Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: Qwen/Qwen3-8B
|
| 3 |
+
tags:
|
| 4 |
+
- unlearning
|
| 5 |
+
- simnpo
|
| 6 |
+
- cleanslate
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# fsid_non_clustered_0.05_threshold_qwen-8b-unlearnt
|
| 10 |
+
|
| 11 |
+
SimNPO-unlearned checkpoint of [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B).
|
| 12 |
+
|
| 13 |
+
**Method**: SimNPO (Simple Noise-contrastive Policy Optimization)
|
| 14 |
+
**Forget set threshold**: 0.05 memorization fraction
|
| 15 |
+
**Training**: FSDP (2x A100-80GB), gradient checkpointing, 1 epoch
|
| 16 |
+
|
| 17 |
+
See [unlearning-cleanslate/fsid_qwen-8b](https://huggingface.co/datasets/unlearning-cleanslate/fsid_qwen-8b) for evaluation artifacts, scores, and validation results.
|