vllm-cpu-amxbf16 0.9.1__cp312-cp312-manylinux_2_17_x86_64.whl

vllm/transformers_utils/configs/nemotron.py

@@ -0,0 +1,205 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Copyright 2024 HuggingFace Inc. team. All rights reserved.
+# Copyright (c) 2024, NVIDIA CORPORATION. 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.
+"""Nemotron model configuration"""
+
+from transformers import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class NemotronConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a
+    [`NemotronModel`]. It is used to instantiate an Nemotron model
+    according to the specified arguments, defining the model architecture.
+    Instantiating a configuration with the defaults will yield a similar
+    configuration to that of the Nemotron-8B.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be
+    used to control the model outputs. Read the documentation from
+    [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 256000):
+            Vocabulary size of the Nemotron model. Defines the number of
+            different tokens that can be represented by the
+            `inputs_ids` passed when calling [`NemotronModel`]
+        hidden_size (`int`, *optional*, defaults to 6144):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 24576):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 48):
+            Number of attention heads for each attention layer in the
+            Transformer decoder.
+        head_dim (`int`, *optional*):
+            Projection weights dimension in multi-head attention. Set to
+            hidden_size // num_attention_heads if None
+        num_key_value_heads (`int`, *optional*):
+            This is the number of key_value heads that should be used to
+            implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use
+            Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention
+            (MQA) otherwise GQA is used. When converting a multi-head
+            checkpoint to a GQA checkpoint, each group key and value
+            head should be constructed by meanpooling all the original
+            heads within that group. For more details checkout 
+            [this paper](https://arxiv.org/pdf/2305.13245.pdf). If it
+            is not specified, will default to `num_attention_heads`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"relu2"`):
+            The non-linear activation function (function or string) in the
+            decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 4096):
+            The maximum sequence length that this model might ever be used
+            with.
+        initializer_range (`float`, *optional*, defaults to 0.0134):
+            The standard deviation of the truncated_normal_initializer for
+            initializing all weight matrices.
+        norm_eps (`float`, *optional*, defaults to 1e-05):
+            The epsilon used by the normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values
+            attentions (not used by all models). Only relevant if
+            `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 2):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 3):
+            End of stream token id.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        partial_rotary_factor (`float`, *optional*, defaults to 0.5):
+            Percentage of the query and keys which will have rotary embedding.
+        attention_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output
+            projection layers during self-attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        mlp_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use a bias in up_proj and down_proj layers in the MLP
+            layers.
+
+    ```python
+    >>> from transformers import NemotronModel, NemotronConfig
+    >>> # Initializing a Nemotron nemotron-15b style configuration
+    >>> configuration = NemotronConfig()
+    >>> # Initializing a model from the nemotron-15b style configuration
+    >>> model = NemotronModel(configuration)
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "nemotron"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=256000,
+        hidden_size=6144,
+        intermediate_size=24576,
+        num_hidden_layers=32,
+        num_attention_heads=48,
+        head_dim=None,
+        num_key_value_heads=None,
+        hidden_act="relu2",
+        max_position_embeddings=4096,
+        initializer_range=0.0134,
+        norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=2,
+        eos_token_id=3,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        partial_rotary_factor=0.5,
+        attention_bias=False,
+        attention_dropout=0.0,
+        mlp_bias=False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        head_dim = head_dim or kwargs.get("kv_channels")
+        self.head_dim = head_dim if head_dim is not None else (
+            hidden_size // num_attention_heads)
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.norm_eps = norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        # for backward compatibility
+        partial_rotary_factor = kwargs.get("rope_percent") or kwargs.get(
+            "rope_percentage") or partial_rotary_factor
+        self.partial_rotary_factor = partial_rotary_factor
+        self._rope_scaling_validation()
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        self.mlp_bias = mlp_bias
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    def _rope_scaling_validation(self):
+        """
+        Validate the `rope_scaling` configuration.
+        """
+        if self.rope_scaling is None:
+            return
+
+        if not isinstance(self.rope_scaling, dict) or len(
+                self.rope_scaling) != 2:
+            raise ValueError(
+                "`rope_scaling` must be a dictionary with two fields, "
+                f"`type` and `factor`, got {self.rope_scaling}")
+        rope_scaling_type = self.rope_scaling.get("type", None)
+        rope_scaling_factor = self.rope_scaling.get("factor", None)
+        if rope_scaling_type is None or rope_scaling_type not in [
+                "linear", "dynamic"
+        ]:
+            raise ValueError(
+                "`rope_scaling`'s type field must be one of ['linear', "
+                f"'dynamic'], got {rope_scaling_type}")
+        if rope_scaling_factor is None or not isinstance(
+                rope_scaling_factor, float) or rope_scaling_factor <= 1.0:
+            raise ValueError(
+                "`rope_scaling`'s factor field must be a float > 1, got "
+                f"{rope_scaling_factor}")

vllm/transformers_utils/configs/nemotron_h.py

@@ -0,0 +1,258 @@
+# SPDX-License-Identifier: Apache-2.0
+
+# Copyright 2024 HuggingFace Inc. team. All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION. 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.
+"""NemotronH model configuration"""
+
+import regex as re
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class NemotronHConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a
+    [`NemotronHModel`]. It is used to instantiate a NemotronH model according
+    to the specified arguments, defining the model architecture. Instantiating
+    a configuration with the defaults will yield a similar configuration to
+    that of the NemotronH-v0.1 model.
+    Args:
+        vocab_size (`int`, *optional*, defaults to 131072):
+            Vocabulary size of the NemotronH model. Defines the number of
+            different tokens that can be represented by the `inputs_ids`
+            passed when calling [`NemotronHModel`]
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be
+            tied. Note that this is only relevant if the model has a output
+            word embedding layer.
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 21504):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 52):
+            Number of hidden layers in the Transformer encoder.
+        hybrid_override_pattern (`str`, *optional*, defaults to
+            `"M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M-"`):
+            The pattern of the hybrid model. The pattern is a string of
+            characters where each character represents
+            M: Mamba2, *: Attention, -: MLP
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the
+            Transformer encoder.
+        attention_head_dim (`int`, *optional*, defaults to 128):
+            Dimension of each attention head.
+        num_key_value_heads (`int`, *optional*, defaults to 8):
+            This is the number of key_value heads that should be used to
+            implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use
+            Multi Head Attention (MHA), if `num_key_value_heads=1` the model
+            will use Multi Query Attention (MQA) otherwise GQA is used.
+        mlp_hidden_act (`str`, *optional*, defaults to "relu2"):
+            The non-linear activation function in the MLP layers.
+        attention_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in attention layers.
+        mlp_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in MLP layers.
+        use_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in the model.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for
+            initializing all weight matrices.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        residual_in_fp32 (`bool`, *optional*, defaults to `False`):
+            Whether or not residuals should be in `float32`. If set to `False`
+            residuals will keep the same `dtype` as the rest of the model.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values
+            attentions (not used by all models). Only relevant if
+            `config.is_decoder=True`.
+        num_logits_to_keep (`int` or `None`, *optional*, defaults to 1):
+            Number of prompt logits to calculate during generation. If `None`,
+            all logits will be calculated. If an integer value, only last
+            `num_logits_to_keep` logits will be calculated.
+        pad_token_id (`int`, *optional*, defaults to 0):
+            The id of the padding token.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            The id of the "beginning-of-sequence" token.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            The id of the "end-of-sequence" token.
+        sliding_window (`int`, *optional*, defaults to None):
+            Sliding window attention window size.
+        max_position_embeddings (`int`, *optional*, defaults to 4096):
+            The maximum sequence length that this model might ever be used
+            with.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        hidden_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the hidden states.
+        use_mamba_kernels (`bool`, *optional*, defaults to `True`):
+            Flag indicating whether or not to use the fast mamba kernels.
+            These are available only if `mamba-ssm` and `causal-conv1d`
+            are installed, and the mamba modules are running on a CUDA device.
+        ssm_state_size (`int`, *optional*, defaults to 128):
+            The dimension of the mamba state space latents.
+        mamba_num_heads (`int`, *optional*, defaults to 128):
+            Number of heads in Mamba layers.
+        mamba_n_groups (`int`, *optional*, defaults to 8):
+            Number of groups in Mamba layers.
+        mamba_head_dim (`int`, *optional*, defaults to 64):
+            Dimension of each Mamba head.
+        mamba_d_conv (`int`, *optional*, defaults to 4):
+            The size of the mamba convolution kernel.
+        mamba_expand (`int`, *optional*, defaults to 2):
+            Expanding factor used to determine the mamba intermediate size.
+        mamba_hidden_act (`str`, *optional*, defaults to "silu"):
+            The non-linear activation function in the Mamba layers.
+        mamba_dt_min (`float`, *optional*, defaults to 0.001):
+            Minimum value for the time step in Mamba.
+        mamba_dt_max (`float`, *optional*, defaults to 0.1):
+            Maximum value for the time step in Mamba.
+        mamba_dt_limit (`tuple`, *optional*, defaults to (0.0, float("inf"))):
+            Limits for the time step in Mamba.
+        mamba_dt_init_floor (`float`, *optional*, defaults to 1e-4):
+            Floor value for time step initialization in Mamba.
+        mamba_conv_bias (`bool`, *optional*, defaults to `True`):
+            Whether to use bias in the convolution layer of the mamba mixer
+            block.
+        mamba_proj_bias (`bool`, *optional*, defaults to `False`):
+            Whether to use bias in the input and output projections of the
+            mamba mixer block.
+        mamba_chunk_size (`int`, *optional*, defaults to 256):
+            Size of chunks for Mamba processing.
+        rescale_prenorm_residual (`bool`, *optional*, defaults to `True`):
+            Whether to rescale the pre-normalization residual connections.
+    """
+
+    model_type = "nemotron_h"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=131072,
+        tie_word_embeddings=False,
+        hidden_size=4096,
+        intermediate_size=21504,
+        num_hidden_layers=52,
+        hybrid_override_pattern="M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M-",
+        num_attention_heads=32,
+        attention_head_dim=128,
+        num_key_value_heads=8,  # nemo: num_query_groups
+        mlp_hidden_act="relu2",
+        attention_bias=False,
+        mlp_bias=False,
+        use_bias=False,
+        initializer_range=0.02,  # nemo: init_method_std
+        layer_norm_epsilon=1e-5,  # nemo: layernorm_epsilon
+        residual_in_fp32=False,  #  Megatron Core default value
+        use_cache=True,
+        num_logits_to_keep=1,
+        pad_token_id=0,
+        bos_token_id=1,
+        eos_token_id=2,
+        sliding_window=None,
+        max_position_embeddings=4096,
+        attention_dropout=0.0,
+        hidden_dropout=0.0,  # * ADDED
+        use_mamba_kernels=True,
+        ssm_state_size=128,  # mamba_state_size
+        mamba_num_heads=128,
+        mamba_n_groups=8,  # nemo: mamba_ssm_ngroups = num_heads
+        mamba_head_dim=64,
+        mamba_d_conv=4,
+        mamba_expand=2,
+        mamba_hidden_act="silu",
+        mamba_dt_min=0.001,
+        mamba_dt_max=0.1,
+        mamba_dt_limit=(0.0, float("inf")),
+        mamba_dt_init_floor=1e-4,
+        mamba_conv_bias=True,
+        mamba_proj_bias=False,
+        mamba_chunk_size=256,
+        rescale_prenorm_residual=True,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.tie_word_embeddings = tie_word_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.hybrid_override_pattern = hybrid_override_pattern
+        self.num_attention_heads = num_attention_heads
+        self.attention_head_dim = attention_head_dim
+        self.sliding_window = sliding_window
+        self.max_position_embeddings = max_position_embeddings
+        self.attention_dropout = attention_dropout
+        self.hidden_dropout = hidden_dropout
+
+        # Validate hybrid_override_pattern
+        # M: Mamba2, *: Attention, -: MLP
+        assert len(self.hybrid_override_pattern) == self.num_hidden_layers, (
+            "hybrid_override_pattern must have same length as "
+            "num_hidden_layers")
+        assert re.match(r"^[*-M]+$", self.hybrid_override_pattern), (
+            "hybrid_override_pattern must only contain characters "
+            "'M', '*', or '-'")
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.mlp_hidden_act = mlp_hidden_act
+        self.attention_bias = attention_bias
+        self.mlp_bias = mlp_bias
+        self.use_bias = use_bias
+        self.initializer_range = initializer_range
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.residual_in_fp32 = residual_in_fp32
+
+        self.use_cache = use_cache
+        self.num_logits_to_keep = num_logits_to_keep
+
+        self.use_mamba_kernels = use_mamba_kernels
+        self.n_groups = mamba_n_groups
+        self.mamba_head_dim = mamba_head_dim
+        self.ssm_state_size = ssm_state_size
+        self.mamba_num_heads = mamba_num_heads
+        self.conv_kernel = mamba_d_conv
+        self.expand = mamba_expand
+        self.mamba_hidden_act = mamba_hidden_act
+        self.time_step_min = mamba_dt_min
+        self.time_step_max = mamba_dt_max
+        self.time_step_limit = mamba_dt_limit
+        self.time_step_floor = mamba_dt_init_floor
+        self.use_conv_bias = mamba_conv_bias
+        self.mamba_proj_bias = mamba_proj_bias
+        self.chunk_size = mamba_chunk_size
+        self.rescale_prenorm_residual = rescale_prenorm_residual
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    @property
+    def layers_block_type(self):
+        return [
+            "mamba" if self.hybrid_override_pattern[i] == "M" else
+            "attention" if self.hybrid_override_pattern[i] == "*" else "mlp"
+            for i in range(self.num_hidden_layers)
+        ]

vllm/transformers_utils/configs/nvlm_d.py

@@ -0,0 +1,15 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# Adapted from
+# https://huggingface.co/nvidia/NVLM-D-72B/blob/main/configuration_nvlm_d.py
+# --------------------------------------------------------
+# NVLM-D
+# Copyright (c) 2024 NVIDIA
+# Licensed under Apache 2.0 License [see LICENSE for details]
+# --------------------------------------------------------
+from .internvl import InternVLChatConfig
+
+
+class NVLM_D_Config(InternVLChatConfig):
+    model_type = 'NVLM_D'

vllm/transformers_utils/configs/ovis.py

@@ -0,0 +1,184 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+# yapf: disable
+# ruff: noqa: E501
+# copied from https://huggingface.co/AIDC-AI/Ovis2-1B/blob/main/configuration_aimv2.py
+# and https://huggingface.co/AIDC-AI/Ovis2-1B/blob/main/configuration_ovis.py
+from typing import Any, Optional, Union
+
+from transformers import AutoConfig, PretrainedConfig
+
+
+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
+
+
+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)
+
+
+# ----------------------------------------------------------------------
+#                     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 = dict[str, Any]()
+        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]
+
+
+class SiglipVisualTokenizerConfig(BaseVisualTokenizerConfig):
+    model_type = "siglip_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("siglip_visual_tokenizer", SiglipVisualTokenizerConfig)
+AutoConfig.register("aimv2_visual_tokenizer", Aimv2VisualTokenizerConfig)
+
+
+# ----------------------------------------------------------------------
+#                           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)
+
+        # map llm_config to text_config
+        self.text_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