File size: 1,386 Bytes
cd78bf4
 
 
 
 
 
 
 
 
 
 
 
e2bc689
 
 
 
 
 
 
cd78bf4
 
 
 
 
 
 
 
 
 
 
 
 
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
---
base_model: Undi95/Meta-Llama-3.1-8B-Claude
library_name: transformers
quantized_by: InferenceIllusionist
tags:
- iMat
- gguf
- llama3
---
<img src="https://i.imgur.com/P68dXux.png" width="400"/>

# Meta-Llama-3.1-8B-Claude-iMat-GGUF
>[!TIP]
><h2>7/28 Update:</h2>
>
>* Reconverted using llama.cpp [b3479](https://github.com/ggerganov/llama.cpp/releases?page=1), adds llama 3.1 rope scaling factors to llama conversion and inference, improving results for context windows above 8192
>* Importance matrix re-calculated with updated fp16 gguf
>* If using Kobold.cpp make sure you are on [v1.71.1](https://github.com/LostRuins/koboldcpp/releases/tag/v1.71.1) or later to take advantage of rope scaling


Quantized from Meta-Llama-3.1-8B-Claude fp16
* Weighted quantizations were creating using fp16 GGUF and groups_merged.txt in 88 chunks and n_ctx=512
* Static fp16 will also be included in repo
* For a brief rundown of iMatrix quant performance please see this [PR](https://github.com/ggerganov/llama.cpp/pull/5747)
* <i>All quants are verified working prior to uploading to repo for your safety and convenience</i>

<b>KL-Divergence Reference Chart</b>
 (Click on image to view in full size)
[<img src="https://i.imgur.com/mV0nYdA.png" width="920"/>](https://i.imgur.com/mV0nYdA.png)


Original model card can be found [here](https://huggingface.co/Undi95/Meta-Llama-3.1-8B-Claude)