File size: 1,705 Bytes
5c59423 925065d 5c59423 925065d 5c59423 8289085 5c59423 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 | ---
tags:
- heretic
- uncensored
- abliterated
- gguf
license: mit
base_model: microsoft/phi-4
---
# phi-4-heretic
Abliterated (uncensored) version of [microsoft/phi-4](https://huggingface.co/microsoft/phi-4),
created using [Heretic](https://github.com/p-e-w/heretic) and converted to GGUF.
## Abliteration Quality
| Metric | Value |
|:-------|------:|
| Refusals | 4/100 |
| KL Divergence | 0.0499 |
| Rounds | 2 |
Lower refusals = fewer refused prompts. Lower KL divergence = closer to original model behavior.
## Available Quantizations
| Quantization | File | Size |
|:-------------|:-----|-----:|
| Q8_0 | [phi-4-heretic-Q8_0.gguf](./phi-4-heretic-Q8_0.gguf) | 14.51 GB |
| Q6_K | [phi-4-heretic-Q6_K.gguf](./phi-4-heretic-Q6_K.gguf) | 11.20 GB |
| Q4_K_M | [phi-4-heretic-Q4_K_M.gguf](./phi-4-heretic-Q4_K_M.gguf) | 8.43 GB |
## Usage with Ollama
```bash
ollama run hf.co/ThalisAI/phi-4-heretic:Q8_0
ollama run hf.co/ThalisAI/phi-4-heretic:Q6_K
ollama run hf.co/ThalisAI/phi-4-heretic:Q4_K_M
```
## Full Precision Weights
This repo contains GGUF quantizations only. For full-precision bf16 weights, see the original model at [microsoft/phi-4](https://huggingface.co/microsoft/phi-4).
## About
This model was processed by the **Apostate** automated abliteration pipeline:
1. The source model was loaded in bf16
2. Heretic's optimization-based abliteration was applied to remove refusal behavior
3. The merged model was converted to GGUF format using llama.cpp
4. Multiple quantization levels were generated
The abliteration process uses directional ablation to remove the model's refusal directions
while minimizing KL divergence from the original model's behavior on harmless prompts.
|