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
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
|
package llm
import "iter"
type TokenBuffer struct {
tokenizer Tokenizer
window int
stride int
buffer []int64
document int
position int
includeTail bool
}
func NewTokenBuffer(tokenizer Tokenizer, window, stride int) *TokenBuffer {
if stride > window {
panic("stride exceeds window")
}
return &TokenBuffer{
tokenizer: tokenizer,
window: window,
stride: stride,
buffer: make([]int64, 0, 2*window),
document: -1,
position: 0,
includeTail: true,
}
}
func (tb *TokenBuffer) IncludeTail() bool {
return tb.includeTail
}
func (tb *TokenBuffer) SetIncludeTail(includeTail bool) {
tb.includeTail = includeTail
}
func (tb *TokenBuffer) Position() int {
return tb.position
}
func (tb *TokenBuffer) Push(document int, text string) iter.Seq2[[]int64, int] {
return func(yield func([]int64, int) bool) {
if tb.document != -1 && document != tb.document {
tail, seen := tb.Tail()
if len(tail) > 0 && !yield(tail, seen) {
return
}
}
tb.document = document
if text == "" {
return
}
ids := toInt64(tb.tokenizer.Tokenize(text))
tb.buffer = append(tb.buffer, ids...)
for len(tb.buffer) >= tb.window {
w := make([]int64, tb.window)
copy(w, tb.buffer[:tb.window])
if !yield(w, min(tb.Position(), tb.window-tb.stride)) {
return
}
tb.position += tb.stride
tb.buffer = append(tb.buffer[:0], tb.buffer[tb.stride:]...)
}
}
}
func (tb *TokenBuffer) Tail() ([]int64, int) {
seen := min(tb.Position(), min(len(tb.buffer), tb.window-tb.stride))
tb.document = -1
tb.position = 0
if !tb.includeTail || len(tb.buffer) == 0 {
tb.buffer = tb.buffer[:0]
return nil, seen
}
w := make([]int64, len(tb.buffer))
copy(w, tb.buffer)
tb.buffer = tb.buffer[:0]
return w, seen
}
|
