configuration_centurio.py

# coding=utf-8

from transformers.configuration_utils import PretrainedConfig
from transformers import logging, AutoConfig
from transformers import CONFIG_MAPPING


logger = logging.get_logger(__name__)


class CenturioConfig(PretrainedConfig):
    r"""
    Based on LlavaConfig.

    Args:
        vision_config (`str`,  *optional*, timm model, defaults to `vit_so400m_patch14_siglip_384`):
            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.
        adapter_type (`str`, *optional*, defaults to `multiscale-pool`):
            The adapter type.
        adapter_config (`dict`, *optional*, defaults to `None`):
"""

    model_type = "centurio"
    is_composition = True

    def __init__(
        self,
        timm_model="vit_so400m_patch14_siglip_384",
        image_hidden_size=1024,
        text_config=None,
        ignore_index=-100,
        image_token_index=32000,
        adapter_type="multiscale-pool",
        adapter_config=dict(),
        **kwargs,
    ):
        self.ignore_index = ignore_index
        self.image_token_index = image_token_index
        self.adapter_type = adapter_type
        self.adapter_config = adapter_config

        self.timm_model = timm_model
        self.image_hidden_size = image_hidden_size

        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__(**kwargs)


# q = CenturioConfig(
#     text_config=AutoConfig.from_pretrained("Qwen/Qwen2.5-7B-Instruct"),
#     image_token_index=151665,
#     adapter_type="multiscale-pool",
#     adapter_config=dict(adapter_multi_scale=2),
#     attn_implementation="flash_attention_2"
# )
# q.save_pretrained("centurio_qwen")
#
# a = CenturioConfig(
#     text_config=AutoConfig.from_pretrained("CohereForAI/aya-expanse-8b"),
#     image_token_index=255029,
#     adapter_type="multiscale-pool",
#     adapter_config=dict(adapter_multi_scale=2),
#     attn_implementation="flash_attention_2"
# )
# a.save_pretrained("centurio_aya")