| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| """ Spec-Vision model configuration""" |
|
|
| from typing import Dict, Optional, Union |
|
|
| from transformers.configuration_utils import PretrainedConfig |
| from transformers.utils import logging |
|
|
| logger = logging.get_logger(__name__) |
|
|
|
|
| class SpecVisionConfig(PretrainedConfig): |
| r""" |
| This is the configuration class to store the configuration of a [`SpecVisionModel`]. It is used to instantiate a Spec-Vision |
| 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: |
| vocab_size (`int`, *optional*, defaults to 32064): |
| Vocabulary size of the model. Defines the number of different tokens that can be represented by the |
| `inputs_ids` passed when calling [`SpecVisionModel`]. |
| hidden_size (`int`, *optional*, defaults to 3072): |
| Dimension of the hidden representations. |
| intermediate_size (`int`, *optional*, defaults to 8192): |
| 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 32): |
| Number of attention heads for each attention layer in the Transformer decoder. |
| num_key_value_heads (`int`, *optional*): |
| Number of key/value heads for implementing Grouped Query Attention. |
| resid_pdrop (`float`, *optional*, defaults to 0.0): |
| Dropout probability for MLP outputs. |
| embd_pdrop (`float`, *optional*, defaults to 0.0): |
| The dropout ratio for embeddings. |
| attention_dropout (`float`, *optional*, defaults to 0.0): |
| The dropout ratio after computing attention scores. |
| hidden_act (`str`, *optional*, defaults to `"silu"`): |
| The non-linear activation function 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 0.02): |
| The standard deviation for initializing all weight matrices. |
| rms_norm_eps (`float`, *optional*, defaults to 1e-5): |
| The epsilon value used for RMSNorm. |
| use_cache (`bool`, *optional*, defaults to `True`): |
| Whether to use the past key/values attentions for faster inference. |
| rope_theta (`float`, *optional*, defaults to 10000.0): |
| The base period of the RoPE embeddings. |
| rope_scaling (`dict`, *optional*): |
| Configuration for RoPE scaling strategy. |
| embd_layer (`dict`, *optional*): |
| Configuration for the embedding layer, including image embedding settings. |
| """ |
| model_type = "spec_vision" |
| keys_to_ignore_at_inference = ["past_key_values"] |
|
|
| def __init__( |
| self, |
| vocab_size: int = 32064, |
| hidden_size: int = 3072, |
| intermediate_size: int = 8192, |
| num_hidden_layers: int = 32, |
| num_attention_heads: int = 32, |
| num_key_value_heads: Optional[int] = None, |
| resid_pdrop: float = 0.0, |
| embd_pdrop: float = 0.0, |
| attention_dropout: float = 0.0, |
| hidden_act: str = "silu", |
| max_position_embeddings: int = 4096, |
| initializer_range: float = 0.02, |
| rms_norm_eps: float = 1e-5, |
| use_cache: bool = True, |
| rope_theta: float = 10000.0, |
| rope_scaling: Optional[Dict] = None, |
| embd_layer: Dict[str, Union[str, bool]] = { |
| "embedding_cls": "image", |
| "hd_transform_order": "sub_glb", |
| "projection_cls": "mlp", |
| "use_hd_transform": True, |
| "with_learnable_separator": True |
| }, |
| bos_token_id: int = 1, |
| eos_token_id: int = 32000, |
| pad_token_id: int = 32000, |
| tie_word_embeddings: bool = False, |
| **kwargs, |
| ): |
| self.vocab_size = vocab_size |
| 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_key_value_heads = num_key_value_heads or num_attention_heads |
| self.resid_pdrop = resid_pdrop |
| self.embd_pdrop = embd_pdrop |
| self.attention_dropout = attention_dropout |
| self.hidden_act = hidden_act |
| self.max_position_embeddings = max_position_embeddings |
| self.initializer_range = initializer_range |
| self.rms_norm_eps = rms_norm_eps |
| self.use_cache = use_cache |
| self.rope_theta = rope_theta |
| self.rope_scaling = rope_scaling |
| self.embd_layer = embd_layer |
|
|
| super().__init__( |
| bos_token_id=bos_token_id, |
| eos_token_id=eos_token_id, |
| pad_token_id=pad_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) != 3: |
| raise ValueError( |
| "`rope_scaling` must be a dictionary with three fields, `type`, `short_factor` and `long_factor`, " |
| f"got {self.rope_scaling}" |
| ) |
| |
| rope_scaling_type = self.rope_scaling.get("type", None) |
| rope_scaling_short_factor = self.rope_scaling.get("short_factor", None) |
| rope_scaling_long_factor = self.rope_scaling.get("long_factor", None) |
| |
| if rope_scaling_type is None or rope_scaling_type not in ["su", "yarn"]: |
| raise ValueError(f"`rope_scaling`'s type field must be one of ['su', 'yarn'], got {rope_scaling_type}") |
| |
| head_dim = self.hidden_size // self.num_attention_heads // 2 |
| |
| for factor, name in [(rope_scaling_short_factor, "short_factor"), (rope_scaling_long_factor, "long_factor")]: |
| if not (isinstance(factor, list) and all(isinstance(x, (int, float)) for x in factor)): |
| raise ValueError(f"`rope_scaling`'s {name} field must be a list of numbers, got {factor}") |
| if len(factor) != head_dim: |
| raise ValueError(f"`rope_scaling`'s {name} field must have length {head_dim}, got {len(factor)}") |
