theluyuan/DiffSynth-Studio
1
0
Fork 0
mirror of https://github.com/modelscope/DiffSynth-Studio.git synced 2026-03-18 22:08:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

update lora loading in docs

Browse Source
This commit is contained in:
Artiprocher
2026-02-10 10:48:44 +08:00
parent dc94614c80
commit ff10fde47f
4 changed files with 138 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

63
docs/en/Pipeline_Usage/Model_Inference.md
View File

@@ -102,4 +102,65 @@ image.save("image.jpg")
Each model `Pipeline` has different input parameters. Please refer to the documentation for each model.
If the model parameters are too large, causing insufficient VRAM, please enable [VRAM management](/docs/en/Pipeline_Usage/VRAM_management.md).
If the model parameters are too large, causing insufficient VRAM, please enable [VRAM management](/docs/en/Pipeline_Usage/VRAM_management.md).
## Loading LoRA
LoRA is a lightweight model training method that produces a small number of parameters to extend model capabilities. DiffSynth-Studio supports two ways to load LoRA: cold loading and hot loading.
* Cold loading: When the base model does not have [VRAM management](/docs/en/Pipeline_Usage/VRAM_management.md) enabled, LoRA will be fused into the base model weights. In this case, inference speed remains unchanged, but LoRA cannot be unloaded after loading.
```python
from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
import torch
pipe = QwenImagePipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors"),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="text_encoder/model*.safetensors"),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="vae/diffusion_pytorch_model.safetensors"),
],
tokenizer_config=ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/"),
)
lora = ModelConfig(model_id="DiffSynth-Studio/Qwen-Image-LoRA-ArtAug-v1", origin_file_pattern="model.safetensors")
pipe.load_lora(pipe.dit, lora, alpha=1)
prompt = "Exquisite portrait, underwater girl, blue dress flowing, hair floating, translucent light, bubbles surrounding, peaceful face, intricate details, dreamy and ethereal."
image = pipe(prompt, seed=0, num_inference_steps=40)
image.save("image.jpg")
```
* Hot loading: When the base model has [VRAM management](/docs/en/Pipeline_Usage/VRAM_management.md) enabled, LoRA will not be fused into the base model weights. In this case, inference speed will be slower, but LoRA can be unloaded through `pipe.clear_lora()` after loading.
```python
from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
import torch
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cuda",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = QwenImagePipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors", **vram_config),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="text_encoder/model*.safetensors"),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="vae/diffusion_pytorch_model.safetensors"),
],
tokenizer_config=ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/"),
)
lora = ModelConfig(model_id="DiffSynth-Studio/Qwen-Image-LoRA-ArtAug-v1", origin_file_pattern="model.safetensors")
pipe.load_lora(pipe.dit, lora, alpha=1)
prompt = "Exquisite portrait, underwater girl, blue dress flowing, hair floating, translucent light, bubbles surrounding, peaceful face, intricate details, dreamy and ethereal."
image = pipe(prompt, seed=0, num_inference_steps=40)
image.save("image.jpg")
pipe.clear_lora()
```

9
docs/en/QA.md
View File

@@ -25,4 +25,11 @@ Even with suitable hardware conditions, we currently have no plans to support na
* The main challenge of native FP8 precision training is precision overflow caused by gradient explosion. To ensure training stability, the model structure needs to be redesigned accordingly. However, no model developers are willing to do so at present.
* Additionally, models trained with native FP8 precision can only be computed with BF16 precision during inference without Hopper architecture GPUs, theoretically resulting in generation quality inferior to FP8.
Therefore, native FP8 precision training technology is extremely immature. We will observe the technological developments in the open-source community.
Therefore, native FP8 precision training technology is extremely immature. We will observe the technological developments in the open-source community.
## How to dynamically load LoRA models during inference?
We support two loading methods for LoRA models. See [LoRA Loading](/docs/en/Pipeline_Usage/Model_Inference.md#loading-lora) for details:
* Cold Loading: When [VRAM Management](/docs/en/Pipeline_Usage/VRAM_management.md) is not enabled for the base model, LoRA will be fused into the base model weights. In this case, inference speed remains unchanged, and LoRA cannot be unloaded after loading.
* Hot Loading: When [VRAM Management](/docs/en/Pipeline_Usage/VRAM_management.md) is enabled for the base model, LoRA will not be fused into the base model weights. In this case, inference speed will slow down, and LoRA can be unloaded after loading via `pipe.clear_lora()`.

61
docs/zh/Pipeline_Usage/Model_Inference.md
View File

@@ -103,3 +103,64 @@ image.save("image.jpg")
每个模型 `Pipeline` 的输入参数不同，请参考各模型的文档。
如果模型参数量太大，导致显存不足，请开启[显存管理](/docs/zh/Pipeline_Usage/VRAM_management.md)。
## 加载 LoRA
LoRA 是一种轻量化的模型训练方式，产生少量参数，扩展模型的能力。DiffSynth-Studio 的 LoRA 加载有两种方式：冷加载和热加载。
* 冷加载：当基础模型未开启[显存管理](/docs/zh/Pipeline_Usage/VRAM_management.md)时，LoRA 会融合进基础模型权重，此时推理速度没有变化，LoRA 加载后无法卸载。
```python
from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
import torch
pipe = QwenImagePipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors"),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="text_encoder/model*.safetensors"),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="vae/diffusion_pytorch_model.safetensors"),
],
tokenizer_config=ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/"),
)
lora = ModelConfig(model_id="DiffSynth-Studio/Qwen-Image-LoRA-ArtAug-v1", origin_file_pattern="model.safetensors")
pipe.load_lora(pipe.dit, lora, alpha=1)
prompt = "精致肖像，水下少女，蓝裙飘逸，发丝轻扬，光影透澈，气泡环绕，面容恬静，细节精致，梦幻唯美。"
image = pipe(prompt, seed=0, num_inference_steps=40)
image.save("image.jpg")
```
* 热加载：当基础模型开启[显存管理](/docs/zh/Pipeline_Usage/VRAM_management.md)时，LoRA 不会融合进基础模型权重，此时推理速度会变慢，LoRA 加载后可通过 `pipe.clear_lora()` 卸载。
```python
from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
import torch
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cuda",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = QwenImagePipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors", **vram_config),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="text_encoder/model*.safetensors"),
ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="vae/diffusion_pytorch_model.safetensors"),
],
tokenizer_config=ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/"),
)
lora = ModelConfig(model_id="DiffSynth-Studio/Qwen-Image-LoRA-ArtAug-v1", origin_file_pattern="model.safetensors")
pipe.load_lora(pipe.dit, lora, alpha=1)
prompt = "精致肖像，水下少女，蓝裙飘逸，发丝轻扬，光影透澈，气泡环绕，面容恬静，细节精致，梦幻唯美。"
image = pipe(prompt, seed=0, num_inference_steps=40)
image.save("image.jpg")
pipe.clear_lora()
```

7
docs/zh/QA.md
View File

@@ -26,3 +26,10 @@
* 此外，使用原生 FP8 精度训练的模型，在推理时若没有 Hopper 架构 GPU，则只能以 BF16 精度进行计算，理论上其生成效果反而不如 FP8。
因此，原生 FP8 精度训练技术是极不成熟的，我们静观开源社区的技术发展。
## 如何在推理时动态加载 LoRA 模型？
我们支持 LoRA 模型的两种加载方式，详见[LoRA 加载](/docs/zh/Pipeline_Usage/Model_Inference.md#加载-lora)：
* 冷加载：当基础模型未开启[显存管理](/docs/zh/Pipeline_Usage/VRAM_management.md)时，LoRA 会融合进基础模型权重，此时推理速度没有变化，LoRA 加载后无法卸载。
* 热加载：当基础模型开启[显存管理](/docs/zh/Pipeline_Usage/VRAM_management.md)时，LoRA 不会融合进基础模型权重，此时推理速度会变慢，LoRA 加载后可通过 `pipe.clear_lora()` 卸载。