Have a ComfyUI workflow? Get a readable Workflow Doc in your browser.Make a workflow readable.Try it now ↗
This is a gemma-4-E2B-it fine-tune, produced through P-E-W's Heretic (v1.2.0) abliteration engine with Arbitrary-Rank Ablation enabled.
Model source
Source excerpt
This is a gemma-4-E2B-it fine-tune, produced through P-E-W's Heretic (v1.2.0) abliteration engine with Arbitrary-Rank Ablation enabled.
Sources
1 sourceVerified Sep 15
Model artifacts
1 artifactSource excerpts
2 excerpts--- library_name: transformers license: apache-2.0 license_link: https://ai.google.dev/gemma/docs/gemma_4_license pipeline_tag: any-to-any tags: - heretic - uncensored - decensored - abliterated - ara base_model: - google/gemma-4-E2B-it --- This is a **gemma-4-E2B-it** fine-tune, produced through P-E-W's [Heretic](https://github.com/p-e-w/heretic) (v1.2.0) abliteration engine with [Arbitrary-Rank Ablation](https://github.com/p-e-w/heretic/pull/211) enabled. **Note:** Model's chat template is updated with Google's fixed version as of May 4th, 2026. --- <img src="https://img.shields.io/badge/RENEGADE_CHAPTER-ARA-4DD2AE?style=flat-square&labelColor=101010" align="right" width="300"> **Heretication Results** | Score Metric | Value | Parameter | Value | | :--- | :--- | :--- | :--- | | **Refusals** | 7/104 | **start_layer_index** | 16 | | **KL Divergence** | 0.0528 | **end_layer_index** | 30 | | **Initial Refusals** | 408/416 | **preserve_good_behavior_weight** | 0.4806 | ||| **steer_bad_behavior_weight** | 0.0002 | ||| **overcorrect_relative_weight** | 0.9682 | ||| **neighbor_count** | 8 | --- **Appendix** > Empty system prompt. <details> <summary>Heretication Rituals</summary> ``` [Trial 331] Refusals: 0/416, KL divergence: 0.3908 [Trial 328] Refusals: 1/416, KL divergence: 0.1424 [Trial 360] Refusals: 2/416, KL divergence: 0.0708 [Trial 375] Refusals: 3/416, KL divergence: 0.0532 » [Trial 99] Refusals: 7/416, KL divergence: 0.0528 [Trial 147] Refusals: 9/416, KL divergence: 0.0289 [Trial 240] Refusals: 34/416, KL divergence: 0.0227 [Trial 190] Refusals: 40/416, KL divergence: 0.0212 [Trial 185] Refusals: 47/416, KL divergence: 0.0191 [Trial 310] Refusals: 86/416, KL divergence: 0.0173 [Trial 307] Refusals: 167/416, KL divergence: 0.0121 [Trial 313] Refusals: 213/416, KL divergence: 0.0120 [Trial 245] Refusals: 297/416, KL divergence: 0.0112 [Trial 220] Refusals: 333/416, KL divergence: 0.0062 [Trial 308] Refusals: 386/416, KL divergence: 0.0034 [Trial 192] Refusals: 408/416, KL divergence: 0.0000 ``` </details> <details> <summary>PIQA Benchmarks</summary> PIQA benchmarks are considered in final trial selection for release. ``` ┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━┓ ┃ Benchmark ┃ Metric ┃ Value ┃ ┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━┩ │ PIQA Base │ acc,none │ 0.5560 │ │ │ acc_stderr,none │ 0.0116 │ │ │ acc_norm,non...
Source context: 19 downloads · 3 likes · Pipeline any-to-any · Library transformers · Repo MuXodious/gemma-4-E2B-it-ARA-heresy
