File size: 3,537 Bytes
5e5b263 b8d51ff 99fcdba da3d012 5e5b263 da3d012 5e5b263 | 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 57 58 59 60 61 62 63 64 65 66 67 68 | ---
base_model: llmfan46/gemma-3-12b-it-ultra-uncensored-heretic
base_model_relation: quantized
quantized_by: DBMe
library_name: exllamav3
pipeline_tag: text-generation
license: gemma
tags:
- exl3
---
# DBMe/gemma-3-12b-it-ultra-uncensored-heretic-exl3
EXL3 (ExLlamaV3) quantizations of [llmfan46/gemma-3-12b-it-ultra-uncensored-heretic](https://huggingface.co/llmfan46/gemma-3-12b-it-ultra-uncensored-heretic). All credit for the original model goes to the original authors.
## 📊 Available Quantizations & VRAM
The model weights are stored in separate branches. **Please switch to a branch to download.**
*Note: VRAM estimates include PyTorch context overhead (~0.8GB) and assume an unquantized FP16 KV cache.*
| Target BPW | Head BPW | Branch (Download Link) | WikiText-2 PPL (512 ctx)¹ | 2K ctx | 4K ctx | 8K ctx | 16K ctx | 32K ctx |
|---|---|---|---|---|---|---|---|---|
| 4.0 | h6 | [4.0bpw_h6](https://huggingface.co/DBMe/gemma-3-12b-it-ultra-uncensored-heretic-exl3/tree/4.0bpw_h6) | 9.7746 | ~9.93 GB | ~10.68 GB | ~12.18 GB | ~15.18 GB | ~21.18 GB |
| 5.0 | h6 | [5.0bpw_h6](https://huggingface.co/DBMe/gemma-3-12b-it-ultra-uncensored-heretic-exl3/tree/5.0bpw_h6) | 9.7767 | ~11.19 GB | ~11.94 GB | ~13.44 GB | ~16.44 GB | ~22.44 GB |
| 6.0 | h6 | [6.0bpw_h6](https://huggingface.co/DBMe/gemma-3-12b-it-ultra-uncensored-heretic-exl3/tree/6.0bpw_h6) | 9.7294 | ~12.44 GB | ~13.19 GB | ~14.69 GB | ~17.69 GB | ~23.69 GB |
| 8.0 | h8 | [8.0bpw_h8](https://huggingface.co/DBMe/gemma-3-12b-it-ultra-uncensored-heretic-exl3/tree/8.0bpw_h8) | 9.7336 | ~15.18 GB | ~15.93 GB | ~17.43 GB | ~20.43 GB | ~26.43 GB |
¹ *Evaluated against WikiText-2 with ExLlamaV3 using a strided 512-token context window (-c 512) in llama.cpp parity mode (-g). Lower is better.*
*(Higher BPW = higher quality, lower BPW = fits in less VRAM).*
## 📥 How to Download
It's recommended to use the `huggingface-cli` to download specific branches. *(Do not use `git clone` as it will download all branches!)*
Ensure you have the CLI installed:
```bash
pip install -U "huggingface_hub[cli]"
```
Download a specific branch (e.g., `4.0bpw_h6`):
```bash
# Example: Downloading the 4.0bpw_h6 branch
huggingface-cli download DBMe/gemma-3-12b-it-ultra-uncensored-heretic-exl3 --revision 4.0bpw_h6 --local-dir gemma-3-12b-it-ultra-uncensored-heretic-exl3-4.0bpw_h6
```
## 💻 Supported Engines
These models are highly optimized for modern GPUs and can be run using:
* **[TabbyAPI](https://github.com/theroyallab/tabbyAPI):** A fast, OpenAI-compatible API server. *(Set `model_name: "gemma-3-12b-it-ultra-uncensored-heretic-exl3-<BranchName>"` in your config)*
* **[Text-Generation-WebUI](https://github.com/oobabooga/text-generation-webui):** A local web interface. *(Select the `exllamav3` loader)*
* **[ExLlamaV3 (Native)](https://github.com/turboderp-org/exllamav3):** Python library for custom integration.
### 📈 Perplexity Degradation Curve
*(Lower is better)*
<details>
<summary><b>⚙️ Advanced: Quantization Environment & Settings</b></summary>
### 🔬 Quantization Settings
- **Codebook:** mcg
- **Output Scales:** always
- **Calibration Rows:** 250
- **Calibration Cols:** 2048
- **Calibration Dataset:** ExLlamaV3 Default (Wiki/C4/Code)
- **High Quality (HQ) Mode:** False
- **ExLlamaV3:** `0.0.29` (Commit: `cb1a436`)
- **Hardware:** `NVIDIA RTX PRO 6000 Blackwell Server Edition`
</details>
|