FearL0rd
9aba0997ac
Keep replica blocks out of the nn.Module tree so model.to(cuda:0) cannot steal cuda:1 weights. Reuse the live home GPU model, skip wrapping it so ComfyUI can apply LoRA, and bake adapters only onto independent clones. Preserve operations in clone config for the fast incremental path.
6 lines
47 B
Plaintext
6 lines
47 B
Plaintext
*.zip
|
|
__pycache__/
|
|
*.pyc
|
|
.pytest_cache/
|
|
tests/
|