Downloads · 30 days
81
6% of all-time downloads
RichardErkhov/Undi95_-_Unholy-v1-12L-13B-gguf
Undi95_-_Unholy-v1-12L-13B-gguf is a machine learning model from RichardErkhov. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Downloads · 30 days
81
6% of all-time downloads
All-time downloads
1.4K
Public
Repo size
147 GB
Likes
0
Public
Click a slice to open those files.
.gguf171 GB · 100%
From the Hugging Face model README
Quantization made by Richard Erkhov.
Unholy-v1-12L-13B - GGUF
| Name | Quant method | Size |
|---|---|---|
| Unholy-v1-12L-13B.Q2_K.gguf | Q2_K | 4.52GB |
| Unholy-v1-12L-13B.IQ3_XS.gguf | IQ3_XS | 4.99GB |
| Unholy-v1-12L-13B.IQ3_S.gguf | IQ3_S | 5.27GB |
| Unholy-v1-12L-13B.Q3_K_S.gguf | Q3_K_S | 5.27GB |
| Unholy-v1-12L-13B.IQ3_M.gguf | IQ3_M | 5.57GB |
| Unholy-v1-12L-13B.Q3_K.gguf | Q3_K | 5.9GB |
| Unholy-v1-12L-13B.Q3_K_M.gguf | Q3_K_M | 5.9GB |
| Unholy-v1-12L-13B.Q3_K_L.gguf | Q3_K_L | 6.45GB |
| Unholy-v1-12L-13B.IQ4_XS.gguf | IQ4_XS | 6.54GB |
| Unholy-v1-12L-13B.Q4_0.gguf | Q4_0 | 6.86GB |
| Unholy-v1-12L-13B.IQ4_NL.gguf | IQ4_NL | 6.9GB |
| Unholy-v1-12L-13B.Q4_K_S.gguf | Q4_K_S | 6.91GB |
| Unholy-v1-12L-13B.Q4_K.gguf | Q4_K | 7.33GB |
| Unholy-v1-12L-13B.Q4_K_M.gguf | Q4_K_M | 7.33GB |
| Unholy-v1-12L-13B.Q4_1.gguf | Q4_1 | 7.61GB |
| Unholy-v1-12L-13B.Q5_0.gguf | Q5_0 | 8.36GB |
| Unholy-v1-12L-13B.Q5_K_S.gguf | Q5_K_S | 8.36GB |
| Unholy-v1-12L-13B.Q5_K.gguf | Q5_K | 8.6GB |
| Unholy-v1-12L-13B.Q5_K_M.gguf | Q5_K_M | 8.6GB |
| Unholy-v1-12L-13B.Q5_1.gguf | Q5_1 | 9.1GB |
| Unholy-v1-12L-13B.Q6_K.gguf | Q6_K | 9.95GB |
| Unholy-v1-12L-13B.Q8_0.gguf | Q8_0 | 12.88GB |
license: cc-by-nc-4.0 tags:

[HIGHLY EXPERIMENTAL]
(Sister model: https://huggingface.co/Undi95/Unholy-v1-10L-13B)
Use at your own risk, I'm not responsible for any usage of this model, don't try to do anything this model tell you to do.
Uncensored.
If you are censored, it's maybe because of keyword like "assistant", "Factual answer", or other "sweet words" like I call them that trigger the censoring accross all the layer of the model (since they're all trained on some of them in a way).
12L : This is a test project, uukuguy/speechless-llama2-luban-orca-platypus-13b and jondurbin/spicyboros-13b-2.2 was used for a merge, then, I deleted the first 8 layers to add 8 layers of MLewd at the beginning, and do the same from layers 16 to 20, trying to break all censoring possible, before merging the output with MLewd at 0.33 weight.
<!-- description start -->This repo contains fp16 files of Unholy v1, an uncensored model.
<!-- description end --> <!-- description start -->Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{prompt}
### Response:
Exemple:

Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 50.65 |
| ARC (25-shot) | 63.57 |
| HellaSwag (10-shot) | 83.75 |
| MMLU (5-shot) | 58.08 |
| TruthfulQA (0-shot) | 51.09 |
| Winogrande (5-shot) | 77.27 |
| GSM8K (5-shot) | 11.07 |
| DROP (3-shot) | 9.73 |