Downloads · 30 days
10
29% of all-time downloads
shirasko/qwen3.5-2b-crisp-culture-of-greece
qwen3.5-2b-crisp-culture-of-greece is a text generation model from shirasko. Use it when you need the model to write or continue text. It is set up for peft.
Downloads · 30 days
10
29% of all-time downloads
All-time downloads
34
Public
Repo size
24.9 MB
Likes
0
Public
Click a slice to open those files.
.json20 MB · 80%
From the Hugging Face model README
| Field | Value |
|---|---|
| Unlearning method | CRISP |
| Base model | Qwen/Qwen3.5-2B |
| Target concept | Culture of Greece |
| Checkpoint type | LoRA Adapter |
| Rank / seed | 100 / 42 |
| Train eval protocol | mc |
Selected hyperparameters (from unlearned_checkpoints.json):
| Parameter | Value |
|---|---|
alpha | 20 |
delta_embed | 0 |
k_features | 20 |
k_features_embed | 0 |
layer_hi | 14 |
layer_lo | 4 |
layer_step | 2 |
lora_rank | 4 |
lr | 0.0001 |
n_tokens_edited | 0 |
num_epochs | 2 |
Headline scores used for checkpoint selection:
| Metric | Train (after unlearning) | Test (after unlearning) |
|---|---|---|
| Efficacy | 0.488 | 0.195 |
| Specificity | 0.903 | 0.848 |
| Harmonic mean | 0.633 | 0.317 |
| Relearning QA (MC) | — | 0.68 |
From evaluation/score_comparison.csv:
| Metric | Baseline (train) | After unlearn (train) | Baseline (test) | After unlearn (test) |
|---|---|---|---|---|
| QA accuracy | 0.66 | 0.46 | 0.66 | 0.58 |
| QA fraction | 1 | 0.512 | 1 | 0.805 |
| SimDom accuracy | 0.88 | 0.84 | 0.82 | 0.74 |
| SimDom fraction | 1 | 0.937 | 1 | 0.86 |
| MMLU accuracy | 0.56 | 0.52 | 0.588 | 0.533 |
| MMLU fraction | 1 | 0.871 | 1 | 0.837 |
| File | Description |
|---|---|
unlearned_checkpoints.json | Checkpoint metadata & hyperparameters |
evaluation/evaluation_summary.json | Full evaluation payload (train/test/relearning) |
evaluation/score_comparison.csv | Baseline vs. unlearned comparison table |