matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-12 16:57:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

model: Don't unconditionally add special tokens

Browse Source 
We sometimes tokenize partial strings. For example, with
multimodal inputs, we split the input string around the images
and then tokenize each piece. In these cases, we should only add
the special tokens on the first piece.
This commit is contained in:
Jesse Gross
2025-03-05 13:27:53 -08:00
committed by Jesse Gross
parent e2252d0fc6
commit b70fc4d51e
4 changed files with 12 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
model/process_text.go
View File

@@ -19,7 +19,7 @@ const (
)
type TextProcessor interface {
Encode(string) ([]int32, error)
Encode(s string, addSpecial bool) ([]int32, error)
Decode([]int32) (string, error)
Is(int32, Special) bool
}
@@ -144,7 +144,7 @@ type merge struct {
runes []rune
}
func (bpe BytePairEncoding) Encode(s string) ([]int32, error) {
func (bpe BytePairEncoding) Encode(s string, addSpecial bool) ([]int32, error) {
fragments := []fragment{{value: s}}
for _, special := range bpe.vocab.SpecialVocabulary() {
// TODO: process special tokens concurrently
@@ -282,7 +282,7 @@ func (bpe BytePairEncoding) Encode(s string) ([]int32, error) {
}
}
if len(ids) > 0 {
if addSpecial && len(ids) > 0 {
if bpe.vocab.AddBOS {
if ids[0] == bpe.vocab.BOS {
slog.Warn("adding bos token to prompt which already has it", "id", bpe.vocab.BOS)