Function _load_lora_into_text_encoder

src/diffusers/loaders/lora_base.py:321–432

(
    state_dict,
    network_alphas,
    text_encoder,
    prefix=None,
    lora_scale=1.0,
    text_encoder_name="text_encoder",
    adapter_name=None,
    _pipeline=None,
    low_cpu_mem_usage=False,
    hotswap: bool = False,
    metadata=None,
)

def _load_lora_into_text_encoder(
    state_dict,
    network_alphas,
    text_encoder,
    prefix=None,
    lora_scale=1.0,
    text_encoder_name="text_encoder",
    adapter_name=None,
    _pipeline=None,
    low_cpu_mem_usage=False,
    hotswap: bool = False,
    metadata=None,
):
    from ..hooks.group_offloading import _maybe_remove_and_reapply_group_offloading

    if not USE_PEFT_BACKEND:
        raise ValueError("PEFT backend is required for this method.")

    if network_alphas and metadata:
        raise ValueError("`network_alphas` and `metadata` cannot be specified both at the same time.")

    peft_kwargs = {}
    if low_cpu_mem_usage:
        if not is_peft_version(">=", "0.13.1"):
            raise ValueError(
                "`low_cpu_mem_usage=True` is not compatible with this `peft` version. Please update it with `pip install -U peft`."
            )
        if not is_transformers_version(">", "4.45.2"):
            # Note from sayakpaul: It's not in `transformers` stable yet.
            # https://github.com/huggingface/transformers/pull/33725/
            raise ValueError(
                "`low_cpu_mem_usage=True` is not compatible with this `transformers` version. Please update it with `pip install -U transformers`."
            )
        peft_kwargs["low_cpu_mem_usage"] = low_cpu_mem_usage

    # If the serialization format is new (introduced in https://github.com/huggingface/diffusers/pull/2918),
    # then the `state_dict` keys should have `unet_name` and/or `text_encoder_name` as
    # their prefixes.
    prefix = text_encoder_name if prefix is None else prefix

    # Safe prefix to check with.
    if hotswap and any(text_encoder_name in key for key in state_dict.keys()):
        raise ValueError("At the moment, hotswapping is not supported for text encoders, please pass `hotswap=False`.")

    # Load the layers corresponding to text encoder and make necessary adjustments.
    if prefix is not None:
        state_dict = {k.removeprefix(f"{prefix}."): v for k, v in state_dict.items() if k.startswith(f"{prefix}.")}
        if metadata is not None:
            metadata = {k.removeprefix(f"{prefix}."): v for k, v in metadata.items() if k.startswith(f"{prefix}.")}

    if len(state_dict) > 0:
        logger.info(f"Loading {prefix}.")
        rank = {}
        state_dict = convert_state_dict_to_diffusers(state_dict)

        # convert state dict
        state_dict = convert_state_dict_to_peft(state_dict)