File size: 1,434 Bytes
9dd0340
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
{}
---
# GGUF quants for [**abacusai/Liberated-Qwen1.5-72B**](https://huggingface.co/abacusai/Liberated-Qwen1.5-72B) using [llama.cpp](https://github.com/ggerganov/llama.cpp)

**Terms of Use**: Please check the [**original model**](https://huggingface.co/abacusai/Liberated-Qwen1.5-72B)

<picture>
<img alt="cthulhu" src="https://huggingface.co/neopolita/common/resolve/main/profile.png">
</picture>

## Quants

* `q2_k`: Uses Q4_K for the attention.vw and feed_forward.w2 tensors, Q2_K for the other tensors.
* `q3_k_s`: Uses Q3_K for all tensors
* `q3_k_m`: Uses Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K
* `q3_k_l`: Uses Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K
* `q4_0`: Original quant method, 4-bit.
* `q4_1`: Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.
* `q4_k_s`: Uses Q4_K for all tensors
* `q4_k_m`: Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q4_K
* `q5_0`: Higher accuracy, higher resource usage and slower inference.
* `q5_1`: Even higher accuracy, resource usage and slower inference.
* `q5_k_s`:  Uses Q5_K for all tensors
* `q5_k_m`: Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q5_K
* `q6_k`: Uses Q8_K for all tensors
* `q8_0`: Almost indistinguishable from float16. High resource use and slow. Not recommended for most users.