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
|
package main
import (
"fmt"
"log"
"go.jknobloc.com/x/dataset"
"go.jknobloc.com/x/llmc"
"go.jknobloc.com/x/shelf"
"go.jknobloc.com/x/tokenizer/bpe"
)
func tinyShakespeare() {
var reader *dataset.FileReader
if r, err := dataset.NewFileReader(shelf.Abs("data/tinyshakespeare/input.txt"), "*.txt"); err != nil {
log.Fatal(err)
} else {
reader = r
}
reader.SetDelimiter("\n\n")
var tokenizer *bpe.Tokenizer
v := shelf.Abs("models/gpt2/vocab.json")
m := shelf.Abs("models/gpt2/merges.txt")
if t, err := bpe.NewTokenizerFromFiles(v, m, bpe.DefaultConfig()); err != nil {
log.Fatal(err)
} else {
tokenizer = t
}
result := make([]int64, 0)
for _, doc := range reader.Texts() {
ids := tokenizer.Tokenize(doc)
toks := make([]int64, len(ids)+1)
toks[0] = 50256 // end of text
for i, id := range ids {
toks[i+1] = int64(id)
}
result = append(result, toks...)
}
val := llmc.DataFile[int64]{
Model: llmc.GPT2,
Tokens: result[:32768],
}
train := llmc.DataFile[int64]{
Model: llmc.GPT2,
Tokens: result[32768:],
}
if _, err := llmc.Serialize(&val, shelf.Abs("llmc/tinyshakespeare/tiny_shakespeare_val.bin")); err != nil {
log.Fatal(err)
}
if _, err := llmc.Serialize(&train, shelf.Abs("llmc/tinyshakespeare/tiny_shakespeare_train.bin")); err != nil {
log.Fatal(err)
}
fmt.Println(val)
}
|
