File size: 739 Bytes
19089d1
 
c5bd2db
 
 
 
 
 
 
 
 
 
 
 
 
 
19089d1
38f4b1d
 
 
 
7259d33
b83031a
 
 
7259d33
b83031a
7259d33
b83031a
7259d33
b83031a
 
 
7259d33
b83031a
7259d33
b83031a
7259d33
b83031a
7259d33
b83031a
38f4b1d
 
 
b83031a
38f4b1d
b0266ff
 
 
 
0fc188e
 
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
---
license: mit
language:
- en
tags:
- Q
- Q2
- Q3
- Q4
- Q5
- Q6
- Q12
- Q8
- Q16
- Q32
- Dront
---

DrontGPTQ is a simple quantizer for GPT-like models.

Supports:

 2 bit ✅

 3 bit ✅
 
 4 bit ✅
 
 5 bit ✅
 
 6 bit ✅

 7 bit ❌
 
 8 bit ✅
 
 12 bit ✅
 
 16 bit ✅
 
 32 bit ✅

Consists of just two files.

DrontQ.py – a very simple quantizer. Converts HF F32 to .dq format (Dront quant).

DQuse.py – a simple way to run .dq models.


*To use this, scroll to the end of the files; you will find the main variables there, which you should adjust to your needs.*
*Quantizes local models—specifically GPT-like ones.*
**Most Dront gpt-like models will also be released in DQ quantization! Especially the new VdrontV**