froggeric commited on
Commit
c3f1ad9
·
verified ·
1 Parent(s): fc98cf0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +51 -0
README.md CHANGED
@@ -5,6 +5,57 @@ language:
5
 
6
  # Input files for generating the Importance Matrix
7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8
  ## How to quantize with an imatrix in llama.cpp
9
 
10
  1. Get one of the input files collected here, or eleswhere.
 
5
 
6
  # Input files for generating the Importance Matrix
7
 
8
+
9
+ ## Which file to use for generating the importance matrix
10
+
11
+ Not all importance matrices are equal. The best results are obtained when using a source file similar to the
12
+ training data. Size also matters: the bigger the model, the bigger the source file needs to be to make an impact.
13
+ Multiple input files can be combined if needed, for example with the cat command:
14
+ ```
15
+ cat technical.txt multilingual.txt wiki.txt >custom.txt
16
+ ```
17
+
18
+ You will find below descriptions for the various input files provided, to help you choose the correct one.
19
+
20
+ ## Community provided files
21
+
22
+ ### 8k_random_data
23
+
24
+ ### 20k_random_data
25
+
26
+ ### groups_merged
27
+
28
+ ### group_10_merged
29
+
30
+ ### ptb.train
31
+
32
+ ## exllamav2 calibration data
33
+
34
+ https://github.com/turboderp/exllamav2/tree/master/conversion/standard_cal_data
35
+
36
+ ### c4
37
+
38
+ ### code
39
+
40
+ Programming
41
+
42
+ ### multilingual
43
+
44
+ English, Arabic, Chinese, French, German, Japanese, Polish, Russian, Spanish, Swedish, Turkish, Hebrew,
45
+ Macedonian, Norwegian, Lithuanian, Greek, Italian, Afrikaans, Dutch, Danish.
46
+
47
+ ### technical
48
+
49
+ Technical writing.
50
+
51
+ ### tiny
52
+
53
+ Very short stories.
54
+
55
+ ### wiki
56
+
57
+ Wikipedia dump.
58
+
59
  ## How to quantize with an imatrix in llama.cpp
60
 
61
  1. Get one of the input files collected here, or eleswhere.