File size: 843 Bytes
34d7aa7
 
 
 
 
 
 
 
 
 
 
 
 
82e7306
34d7aa7
 
 
 
80a9cdf
82e7306
34d7aa7
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
---
library_name: transformers
tags:
- llama
- facebook
- meta
- llama-3
- conversational
- text-generation-inference
---
Official [AQLM](https://arxiv.org/abs/2401.06118) quantization of [meta-llama/Meta-Llama-3.1-8B-Instruct
](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct) finetuned with [PV-Tuning](https://arxiv.org/abs/2405.14852).

For this quantization, we used 1 codebook of 16 bits and groupsize of 8.

Results: 
| Model      | Quantization | MMLU (5-shot) | ArcC| ArcE| Hellaswag | PiQA | Winogrande | Model size, Gb |
|------|------|-------|------|------|------|------|------|------|
|meta-llama/Meta-Llama-3.1-8B-Instruct| None | 0.6817 | 0.5162 | 0.8186 | 0.5909 | 0.8014	| 0.7364 | 16.1 |
|  | 1x16g8 | 0.5917 | 0.4377 | 0.7681 | 0.5233	| 0.7748 | 0.6867 | 4.1 |

**Note**

We used `lm-eval=0.4.0` for evaluation.