Views
No views yet
A photo-realistic image of a cat6.00.030FlowMatchEulerDiscreteScheduler421024x10241{
2 "bypass_mode": true,
3 "algo": "lokr",
4 "multiplier": 1.0,
5 "full_matrix": true,
6 "linear_dim": 10000,
7 "linear_alpha": 1,
8 "factor": 4,
9 "apply_preset": {
10 "target_module": [
11 "Attention",
12 "FeedForward"
13 ],
14 "module_algo_map": {
15 "FeedForward": {
16 "factor": 4
17 },
18 "Attention": {
19 "factor": 2
20 }
21 }
22 }
23}1import torch
2from diffusers import DiffusionPipeline
3from lycoris import create_lycoris_from_weights
4
5
6def download_adapter(repo_id: str):
7 import os
8 from huggingface_hub import hf_hub_download
9 adapter_filename = "pytorch_lora_weights.safetensors"
10 cache_dir = os.environ.get('HF_PATH', os.path.expanduser('~/.cache/huggingface/hub/models'))
11 cleaned_adapter_path = repo_id.replace("/", "_").replace("\\", "_").replace(":", "_")
12 path_to_adapter = os.path.join(cache_dir, cleaned_adapter_path)
13 path_to_adapter_file = os.path.join(path_to_adapter, adapter_filename)
14 os.makedirs(path_to_adapter, exist_ok=True)
15 hf_hub_download(
16 repo_id=repo_id, filename=adapter_filename, local_dir=path_to_adapter
17 )
18
19 return path_to_adapter_file
20
21model_id = 'stabilityai/stable-diffusion-3.5-medium'
22adapter_repo_id = 'bghira/sd35m-photo-154_seqlen'
23adapter_filename = 'pytorch_lora_weights.safetensors'
24adapter_file_path = download_adapter(repo_id=adapter_repo_id)
25pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16
26lora_scale = 1.0
27wrapper, _ = create_lycoris_from_weights(lora_scale, adapter_file_path, pipeline.transformer)
28wrapper.merge_to()
29
30prompt = "A photo-realistic image of a cat"
31negative_prompt = 'ugly, cropped, blurry, low-quality, mediocre average'
32
33## Optional: quantise the model to save on vram.
34## Note: The model was quantised during training, and so it is recommended to do the same during inference time.
35from optimum.quanto import quantize, freeze, qint8
36quantize(pipeline.transformer, weights=qint8)
37freeze(pipeline.transformer)
38
39pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu') # the pipeline is already in its target precision level
40image = pipeline(
41 prompt=prompt,
42 negative_prompt=negative_prompt,
43 num_inference_steps=30,
44 generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(42),
45 width=1024,
46 height=1024,
47 guidance_scale=6.0,
48).images[0]
49image.save("output.png", format="PNG")