Downloads · 30 days
8
10% of all-time downloads
JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget01_PDU
tofu_Llama-3.2-3B-Instruct_forget01_PDU is a text generation model from JoaoBoer. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as llama3.2.
open-unlearning/tofuLlama-3.2-3B-Instructfull unlearned on the TOFU forget01 split with PDU, trained with the open-unlearning framework. Used as a weight-unlearning baseline / draft model in the Speculative-Decoding-U…
Downloads · 30 days
8
10% of all-time downloads
All-time downloads
80
Public
Parameters
3.2B
6.4 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors6.4 GB · 100%
From the Hugging Face model README
open-unlearning/tofu_Llama-3.2-3B-Instruct_full unlearned on the TOFU forget01 split with PDU, trained with the open-unlearning framework. Used as a weight-unlearning baseline / draft model in the Speculative-Decoding-Unlearning project.
Full training config: .hydra/config.yaml. TOFU evaluation outputs: evals/.
gamma: 1.0
alpha: 100
retain_loss_type: NLL
retain_loss_eps: 0.3
primal_dual: True
dual_step_size: 5
dual_update_upon: step
dual_warmup_epochs: 5
loss_names: ['forget_loss', 'retain_loss']
| metric | value |
|---|---|
| exact_memorization | 0.4894 |
| extraction_strength | 0.0565 |
| forget_Q_A_PARA_Prob | 0.0206 |
| forget_Q_A_gibberish | 0.5511 |
| forget_quality | 0.7659 |
| forget_truth_ratio | 0.6825 |
| mia_loss | 0.1997 |
| mia_min_k | 0.3956 |
| mia_min_k_plus_plus | 0.9938 |
| mia_zlib | 0.2619 |
| model_utility | 0.6674 |
| privleak | 36.5819 |