mirror of
https://github.com/modelscope/DiffSynth-Studio.git
synced 2026-03-19 06:32:27 +00:00
20 lines
786 B
Python
20 lines
786 B
Python
import torch
|
|
|
|
def decomposite(tensor_A, tensor_B, rank):
|
|
dtype, device = tensor_A.dtype, tensor_A.device
|
|
weight = tensor_B @ tensor_A
|
|
U, S, V = torch.pca_lowrank(weight.float(), q=rank)
|
|
tensor_A = (V.T).to(dtype=dtype, device=device).contiguous()
|
|
tensor_B = (U @ torch.diag(S)).to(dtype=dtype, device=device).contiguous()
|
|
return tensor_A, tensor_B
|
|
|
|
def reset_lora_rank(lora, rank):
|
|
lora_merged = {}
|
|
keys = [i for i in lora.keys() if ".lora_A." in i]
|
|
for key in keys:
|
|
tensor_A = lora[key]
|
|
tensor_B = lora[key.replace(".lora_A.", ".lora_B.")]
|
|
tensor_A, tensor_B = decomposite(tensor_A, tensor_B, rank)
|
|
lora_merged[key] = tensor_A
|
|
lora_merged[key.replace(".lora_A.", ".lora_B.")] = tensor_B
|
|
return lora_merged |