AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

fuse_lora

Browse Source
This commit is contained in:
sayakpaul
2025-09-26 09:05:26 +05:30
parent 769c56af6f
commit ca913f0db4
1 changed files with 15 additions and 435 deletions
Download Patch File Download Diff File Expand all files Collapse all files

450
src/diffusers/loaders/lora_pipeline.py
View File

@@ -982,35 +982,7 @@ class StableDiffusionXLLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -1341,35 +1313,7 @@ class SD3LoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -1602,35 +1546,7 @@ class AuraFlowLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -2163,35 +2079,7 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.lora_state_dict`] for more details.
"""
transformer = getattr(self, self.transformer_name) if not hasattr(self, "transformer") else self.transformer
@@ -2888,35 +2776,7 @@ class CogVideoXLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -3145,35 +3005,7 @@ class Mochi1LoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -3406,35 +3238,7 @@ class LTXVideoLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -3664,35 +3468,7 @@ class SanaLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -3925,35 +3701,7 @@ class HunyuanVideoLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -4187,35 +3935,7 @@ class Lumina2LoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -4519,35 +4239,7 @@ class WanLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -4854,35 +4546,7 @@ class SkyReelsV2LoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -5112,35 +4776,7 @@ class CogView4LoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -5373,35 +5009,7 @@ class HiDreamImageLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,
@@ -5636,35 +5244,7 @@ class QwenImageLoraLoaderMixin(LoraBaseMixin):
**kwargs,
):
r"""
Fuses the LoRA parameters into the original parameters of the corresponding blocks.
<Tip warning={true}>
This is an experimental API.
</Tip>
Args:
components: (`List[str]`): List of LoRA-injectable components to fuse the LoRAs into.
lora_scale (`float`, defaults to 1.0):
Controls how much to influence the outputs with the LoRA parameters.
safe_fusing (`bool`, defaults to `False`):
Whether to check fused weights for NaN values before fusing and if values are NaN not fusing them.
adapter_names (`List[str]`, *optional*):
Adapter names to be used for fusing. If nothing is passed, all active adapters will be fused.
Example:
```py
from diffusers import DiffusionPipeline
import torch
pipeline = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
).to("cuda")
pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
pipeline.fuse_lora(lora_scale=0.7)
```
See [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] for more details.
"""
super().fuse_lora(
components=components,