File size: 3,659 Bytes
e9b7cd7 e45696b 1e8adb2 9ef2e1d e9b7cd7 9ef2e1d e9b7cd7 9ef2e1d e9b7cd7 9ef2e1d e9b7cd7 | 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 69 70 71 72 | ---
base_model: llmfan46/gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic
base_model_relation: quantized
quantized_by: DBMe
library_name: exllamav3
pipeline_tag: text-generation
license: apache-2.0
tags:
- exl3
- exllamav3
- quantized
- text-generation
---
# gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic-exl3-mul1
EXL3 (ExLlamaV3) quantizations of [llmfan46/gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic](https://huggingface.co/llmfan46/gemma-4-31B-it-qat-q4_0-unquantized-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 |
|---|---|---|---|---|---|---|---|---|
| 3.5 | h6 | [3.5bpw_h6](https://huggingface.co/DBMe/gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic-exl3-mul1/tree/3.5bpw_h6) | 2335.9614 | N/A | N/A | N/A | N/A | N/A |
| 3.75 | h6 | [3.75bpw_h6](https://huggingface.co/DBMe/gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic-exl3-mul1/tree/3.75bpw_h6) | 2312.4667 | N/A | N/A | N/A | N/A | N/A |
| 4.0 | h6 | [4.0bpw_h6](https://huggingface.co/DBMe/gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic-exl3-mul1/tree/4.0bpw_h6) | 2097.6864 | N/A | N/A | N/A | N/A | N/A |
| 5.0 | h6 | [5.0bpw_h6](https://huggingface.co/DBMe/gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic-exl3-mul1/tree/5.0bpw_h6) | 1995.7812 | N/A | N/A | N/A | N/A | N/A |
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-4-31B-it-qat-q4_0-unquantized-uncensored-heretic-exl3-mul1 --revision 4.0bpw_h6 --local-dir gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic-exl3-mul1-4.0bpw_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` to the local folder name you downloaded the branch into)*
* **[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:** mul1
- **Output Scales:** always
- **Calibration Rows:** 250
- **Calibration Cols:** 2048
- **Calibration Dataset:** ExLlamaV3 Default (Wiki/C4/Code)
- **High Quality (HQ) Mode:** False
- **ExLlamaV3:** `1.0.0` (Commit: `cb7f2e9`)
- **Hardware:** `NVIDIA RTX PRO 6000 Blackwell Server Edition`
</details>
|