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
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
|
package main
import (
"fmt"
"log"
"math"
"strings"
"sync"
"go.jknobloc.com/x/dataset"
"go.jknobloc.com/x/llm"
)
type pplResult struct {
v float64
n int
}
func perplexity() {
d := data()
m := model()
t := tokenizer()
e := llm.NewEvaluator(m, t, func(job llm.Job, logits [][]float32, tokens []int) pplResult {
p, n := llm.NegLogLikelihood(logits, tokens)
return pplResult{
v: p,
n: n,
}
})
results := e.Results()
var wg sync.WaitGroup
total := float64(0)
n := 0
wg.Add(1)
go func() {
defer wg.Done()
for r := range results {
total += r.v
n += r.n
}
}()
if err := e.Run(d, 1024, 512); err != nil {
log.Fatal(err)
}
wg.Wait()
avg := total / float64(n)
ppl := math.Exp(avg)
fmt.Println(ppl)
}
func joined() dataset.Reader {
miniPile := data()
docs := make([]string, 0)
for d := range miniPile.Texts("text") {
docs = append(docs, d)
}
j := dataset.NewStringReader(strings.Join(docs, "\n\n"))
return j
}
|
