gemma-3-1b-it-Abliterated
An abliterated version of google/gemma-3-1b-it, created using AnlordAbliterator 1.4.0.
The model is provided in the original Hugging Face Transformers / Safetensors format and can be used directly with compatible Transformers-based tooling.
Model
Base model: google/gemma-3-1b-it
Format: Safetensors
Parameters: ~1B
Abliteration tool: AnlordAbliterator
Abliteration Results
The model was processed with the AnlordAbliterator pipeline (200 optimization trials).
| Metric | Before | After |
|---|---|---|
| Refusals | 97 / 104 | 5 / 104 |
| KL divergence | 0 | 0.1010 |
Abliteration log
============================================================
ANLORD ABLITERATOR
============================================================
Model: google/gemma-3-1b-it
Status: SUCCESS
Abliteration:
Initial refusals: 97
Final refusals: 5
KL divergence: 0.1010
Note: refusal rate and KL divergence are measurements from the AnlordAbliterator evaluation pipeline (104 evaluation prompts) and should be treated as evaluation results rather than a guarantee of model behavior on all prompts.
Abliteration Parameters
| Parameter | Value |
|---|---|
| direction_scope | global |
| direction_index | 13.424 |
| attn.o_proj.max_weight | 1.945 |
| attn.o_proj.max_weight_position | 12.909 |
| attn.o_proj.min_weight | 1.729 |
| attn.o_proj.min_weight_distance | 9.663 |
| mlp.down_proj.max_weight | 1.426 |
| mlp.down_proj.max_weight_position | 14.765 |
| mlp.down_proj.min_weight | 0.013 |
| mlp.down_proj.min_weight_distance | 2.431 |
Usage
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "anlord/gemma-3-1b-it-Abliterated"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map="auto"
)
Use the model with the same general workflow as the original google/gemma-3-1b-it model.
Reproduce this ablation
This repository ships a reproduce/ folder with everything needed to reproduce this exact
ablation with AnlordAbliterator 1.4.0: the pinned configuration, all ablation parameters,
the refusals/KL scores, SHA-256 checksums and the full Optuna study journal.
AnlordAbliterator --reproduce anlord/gemma-3-1b-it-Abliterated
The pipeline downloads reproduce/reproduce.json from this repository, restores the recorded
settings and parameters, re-applies the ablation to the original base model, exports the model
and verifies the SHA-256 hashes of the weight files — then re-measures refusals and KL and
compares them against the numbers above.
Note: the base model
google/gemma-3-1b-itis gated on Hugging Face — accept the Gemma terms with your own account so reproduction can download it.
Related Repositories
GGUF versions
All available GGUF quantizations are available here:
anlord/gemma-3-1b-it-Abliterated-GGUF
Available quantizations:
- BF16
- F16
- Q8_0
- Q6_K
- Q5_K_M
- Q5_0
- Q4_K_M
- Q4_0
Abliteration Tool
The model was created using:
Base Model & License
This model is derived from google/gemma-3-1b-it.
The original gemma-3-1b-it model is released under the Gemma Terms of Use and is gated on Hugging Face.
Please refer to the original model repository for the applicable license terms.
Original model: https://huggingface.co/google/gemma-3-1b-it
Disclaimer
This is an abliterated derivative of the original gemma-3-1b-it model.
Abliteration changes the model's refusal behavior and may also affect other aspects of its behavior. Use the model responsibly and evaluate it for your intended use case.