File size: 5,979 Bytes
6407cfd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 | from typing import Any, Optional, List, Union
from transformers import Qwen3Config, Qwen3MoeConfig, Qwen3NextConfig
from transformers.configuration_utils import PretrainedConfig
__all__ = ["Siglip2NavitConfig", "Ovis2_6_Config"]
class Siglip2NavitConfig(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 = "siglip2_navit"
def __init__(
self,
hidden_size: int = 1024,
intermediate_size: int = 4096,
num_hidden_layers: int = 24,
num_attention_heads: int = 16,
num_channels: int = 3,
num_patches: int = -1,
image_size: int = 512,
patch_size: int = 16,
hidden_act: str="gelu_pytorch_tanh",
layer_norm_eps: float = 1e-6,
attention_dropout: float = 0.0,
hidden_stride: int = 2,
window_size: int = 112,
fullatt_block_indexes: Optional[list] = None,
temporal_patch_size: int = 1,
preserve_original_pe: bool = True,
use_rope: bool = True,
**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.num_patches = num_patches
self.patch_size = patch_size
self.image_size = image_size
self.hidden_act = hidden_act
self.attention_dropout = attention_dropout
self.layer_norm_eps = layer_norm_eps
self.hidden_stride = hidden_stride
self.window_size = window_size
self.fullatt_block_indexes = fullatt_block_indexes
self.temporal_patch_size = temporal_patch_size
self.preserve_original_pe = preserve_original_pe
self.use_rope = use_rope
class Ovis2_6_Config(PretrainedConfig):
model_type = "ovis2_6"
sub_configs = dict(llm_config=Qwen3Config, vit_config=Siglip2NavitConfig)
def __init__(self,
llm_config: Optional[Union[Qwen3Config, dict]] = None,
vit_config: Optional[Union[Siglip2NavitConfig, dict]] = None,
visual_vocab_size=65536,
hidden_size=None,
**kwargs
):
super().__init__(**kwargs)
if isinstance(llm_config, dict):
llm_config = Qwen3Config(**llm_config)
self.llm_config = llm_config
if isinstance(vit_config, dict):
vit_config = Siglip2NavitConfig(**vit_config)
self.vit_config = vit_config
self.visual_vocab_size = visual_vocab_size
self.hidden_size = hidden_size
if kwargs.get('attn_implementation'):
self.llm_config._attn_implementation = kwargs['attn_implementation']
self.vit_config._attn_implementation = kwargs['attn_implementation']
class Ovis2_6_Moe_Config(PretrainedConfig):
model_type = "ovis2_6_moe"
sub_configs = dict(llm_config=Qwen3MoeConfig, vit_config=Siglip2NavitConfig)
def __init__(self,
llm_config: Optional[Union[Qwen3MoeConfig, dict]] = None,
vit_config: Optional[Union[Siglip2NavitConfig, dict]] = None,
visual_vocab_size=65536,
hidden_size=None,
**kwargs
):
super().__init__(**kwargs)
if isinstance(llm_config, dict):
llm_config = Qwen3MoeConfig(**llm_config)
self.llm_config = llm_config
if isinstance(vit_config, dict):
vit_config = Siglip2NavitConfig(**vit_config)
self.vit_config = vit_config
self.visual_vocab_size = visual_vocab_size
self.hidden_size = hidden_size
if kwargs.get('attn_implementation'):
self.llm_config._attn_implementation = kwargs['attn_implementation']
self.vit_config._attn_implementation = kwargs['attn_implementation']
class Ovis2_6_Next_Config(PretrainedConfig):
model_type = "ovis2_6_next"
sub_configs = dict(llm_config=Qwen3NextConfig, vit_config=Siglip2NavitConfig)
def __init__(self,
llm_config: Optional[Union[Qwen3NextConfig, dict]] = None,
vit_config: Optional[Union[Siglip2NavitConfig, dict]] = None,
visual_vocab_size=65536,
hidden_size=None,
**kwargs
):
super().__init__(**kwargs)
if isinstance(llm_config, dict):
llm_config = Qwen3NextConfig(**llm_config)
self.llm_config = llm_config
if isinstance(vit_config, dict):
vit_config = Siglip2NavitConfig(**vit_config)
self.vit_config = vit_config
self.visual_vocab_size = visual_vocab_size
self.hidden_size = hidden_size
if kwargs.get('attn_implementation'):
self.llm_config._attn_implementation = kwargs['attn_implementation']
self.vit_config._attn_implementation = kwargs['attn_implementation']
|