nisten commited on
Commit
fd3c28f
1 Parent(s): c935bad

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +12 -4
README.md CHANGED
@@ -17,7 +17,14 @@ cat mix4ns-0000* > mix4ns.gguf
17
  ```
18
  careful this can take 5 minutes or up to 10-15 on slow instances, check progress with ls -la
19
 
 
20
 
 
 
 
 
 
 
21
  # Perplexity benchmarks
22
 
23
  Command I used to run these on 48 core CPU only machine, you can add -ngl 16 to offload 16 layers or more to gpu on your own.
@@ -54,13 +61,13 @@ perplexity regular q4_km (no imatrix): 108.59 seconds per pass
54
  [1]2.6100,[2]3.1304,[3]3.6897,[4]3.3500,[5]2.8118,[6]2.5992,[7]2.4349,[8]2.3816,[9]2.4174,[10]2.3959,[11]2.3988,[12]2.3976,
55
  Final estimate: PPL = 2.3976 +/- 0.07111
56
 
57
- perplexity EdgeQuant iq4-ns edgequant (no imatrix) 84.45 seconds per pass - FILESIZE 77258 MB
58
  [1]2.7195,[2]3.1821,[3]3.7177,[4]3.3017,[5]2.8012,[6]2.6034,[7]2.4318,[8]2.3747,[9]2.4160,[10]2.3931,[11]2.4023,[12]2.4013,
59
  Final estimate: PPL = 2.4013 +/- 0.07116
60
 
61
- perplexity EdgeQuant iq4-ns edgequant (WITH imatrix) 82.76 seconds per pass - FILESIZE 75626 MB ( mix4ns.gguf )
62
- [1]2.7175,[2]3.1616,[3]3.6859,[4]3.3012,[5]2.7780,[6]2.5755,[7]2.4093,[8]2.3523,[9]2.3885,[10]2.3757,[11]2.3859,[12]2.3843,
63
- Final estimate: PPL = 2.3843 +/- 0.07036
64
 
65
  perplexity 2K (no imatrix) 207.70 seconds per pass - FILESIZE 47564MB (mix2k-noimatrix-but-usable-reference.gguf)
66
  [1]2.9401,[2]3.4224,[3]4.0174,[4]3.8503,[5]3.5607,[6]3.4449,[7]3[9]3.5589,[10]3.6546,[11]3.7810,[12]3.7733,
@@ -72,5 +79,6 @@ Final estimate: PPL = 3.7733 +/- 0.13299
72
  command to run these was:
73
  ```
74
  ./main -m mix4ns.gguf -n 256 -t 48 --temp 0.5 --color -p "How to build a city on mars via shipping through aldrin cycler orbits?"
 
75
  ```
76
 
 
17
  ```
18
  careful this can take 5 minutes or up to 10-15 on slow instances, check progress with ls -la
19
 
20
+ # Run with llama.cpp
21
 
22
+ ```
23
+ git clone https://github.com/ggerganov/llama.cpp && cd llama.cpp/ && make -j
24
+
25
+ ./main -m ~/mix4ns-00001-of-00005.gguf -n 256 -t 64 --temp 0.2 --color -p "How to build a city on mars via aldrin cycler orgbits?"
26
+
27
+ ```
28
  # Perplexity benchmarks
29
 
30
  Command I used to run these on 48 core CPU only machine, you can add -ngl 16 to offload 16 layers or more to gpu on your own.
 
61
  [1]2.6100,[2]3.1304,[3]3.6897,[4]3.3500,[5]2.8118,[6]2.5992,[7]2.4349,[8]2.3816,[9]2.4174,[10]2.3959,[11]2.3988,[12]2.3976,
62
  Final estimate: PPL = 2.3976 +/- 0.07111
63
 
64
+ perplexity EdgeQuant iq4-ns (no imatrix) 84.45 seconds per pass - FILESIZE 77258 MB
65
  [1]2.7195,[2]3.1821,[3]3.7177,[4]3.3017,[5]2.8012,[6]2.6034,[7]2.4318,[8]2.3747,[9]2.4160,[10]2.3931,[11]2.4023,[12]2.4013,
66
  Final estimate: PPL = 2.4013 +/- 0.07116
67
 
68
+ perplexity EdgeQuant iq4-ns (WITH imatrix) 82.76 seconds per pass - FILESIZE 73636 MB ( mix4ns.gguf ) //BEST ONE FOR 80GB CARD
69
+ [1]2.7166,[2]3.1720,[3]3.6988,[4]3.3195,[5]2.7949,[6]2.5862,[7]2.4186,[8]2.3621,[9]2.3981,[10]2.3876,[11]2.3971,[12]2.3973,
70
+ Final estimate: PPL = 2.3973 +/- 0.07080
71
 
72
  perplexity 2K (no imatrix) 207.70 seconds per pass - FILESIZE 47564MB (mix2k-noimatrix-but-usable-reference.gguf)
73
  [1]2.9401,[2]3.4224,[3]4.0174,[4]3.8503,[5]3.5607,[6]3.4449,[7]3[9]3.5589,[10]3.6546,[11]3.7810,[12]3.7733,
 
79
  command to run these was:
80
  ```
81
  ./main -m mix4ns.gguf -n 256 -t 48 --temp 0.5 --color -p "How to build a city on mars via shipping through aldrin cycler orbits?"
82
+
83
  ```
84