PEFT (original) (raw)

Refer to the Inference with PEFT tutorial for an overview of how to use PEFT in Diffusers for inference.

class diffusers.loaders.PeftAdapterMixin

( )

A class containing all functions for loading and using adapters weights that are supported in PEFT library. For more details about adapters and injecting them in a base model, check out the PEFTdocumentation.

Install the latest version of PEFT, and use this mixin to:

Attach new adapters in the model.
Attach multiple adapters and iteratively activate/deactivate them.
Activate/deactivate all adapters from the model.
Get a list of the active adapters.

Gets the current list of active adapters of the model.

If you are not familiar with adapters and PEFT methods, we invite you to read more about them on the PEFTdocumentation.

add_adapter

< source >

( adapter_config adapter_name: str = 'default' )

Parameters

adapter_config ([~peft.PeftConfig]) — The configuration of the adapter to add; supported adapters are non-prefix tuning and adaption prompt methods.
adapter_name (str, optional, defaults to "default") — The name of the adapter to add. If no name is passed, a default name is assigned to the adapter.

Adds a new adapter to the current model for training. If no adapter name is passed, a default name is assigned to the adapter to follow the convention of the PEFT library.

If you are not familiar with adapters and PEFT methods, we invite you to read more about them in the PEFTdocumentation.

delete_adapters

< source >

( adapter_names: typing.Union[typing.List[str], str] )

Parameters

adapter_names (Union[List[str], str]) — The names (single string or list of strings) of the adapter to delete.

Delete an adapter’s LoRA layers from the underlying model.

Example:

from diffusers import AutoPipelineForText2Image import torch

Disable all adapters attached to the model and fallback to inference with the base model only.

If you are not familiar with adapters and PEFT methods, we invite you to read more about them on the PEFTdocumentation.

Disables the active LoRA layers of the underlying model.

Example:

from diffusers import AutoPipelineForText2Image import torch

Enable adapters that are attached to the model. The model uses self.active_adapters() to retrieve the list of adapters to enable.

If you are not familiar with adapters and PEFT methods, we invite you to read more about them on the PEFTdocumentation.

Enables the active LoRA layers of the underlying model.

Example:

from diffusers import AutoPipelineForText2Image import torch

enable_lora_hotswap

< source >

( target_rank: int = 128 check_compiled: typing.Literal['error', 'warn', 'ignore'] = 'error' )

Parameters

target_rank (int, optional, defaults to 128) — The highest rank among all the adapters that will be loaded.
check_compiled (str, optional, defaults to "error") — How to handle the case when the model is already compiled, which should generally be avoided. The options are:
- “error” (default): raise an error
- “warn”: issue a warning
- “ignore”: do nothing

Enables the possibility to hotswap LoRA adapters.

Calling this method is only required when hotswapping adapters and if the model is compiled or if the ranks of the loaded adapters differ.

load_lora_adapter

< source >

( pretrained_model_name_or_path_or_dict prefix = 'transformer' hotswap: bool = False **kwargs )

Parameters

pretrained_model_name_or_path_or_dict (str or os.PathLike or dict) — Can be either:
- A string, the model id (for example google/ddpm-celebahq-256) of a pretrained model hosted on the Hub.
- A path to a directory (for example ./my_model_directory) containing the model weights saved with ModelMixin.save_pretrained().
- A torch state dict.
prefix (str, optional) — Prefix to filter the state dict.
cache_dir (Union[str, os.PathLike], optional) — Path to a directory where a downloaded pretrained model configuration is cached if the standard cache is not used.
force_download (bool, optional, defaults to False) — Whether or not to force the (re-)download of the model weights and configuration files, overriding the cached versions if they exist.
proxies (Dict[str, str], optional) — A dictionary of proxy servers to use by protocol or endpoint, for example, {'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}. The proxies are used on each request.
local_files_only (bool, optional, defaults to False) — Whether to only load local model weights and configuration files or not. If set to True, the model won’t be downloaded from the Hub.
token (str or bool, optional) — The token to use as HTTP bearer authorization for remote files. If True, the token generated fromdiffusers-cli login (stored in ~/.huggingface) is used.
revision (str, optional, defaults to "main") — The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier allowed by Git.
subfolder (str, optional, defaults to "") — The subfolder location of a model file within a larger model repository on the Hub or locally.
network_alphas (Dict[str, float]) — The value of the network alpha used for stable learning and preventing underflow. This value has the same meaning as the --network_alpha option in the kohya-ss trainer script. Refer to this link.
low_cpu_mem_usage (bool, optional) — Speed up model loading by only loading the pretrained LoRA weights and not initializing the random weights.
hotswap — (bool, optional) Defaults to False. Whether to substitute an existing (LoRA) adapter with the newly loaded adapter in-place. This means that, instead of loading an additional adapter, this will take the existing adapter weights and replace them with the weights of the new adapter. This can be faster and more memory efficient. However, the main advantage of hotswapping is that when the model is compiled with torch.compile, loading the new adapter does not require recompilation of the model. When using hotswapping, the passed adapter_name should be the name of an already loaded adapter.
If the new adapter and the old adapter have different ranks and/or LoRA alphas (i.e. scaling), you need to call an additional method before loading the adapter:

Loads a LoRA adapter into the underlying model.

save_lora_adapter

< source >

( save_directory adapter_name: str = 'default' upcast_before_saving: bool = False safe_serialization: bool = True weight_name: typing.Optional[str] = None )

Parameters

save_directory (str or os.PathLike) — Directory to save LoRA parameters to. Will be created if it doesn’t exist.
adapter_name — (str, defaults to “default”): The name of the adapter to serialize. Useful when the underlying model has multiple adapters loaded.
upcast_before_saving (bool, defaults to False) — Whether to cast the underlying model to torch.float32 before serialization.
save_function (Callable) — The function to use to save the state dictionary. Useful during distributed training when you need to replace torch.save with another method. Can be configured with the environment variableDIFFUSERS_SAVE_MODE.
safe_serialization (bool, optional, defaults to True) — Whether to save the model using safetensors or the traditional PyTorch way with pickle.
weight_name — (str, optional, defaults to None): Name of the file to serialize the state dict with.

Save the LoRA parameters corresponding to the underlying model.

set_adapter

< source >

( adapter_name: typing.Union[str, typing.List[str]] )

Parameters

adapter_name (Union[str, List[str]])) — The list of adapters to set or the adapter name in the case of a single adapter.

Sets a specific adapter by forcing the model to only use that adapter and disables the other adapters.

If you are not familiar with adapters and PEFT methods, we invite you to read more about them on the PEFTdocumentation.

set_adapters

< source >

( adapter_names: typing.Union[typing.List[str], str] weights: typing.Union[float, typing.Dict, typing.List[float], typing.List[typing.Dict], typing.List[NoneType], NoneType] = None )

Parameters

adapter_names (List[str] or str) — The names of the adapters to use.
adapter_weights (Union[List[float], float], optional) — The adapter(s) weights to use with the UNet. If None, the weights are set to 1.0 for all the adapters.

Set the currently active adapters for use in the UNet.

Example:

from diffusers import AutoPipelineForText2Image import torch

pipeline = AutoPipelineForText2Image.from_pretrained( "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16 ).to("cuda") pipeline.load_lora_weights( "jbilcke-hf/sdxl-cinematic-1", weight_name="pytorch_lora_weights.safetensors", adapter_name="cinematic" ) pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel") pipeline.set_adapters(["cinematic", "pixel"], adapter_weights=[0.5, 0.5])