Layer-pruned Ovis2-8B (PPL-based (Shortened LLaMA), removed 8 layers, 20.9% reduction)

.gitattributes

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

chat_template.jinja

@@ -0,0 +1,54 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- messages[0]['content'] }}
+    {%- else %}
+        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}
+    {%- endif %}
+    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content %}
+            {{- '\n' + message.content }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n{"name": "' }}
+            {{- tool_call.name }}
+            {{- '", "arguments": ' }}
+            {{- tool_call.arguments | tojson }}
+            {{- '}\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

config.json

@@ -0,0 +1,213 @@
+{
+  "architectures": [
+    "Ovis"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_ovis.OvisConfig",
+    "AutoModelForCausalLM": "modeling_ovis.Ovis"
+  },
+  "conversation_formatter_class": "QwenConversationFormatter",
+  "disable_tie_weight": false,
+  "dtype": "bfloat16",
+  "hidden_size": 3584,
+  "llm_attn_implementation": "flash_attention_2",
+  "llm_config": {
+    "_attn_implementation_autoset": true,
+    "_name_or_path": "Qwen/Qwen2.5-7B-Instruct",
+    "add_cross_attention": false,
+    "architectures": [
+      "Qwen2ForCausalLM"
+    ],
+    "attention_dropout": 0.0,
+    "bos_token_id": 151643,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "dtype": "bfloat16",
+    "eos_token_id": 151645,
+    "finetuning_task": null,
+    "hidden_act": "silu",
+    "hidden_size": 3584,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 18944,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_types": [
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 32768,
+    "max_window_layers": 28,
+    "model_type": "qwen2",
+    "num_attention_heads": 28,
+    "num_hidden_layers": 28,
+    "num_key_value_heads": 4,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "pad_token_id": null,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "return_dict": true,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "rope_theta": 1000000.0,
+      "rope_type": "default"
+    },
+    "sep_token_id": null,
+    "sliding_window": null,
+    "task_specific_params": null,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": false,
+    "tokenizer_class": null,
+    "torchscript": false,
+    "use_bfloat16": false,
+    "use_cache": true,
+    "use_sliding_window": false,
+    "vocab_size": 152064
+  },
+  "model_type": "ovis",
+  "multimodal_max_length": 32768,
+  "num_hidden_layers": 20,
+  "transformers_version": "5.3.0",
+  "visual_tokenizer_config": {
+    "_attn_implementation_autoset": true,
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "backbone_config": {
+      "_attn_implementation_autoset": true,
+      "_name_or_path": "apple/aimv2-huge-patch14-448",
+      "add_cross_attention": false,
+      "architectures": [
+        "AIMv2Model"
+      ],
+      "attention_dropout": 0.0,
+      "auto_map": {
+        "AutoConfig": "configuration_aimv2.AIMv2Config",
+        "AutoModel": "modeling_aimv2.AIMv2Model",
+        "FlaxAutoModel": "modeling_flax_aimv2.FlaxAIMv2Model"
+      },
+      "bos_token_id": null,
+      "chunk_size_feed_forward": 0,
+      "cross_attention_hidden_size": null,
+      "decoder_start_token_id": null,
+      "dtype": "bfloat16",
+      "eos_token_id": null,
+      "finetuning_task": null,
+      "hidden_size": 1536,
+      "id2label": {
+        "0": "LABEL_0",
+        "1": "LABEL_1"
+      },
+      "image_size": 448,
+      "intermediate_size": 4096,
+      "is_decoder": false,
+      "is_encoder_decoder": false,
+      "label2id": {
+        "LABEL_0": 0,
+        "LABEL_1": 1
+      },
+      "model_type": "aimv2",
+      "num_attention_heads": 12,
+      "num_channels": 3,
+      "num_hidden_layers": 24,
+      "output_attentions": false,
+      "output_hidden_states": false,
+      "pad_token_id": null,
+      "patch_size": 14,
+      "prefix": null,
+      "problem_type": null,
+      "projection_dropout": 0.0,
+      "pruned_heads": {},
+      "qkv_bias": false,
+      "return_dict": true,
+      "rms_norm_eps": 1e-05,
+      "sep_token_id": null,
+      "task_specific_params": null,
+      "tf_legacy_loss": false,
+      "tie_encoder_decoder": false,
+      "tie_word_embeddings": true,
+      "tokenizer_class": null,
+      "torchscript": false,
+      "use_bfloat16": false,
+      "use_bias": false
+    },
+    "backbone_kwargs": {},
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "depths": null,
+    "drop_cls_token": false,
+    "dtype": null,
+    "eos_token_id": null,
+    "finetuning_task": null,
+    "hidden_stride": 2,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "model_type": "aimv2_visual_tokenizer",
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "pad_token_id": null,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "return_dict": true,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "tau": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenize_function": "softmax",
+    "tokenizer_class": null,
+    "torchscript": false,
+    "use_bfloat16": false,
+    "use_indicators": false,
+    "vocab_size": 65536
+  }
+}

configuration_aimv2.py

@@ -0,0 +1,63 @@
+# copied from https://huggingface.co/apple/aimv2-huge-patch14-448
+from typing import Any
+
+from transformers.configuration_utils import PretrainedConfig
+
+__all__ = ["AIMv2Config"]
+
+
+class AIMv2Config(PretrainedConfig):
+    """This is the configuration class to store the configuration of an [`AIMv2Model`].
+
+    Instantiating a configuration with the defaults will yield a similar configuration
+    to that of the [apple/aimv2-large-patch14-224](https://huggingface.co/apple/aimv2-large-patch14-224).
+
+    Args:
+        hidden_size: Dimension of the hidden representations.
+        intermediate_size: Dimension of the SwiGLU representations.
+        num_hidden_layers: Number of hidden layers in the Transformer.
+        num_attention_heads: Number of attention heads for each attention layer
+            in the Transformer.
+        num_channels: Number of input channels.
+        image_size: Image size.
+        patch_size: Patch size.
+        rms_norm_eps: Epsilon value used for the RMS normalization layer.
+        attention_dropout: Dropout ratio for attention probabilities.
+        projection_dropout: Dropout ratio for the projection layer after the attention.
+        qkv_bias: Whether to add a bias to the queries, keys and values.
+        use_bias: Whether to add a bias in the feed-forward and projection layers.
+        kwargs: Keyword arguments for the [`PretrainedConfig`].
+    """
+
+    model_type: str = "aimv2"
+
+    def __init__(
+        self,
+        hidden_size: int = 1024,
+        intermediate_size: int = 2816,
+        num_hidden_layers: int = 24,
+        num_attention_heads: int = 8,
+        num_channels: int = 3,
+        image_size: int = 224,
+        patch_size: int = 14,
+        rms_norm_eps: float = 1e-5,
+        attention_dropout: float = 0.0,
+        projection_dropout: float = 0.0,
+        qkv_bias: bool = False,
+        use_bias: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(**kwargs)
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_channels = num_channels
+        self.patch_size = patch_size
+        self.image_size = image_size
+        self.attention_dropout = attention_dropout
+        self.rms_norm_eps = rms_norm_eps
+
+        self.projection_dropout = projection_dropout
+        self.qkv_bias = qkv_bias
+        self.use_bias = use_bias

configuration_ovis.py

@@ -0,0 +1,204 @@
+from abc import ABC, abstractmethod
+from typing import List, Dict, Union, Optional
+
+from transformers import PretrainedConfig, AutoConfig, AutoModel
+from .configuration_aimv2 import AIMv2Config
+from .modeling_aimv2 import AIMv2Model
+
+IGNORE_ID = -100
+IMAGE_TOKEN_ID = -200
+IMAGE_TOKEN = "<image>"
+IMAGE_ATOM_ID = -300
+IMAGE_INDICATOR_IDS = [-301, -302, -303, -304, -305]
+
+AutoConfig.register("aimv2", AIMv2Config, exist_ok=True)
+AutoModel.register(AIMv2Config, AIMv2Model)
+
+# ----------------------------------------------------------------------
+#                     Visual Tokenizer Configuration
+# ----------------------------------------------------------------------
+class BaseVisualTokenizerConfig(PretrainedConfig):
+    def __init__(
+        self,
+        vocab_size=16384,
+        tokenize_function="softmax",
+        tau=1.0,
+        depths=None,
+        drop_cls_token=False,
+        backbone_config: Optional[Union[PretrainedConfig, dict]] = None,
+        hidden_stride: int = 1,
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+        self.vocab_size = vocab_size
+        self.tokenize_function = tokenize_function
+        self.tau = tau
+        if isinstance(depths, str):
+            depths = [int(x) for x in depths.split('|')]
+        self.depths = depths
+        self.backbone_kwargs = {}
+        self.drop_cls_token = drop_cls_token
+        if backbone_config is not None:
+            assert isinstance(backbone_config, (PretrainedConfig, dict)), \
+                f"expect `backbone_config` to be instance of PretrainedConfig or dict, but got {type(backbone_config)} type"
+            if not isinstance(backbone_config, PretrainedConfig):
+                model_type = backbone_config['model_type']
+                backbone_config.pop('model_type')
+                backbone_config = AutoConfig.for_model(model_type, **backbone_config)
+        self.backbone_config = backbone_config
+        self.hidden_stride = hidden_stride
+
+
+class Aimv2VisualTokenizerConfig(BaseVisualTokenizerConfig):
+    model_type = "aimv2_visual_tokenizer"
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        if self.drop_cls_token:
+            self.drop_cls_token = False
+        if self.depths:
+            assert len(self.depths) == 1
+            self.backbone_kwargs['num_hidden_layers'] = self.depths[0]
+
+
+AutoConfig.register("aimv2_visual_tokenizer", Aimv2VisualTokenizerConfig, exist_ok=True)
+
+
+# ----------------------------------------------------------------------
+#                           Ovis Configuration
+# ----------------------------------------------------------------------
+class OvisConfig(PretrainedConfig):
+    model_type = "ovis"
+
+    def __init__(
+        self,
+        llm_config: Optional[Union[PretrainedConfig, dict]] = None,
+        visual_tokenizer_config: Optional[Union[PretrainedConfig, dict]] = None,
+        multimodal_max_length=8192,
+        hidden_size=None,
+        conversation_formatter_class=None,
+        llm_attn_implementation=None,
+        disable_tie_weight=False,
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+        if llm_config is not None:
+            assert isinstance(llm_config, (PretrainedConfig, dict)), \
+                f"expect `llm_config` to be instance of PretrainedConfig or dict, but got {type(llm_config)} type"
+            if not isinstance(llm_config, PretrainedConfig):
+                model_type = llm_config['model_type']
+                llm_config.pop('model_type')
+                llm_config = AutoConfig.for_model(model_type, **llm_config)
+        self.llm_config = llm_config
+        if visual_tokenizer_config is not None:
+            assert isinstance(visual_tokenizer_config, (PretrainedConfig, dict)), \
+                f"expect `visual_tokenizer_config` to be instance of PretrainedConfig or dict, but got {type(visual_tokenizer_config)} type"
+            if not isinstance(visual_tokenizer_config, PretrainedConfig):
+                model_type = visual_tokenizer_config['model_type']
+                visual_tokenizer_config.pop('model_type')
+                visual_tokenizer_config = AutoConfig.for_model(model_type, **visual_tokenizer_config)
+        self.visual_tokenizer_config = visual_tokenizer_config
+        self.multimodal_max_length = multimodal_max_length
+        self.hidden_size = hidden_size
+        self.conversation_formatter_class = conversation_formatter_class
+        self.llm_attn_implementation = llm_attn_implementation
+        self.disable_tie_weight = disable_tie_weight
+
+
+# ----------------------------------------------------------------------
+#                         Conversation Formatter
+# ----------------------------------------------------------------------
+class ConversationFormatter(ABC):
+    support_tokenizer_types = None
+
+    def __init__(self, tokenizer):
+        tokenizer_type = type(tokenizer).__name__
+        assert tokenizer_type in self.support_tokenizer_types, \
+            f'Invalid tokenizer type, expected one from `{self.support_tokenizer_types}`, but got `{tokenizer_type}`'
+        self.tokenizer = tokenizer
+        self.image_token = IMAGE_TOKEN
+        self.image_token_id = IMAGE_TOKEN_ID
+        self.ignore_id = IGNORE_ID
+
+    def _tokenize_with_image_symbol(self, text):
+        text_chunks = [self.tokenizer(chunk, add_special_tokens=False).input_ids for chunk in
+                       text.split(self.image_token)]
+        token_ids = []
+        num_chuck = len(text_chunks)
+        for i, chunk in enumerate(text_chunks):
+            token_ids.extend(chunk)
+            if i < num_chuck - 1:
+                token_ids.append(self.image_token_id)
+        return token_ids
+
+    @abstractmethod
+    def format(self, conversations: List[Dict], generation_preface=None):
+        pass
+
+    @abstractmethod
+    def format_query(self, query, generation_preface=""):
+        pass
+
+
+class QwenConversationFormatter(ConversationFormatter):
+    support_tokenizer_types = ['QWenTokenizer', 'Qwen2TokenizerFast', 'Qwen2Tokenizer']
+
+    def __init__(self, tokenizer):
+        super().__init__(tokenizer)
+        self.from2role = {
+            "system": "<|im_start|>system\n",
+            "human": "<|im_start|>user\n",
+            "gpt": "<|im_start|>assistant\n",
+        }
+        self.gpt_token_num = None
+        self.im_end = "<|im_end|>\n"
+        self.default_system_prompt = "You are a helpful assistant."
+
+    def format(self, conversations: List[Dict], generation_preface=None):
+        if self.gpt_token_num is None:
+            self.gpt_token_num = len(self.tokenizer(self.from2role["gpt"], add_special_tokens=False).input_ids)
+
+        if conversations[0]["from"] != "system":
+            conversations.insert(0, {
+                "from": "system",
+                "value": self.default_system_prompt
+            })
+
+        if generation_preface is not None:
+            conversations.append({
+                "from": "gpt",
+                "value": generation_preface
+            })
+
+        prompt = ""
+        input_ids = []
+        labels = []
+        num_conversation = len(conversations)
+        for i, conversation in enumerate(conversations):
+            frm = conversation["from"]
+            role = self.from2role[frm]
+            message = conversation["value"]
+            text = role + message
+            if i < num_conversation - 1 or generation_preface is None:
+                text += self.im_end
+            prompt += text
+            token_ids = self._tokenize_with_image_symbol(text)
+            input_ids.extend(token_ids)
+            label_ids = [self.ignore_id] * len(token_ids)
+            if frm == "gpt" and generation_preface is None:
+                # learning `\n` following `im_end` is meaningless, so the last `\n` token is ignored in label
+                label_ids[self.gpt_token_num:-1] = token_ids[self.gpt_token_num:-1]
+            labels.extend(label_ids)
+
+        assert self._tokenize_with_image_symbol(prompt) == input_ids
+        assert len(input_ids) == len(labels)
+
+        return prompt, input_ids, labels
+
+    def format_query(self, query, generation_preface=""):
+        prompt, input_ids, _ = self.format([{
+            "from": "human",
+            "value": query
+        }], generation_preface=generation_preface)
+
+        return prompt, input_ids

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2f5b587fdc9c375c91766ff53633dd2376db1c80d1cb31c0ba91e7e0df74cc3
+size 14141693348

modeling_aimv2.py

@@ -0,0 +1,198 @@
+# adapted from https://huggingface.co/apple/aimv2-huge-patch14-448 (modification: add gradient checkpoint support)
+from typing import Optional, Tuple, Union
+
+import torch
+from .configuration_aimv2 import AIMv2Config
+from torch import nn
+from torch.nn import functional as F
+from transformers.modeling_outputs import BaseModelOutputWithNoAttention
+from transformers.modeling_utils import PreTrainedModel
+
+__all__ = ["AIMv2Model"]
+
+
+class RMSNorm(nn.Module):
+    def __init__(self, dim: int, eps: float = 1e-6):
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(dim))
+        self.eps = eps
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        output = self._norm(x.float()).type_as(x)
+        return output * self.weight
+
+    def extra_repr(self) -> str:
+        return f"{tuple(self.weight.shape)}, eps={self.eps}"
+
+    def _norm(self, x: torch.Tensor) -> torch.Tensor:
+        return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
+
+
+class AIMv2SwiGLUFFN(nn.Module):
+    def __init__(self, config: AIMv2Config):
+        super().__init__()
+        hidden_features = config.intermediate_size
+        in_features = config.hidden_size
+        bias = config.use_bias
+
+        self.fc1 = nn.Linear(in_features, hidden_features, bias=bias)
+        self.fc2 = nn.Linear(hidden_features, in_features, bias=bias)
+        self.fc3 = nn.Linear(in_features, hidden_features, bias=bias)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = F.silu(self.fc1(x)) * self.fc3(x)
+        x = self.fc2(x)
+        return x
+
+
+class AIMv2PatchEmbed(nn.Module):
+    def __init__(self, config: AIMv2Config):
+        super().__init__()
+        self.proj = nn.Conv2d(
+            config.num_channels,
+            config.hidden_size,
+            kernel_size=(config.patch_size, config.patch_size),
+            stride=(config.patch_size, config.patch_size),
+        )
+        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.proj(x).flatten(2).transpose(1, 2)
+        x = self.norm(x)
+        return x
+
+
+class AIMv2ViTPreprocessor(nn.Module):
+    def __init__(self, config: AIMv2Config):
+        super().__init__()
+        num_patches = (config.image_size // config.patch_size) ** 2
+
+        self.patchifier = AIMv2PatchEmbed(config)
+        self.pos_embed = nn.Parameter(torch.zeros((1, num_patches, config.hidden_size)))
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        tokens = self.patchifier(x)
+        _, N, _ = tokens.shape
+        pos_embed = self.pos_embed.to(tokens.device)
+        tokens = tokens + pos_embed[:, :N]
+        return tokens
+
+
+class AIMv2Attention(nn.Module):
+    def __init__(self, config: AIMv2Config):
+        super().__init__()
+        dim = config.hidden_size
+
+        self.num_heads = config.num_attention_heads
+        self.qkv = nn.Linear(dim, dim * 3, bias=config.qkv_bias)
+        self.attn_drop = nn.Dropout(config.attention_dropout)
+        self.proj = nn.Linear(dim, dim, bias=config.use_bias)
+        self.proj_drop = nn.Dropout(config.projection_dropout)
+
+    def forward(
+        self, x: torch.Tensor, mask: Optional[torch.Tensor] = None
+    ) -> torch.Tensor:
+        B, N, C = x.shape
+        qkv = (
+            self.qkv(x)
+            .reshape(B, N, 3, self.num_heads, C // self.num_heads)
+            .permute(2, 0, 3, 1, 4)
+        )
+        q, k, v = qkv.unbind(0)
+
+        x = F.scaled_dot_product_attention(q, k, v, attn_mask=mask)
+        x = x.transpose(1, 2).contiguous().reshape(B, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class AIMv2Block(nn.Module):
+    def __init__(self, config: AIMv2Config):
+        super().__init__()
+        self.attn = AIMv2Attention(config)
+        self.norm_1 = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.mlp = AIMv2SwiGLUFFN(config)
+        self.norm_2 = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+        self, x: torch.Tensor, mask: Optional[torch.Tensor] = None
+    ) -> torch.Tensor:
+        x = x + self.attn(self.norm_1(x), mask)
+        x = x + self.mlp(self.norm_2(x))
+        return x
+
+
+class AIMv2Transformer(nn.Module):
+    def __init__(self, config: AIMv2Config):
+        super().__init__()
+        self.blocks = nn.ModuleList(
+            [AIMv2Block(config) for _ in range(config.num_hidden_layers)]
+        )
+        self.post_trunk_norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        tokens: torch.Tensor,
+        mask: Optional[torch.Tensor] = None,
+        output_hidden_states: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor, ...]]]:
+        hidden_states = () if output_hidden_states else None
+        for block in self.blocks:
+            if self.gradient_checkpointing and self.training:
+                tokens = self._gradient_checkpointing_func(block.__call__, tokens, mask)
+            else:
+                tokens = block(tokens, mask)
+            if output_hidden_states:
+                hidden_states += (tokens,)
+        tokens = self.post_trunk_norm(tokens)
+        return tokens, hidden_states
+
+
+class AIMv2PretrainedModel(PreTrainedModel):
+    config_class = AIMv2Config
+    base_model_prefix = "aimv2"
+    supports_gradient_checkpointing = True
+    main_input_name = "pixel_values"
+    _no_split_modules = ["AIMv2ViTPreprocessor", "AIMv2Block"]
+    _supports_sdpa = True
+
+
+class AIMv2Model(AIMv2PretrainedModel):
+    def __init__(self, config: AIMv2Config):
+        super().__init__(config)
+        self.preprocessor = AIMv2ViTPreprocessor(config)
+        self.trunk = AIMv2Transformer(config)
+
+    def forward(
+        self,
+        pixel_values: torch.Tensor,
+        mask: Optional[torch.Tensor] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[
+        Tuple[torch.Tensor],
+        Tuple[torch.Tensor, Tuple[torch.Tensor, ...]],
+        BaseModelOutputWithNoAttention,
+    ]:
+        if output_hidden_states is None:
+            output_hidden_states = self.config.output_hidden_states
+        if return_dict is None:
+            return_dict = self.config.use_return_dict
+
+        x = self.preprocessor(pixel_values)
+        x, hidden_states = self.trunk(
+            x, mask, output_hidden_states=output_hidden_states
+        )
+
+        if not return_dict:
+            res = (x,)
+            res += (hidden_states,) if output_hidden_states else ()
+            return res
+
+        return BaseModelOutputWithNoAttention(
+            last_hidden_state=x,
+            hidden_states=hidden_states,
+        )
+

pruning_info.json

@@ -0,0 +1,43 @@
+{
+  "base_model": "AIDC-AI/Ovis2-8B",
+  "method": "layer_pruning_ppl",
+  "method_name": "PPL-based (Shortened LLaMA)",
+  "layers_removed": [
+    4,
+    5,
+    7,
+    8,
+    10,
+    11,
+    13,
+    15
+  ],
+  "n_layers_original": 28,
+  "n_layers_remaining": 20,
+  "params_before_M": 8935.3,
+  "params_after_M": 7070.8,
+  "param_reduction_pct": 20.9,
+  "benchmarks": {
+    "vqav2": {
+      "accuracy": 0.0,
+      "avg_latency_s": 0.5913,
+      "peak_memory_mb": 18675.1,
+      "avg_memory_mb": 18638.7,
+      "throughput_sps": 1.69,
+      "avg_power_w": 289.8,
+      "avg_gpu_util_pct": 94.2,
+      "n_samples": 50,
+      "n_evaluated": 50,
+      "n_skipped": 0,
+      "all_failed": false,
+      "zero_accuracy_warning": true,
+      "metrics": {
+        "exact_match": 0.0,
+        "contains": 0.26,
+        "token_f1": 0.0731,
+        "bleu": 0.0457,
+        "rouge_l": 0.0731
+      }
+    }
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fd169731d2cbde95e10bf356d66d5997fd885dd8dbb6fb4684da3f23b2585d8
+size 11421892

tokenizer_config.json

@@ -0,0 +1,29 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "is_local": false,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}