AUREON: Production-Grade Universal Reversible Learning
Project description
# Reversible: Universal Unlearning with 0.96+ Recovery Guarantee
Fix AI model bias in minutes instead of months.
Installation
\\ash pip install reversible \\
Quick Start
\\python from reversible import Reversible import torch
model = your_pytorch_model optimizer = torch.optim.Adam(model.parameters())
reversible = Reversible(model, optimizer, device='cuda')
CHARGE: Learn baseline
reversible.charge(train_loader, num_epochs=3)
DISCHARGE: Unlearn bias
reversible.discharge(bad_loader, num_epochs=2)
RECHARGE: Recover performance
reversible.recharge(good_loader, num_epochs=10)
VERIFY: Check recovery ratio
metrics = reversible.verify(test_loader) print(f'Recovery Ratio: {metrics["recovery_ratio"]:.3f}') # ~0.96 expected \\
Key Results
- Recovery Ratio: 0.96 ✅ (NLP & Vision validated)
- k-ratio: 0.671 (Hyperuniversal across 66M-355M params)
- Domains tested: 4 (SST-2, BANKING77, RTE, MRPC + Vision + RL)
- Status: Production Ready
License
MIT License
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file reversible_unlearning-0.2.9.tar.gz.
File metadata
- Download URL: reversible_unlearning-0.2.9.tar.gz
- Upload date:
- Size: 4.4 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.13.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
51d45a259c636f7e7737a9c9cb410751023c6dba9039d50a5867bb0f20f4b988
|
|
| MD5 |
ec2c32dd3cb004655bb74d9059fe78f3
|
|
| BLAKE2b-256 |
0a6fd9412a8feead14b837fdf57a78cd311057d034d59eb98adca2205f2b30c6
|
File details
Details for the file reversible_unlearning-0.2.9-py3-none-any.whl.
File metadata
- Download URL: reversible_unlearning-0.2.9-py3-none-any.whl
- Upload date:
- Size: 4.7 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.13.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
7ef273aa4aefa17d799a48280547d449b28fb6b671b3031eed76d2586a18fc2b
|
|
| MD5 |
63e3590f9066c165550cb35519b45d23
|
|
| BLAKE2b-256 |
0abd9a4c27baa910f1502eae81b41f03db4831d7c5a4628103ddb42e7434a77e
|
