length, -1)
+        attn_output = self.proj(attn_output)
+        return attn_output
+
+
+QWEN2_VL_VISION_ATTENTION_CLASSES = {
+    "eager": VisionAttention,
+    "flash_attention_2": VisionFlashAttention2,
+    "sdpa": VisionSdpaAttention,
+}
+
+
+class Qwen2VLVisionBlock(nn.Module):
+    def __init__(self, config, attn_implementation: str = "sdpa") -> None:
+        super().__init__()
+        self.norm1 = LayerNorm(config.embed_dim, eps=1e-6)
+        self.norm2 = LayerNorm(config.embed_dim, eps=1e-6)
+        mlp_hidden_dim = int(config.embed_dim * config.mlp_ratio)
+
+        self.attn = QWEN2_VL_VISION_ATTENTION_CLASSES[attn_implementation](
+            config.embed_dim, num_heads=config.num_heads
+        )
+        self.mlp = VisionMlp(dim=config.embed_dim, hidden_dim=mlp_hidden_dim, hidden_act=config.hidden_act)
+
+    def forward(self, hidden_states, cu_seqlens, rotary_pos_emb) -> torch.Tensor:
+        hidden_states = hidden_states + self.attn(
+            self.norm1(hidden_states), cu_seqlens=cu_seqlens, rotary_pos_emb=rotary_pos_emb
+        )
+        hidden_states = hidden_states + self.mlp(self.norm2(hidden_states))
+        return hidden_states
+
+
+class Qwen2VisionTower(PreTrainedModel):
+    config_class = Qwen2VLVisionConfig
+    _no_split_modules = ["Qwen2VLVisionBlock"]
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, (nn.Linear, nn.Conv3d)):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    def __init__(self, config) -> None:
+        super().__init__(config)
+        self.spatial_merge_size = config.spatial_merge_size
+
+        self.patch_embed = PatchEmbed(
+            patch_size=config.patch_size,
+            temporal_patch_size=config.temporal_patch_size,
+            in_channels=config.in_channels,
+            embed_dim=config.embed_dim,
+        )
+
+        head_dim = config.embed_dim // config.num_heads
+        self.rotary_pos_emb = VisionRotaryEmbedding(head_dim // 2)
+
+        self.blocks = nn.ModuleList(
+            [Qwen2VLVisionBlock(config, "eager") for _ in range(config.depth)]
+        )
+        self.merger = PatchMerger(
+            dim=config.hidden_size, context_dim=config.embed_dim, spatial_merge_size=config.spatial_merge_size
+        )
+
+    def get_dtype(self) -> torch.dtype:
+        return self.blocks[0].mlp.fc2.weight.dtype
+
+    def get_device(self) -> torch.device:
+        return self.blocks[0].mlp.fc2.weight.device
+
+    def rot_pos_emb(self, grid_thw):
+        pos_ids = []
+        for t, h, w in grid_thw:
+            hpos_ids = torch.arange(h).unsqueeze(1).expand(-1, w)
+            hpos_ids = hpos_ids.reshape(
+                h // self.spatial_merge_size,
+                self.spatial_merge_size,
+                w // self.spatial_merge_size,
+                self.spatial_merge_size,
+            )
+            hpos_ids = hpos_ids.permute(0, 2, 1, 3)
+            hpos_ids = hpos_ids.flatten()
+
+            wpos_ids = torch.arange(w).unsqueeze(0).expand(h, -1)
+            wpos_ids = wpos_ids.reshape(
+                h // self.spatial_merge_size,
+                self.spatial_merge_size,
+                w // self.spatial_merge_size,
+                self.spatial_merge_size,
+            )
+            wpos_ids = wpos_ids.permute(0, 2, 1, 3)
+            wpos_ids = wpos_ids.flatten()
+            pos_ids.append(torch.stack([hpos_ids, wpos_ids], dim=-1).repeat(t, 1))
+        pos_ids = torch.cat(pos_ids, dim=0)
+        max_grid_size = grid_thw[:, 1:].max()
+        rotary_pos_emb_full = self.rotary_pos_emb(max_grid_size)
+        rotary_pos_emb = rotary_pos_emb_full[pos_ids].flatten(1)
+        return rotary_pos_emb
+
+    def forward(self, hidden_states: torch.Tensor, grid_thw: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.patch_embed(hidden_states)
+        rotary_pos_emb = self.rot_pos_emb(grid_thw)
+
+        cu_seqlens = torch.repeat_interleave(grid_thw[:, 1] * grid_thw[:, 2], grid_thw[:, 0]).cumsum(
+            dim=0, dtype=torch.int32
+        )
+        cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
+
+        for blk in self.blocks:
+            hidden_states = blk(hidden_states, cu_seqlens=cu_seqlens, rotary_pos_emb=rotary_pos_emb)
+
+        return self.merger(hidden_states)