mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

multi-regexp pretokenizer (#12325)

Browse Source
This commit is contained in:
Michael Yang
2025-09-23 13:21:47 -07:00
committed by GitHub
parent 64883e3c4c
commit a40d427bce
12 changed files with 124 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
model/models/gptoss/model.go
View File

@@ -227,17 +227,6 @@ func New(c fs.Config) (model.Model, error) {
m := Transformer{
TransformerBlocks: make([]TransformerBlock, c.Uint("block_count")),
BytePairEncoding: model.NewBytePairEncoding(
c.String("tokenizer.ggml.pretokenizer",
strings.Join([]string{
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]*[\p{Ll}\p{Lm}\p{Lo}\p{M}]+(?i:'s|'t|'re|'ve|'m|'ll|'d)?`,
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]+[\p{Ll}\p{Lm}\p{Lo}\p{M}]*(?i:'s|'t|'re|'ve|'m|'ll|'d)?`,
`\p{N}{1,3}`,
` ?[^\s\p{L}\p{N}]+[\r\n/]*`,
`\s*[\r\n]+`,
`\s+(?!\S)`,
`\s+`,
}, "|"),
),
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -250,6 +239,15 @@ func New(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
strings.Join([]string{
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]*[\p{Ll}\p{Lm}\p{Lo}\p{M}]+(?i:'s|'t|'re|'ve|'m|'ll|'d)?`,
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]+[\p{Ll}\p{Lm}\p{Lo}\p{M}]*(?i:'s|'t|'re|'ve|'m|'ll|'d)?`,
`\p{N}{1,3}`,
` ?[^\s\p{L}\p{N}]+[\r\n/]*`,
`\s*[\r\n]+`,
`\s+(?!\S)`,
`\s+`,
}, "|"),
),
Options: Options{
hiddenSize: int(c.Uint("embedding_length")),

28
model/models/llama/model.go
View File

@@ -54,10 +54,30 @@ func New(c fs.Config) (model.Model, error) {
}
switch c.String("tokenizer.ggml.model") {
case "gpt2":
processor = model.NewBytePairEncoding(
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
&vocabulary,
)
var pretokenizers []string
switch c.String("tokenizer.ggml.pre") {
case "default":
// no-op use the default bpe pretokenizer
case "qwen2":
pretokenizers = []string{
"(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
}
case "refact":
pretokenizers = []string{
`\p{N}`,
`'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+`,
}
case "tekken":
pretokenizers = []string{
"[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]*[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]+|[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]+[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]*|\\p{N}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
}
default:
// use a llama-style pretokenizer
pretokenizers = []string{
"(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
}
}
processor = model.NewBytePairEncoding(&vocabulary, pretokenizers...)
case "llama":
processor = model.NewSentencePiece(&vocabulary)
default:

3
model/models/llama4/model.go
View File

@@ -34,8 +34,6 @@ func (p *Projector) Forward(ctx ml.Context, visionOutputs ml.Tensor) ml.Tensor {
func New(c fs.Config) (model.Model, error) {
m := Model{
BytePairEncoding: model.NewBytePairEncoding(
c.String("tokenizer.ggml.pretokenizer",
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]*[\p{Ll}\p{Lm}\p{Lo}\p{M}]+(?i:'s|'t|'re|'ve|'m|'ll|'d)?|[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]+[\p{Ll}\p{Lm}\p{Lo}\p{M}]*(?i:'s|'t|'re|'ve|'m|'ll|'d)?|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n/]*|\s*[\r\n]+|\s+(?!\S)|\s+`),
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -48,6 +46,7 @@ func New(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]*[\p{Ll}\p{Lm}\p{Lo}\p{M}]+(?i:'s|'t|'re|'ve|'m|'ll|'d)?|[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]+[\p{Ll}\p{Lm}\p{Lo}\p{M}]*(?i:'s|'t|'re|'ve|'m|'ll|'d)?|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n/]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
),
ImageProcessor: newImageProcessor(c),
VisionModel: newVisionModel(c),

2
model/models/mistral3/model.go
View File

@@ -33,7 +33,6 @@ var _ model.TextProcessor = (*Model)(nil)
func New(c fs.Config) (model.Model, error) {
m := &Model{
BytePairEncoding: model.NewBytePairEncoding(
c.String("tokenizer.ggml.pretokenizer", `[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]*[\p{Ll}\p{Lm}\p{Lo}\p{M}]+|[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]+[\p{Ll}\p{Lm}\p{Lo}\p{M}]*|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n/]*|\s*[\r\n]+|\s+(?!\S)|\s+`),
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -46,6 +45,7 @@ func New(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
`[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]*[\p{Ll}\p{Lm}\p{Lo}\p{M}]+|[^\r\n\p{L}\p{N}]?[\p{Lu}\p{Lt}\p{Lm}\p{Lo}\p{M}]+[\p{Ll}\p{Lm}\p{Lo}\p{M}]*|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n/]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
),
TextModel: newTextModel(c),
VisionModel: newVisionModel(c),

2
model/models/mllama/model.go
View File

@@ -33,7 +33,6 @@ const (
func New(c fs.Config) (model.Model, error) {
m := Model{
BytePairEncoding: model.NewBytePairEncoding(
c.String("tokenizer.ggml.pretokenizer", `(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`),
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -46,6 +45,7 @@ func New(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
),
ImageProcessor: newImageProcessor(c),
VisionModel: newVisionModel(c),

2
model/models/qwen2/model.go
View File

@@ -139,7 +139,6 @@ func New(c fs.Config) (model.Model, error) {
m := Model{
Layers: make([]DecoderLayer, c.Uint("block_count")),
BytePairEncoding: model.NewBytePairEncoding(
c.String("tokenizer.ggml.pretokenizer", `(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`),
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -152,6 +151,7 @@ func New(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
),
Options: Options{
hiddenSize: int(c.Uint("embedding_length")),

2
model/models/qwen25vl/model.go
View File

@@ -29,7 +29,6 @@ var _ model.MultimodalProcessor = (*Model)(nil)
func New(c fs.Config) (model.Model, error) {
m := &Model{
BytePairEncoding: model.NewBytePairEncoding(
c.String("tokenizer.ggml.pretokenizer", `(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`),
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -42,6 +41,7 @@ func New(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
),
TextModel: NewTextModel(c),
VisionModel: newVisionModel(c),

2
model/models/qwen3/embed.go
View File

@@ -35,7 +35,6 @@ func newEmbed(c fs.Config) (model.Model, error) {
}
m := embedModel{
BytePairEncoding: model.NewBytePairEncoding(
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -48,6 +47,7 @@ func newEmbed(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
),
Model: &Model{
Layers: layers,

2
model/models/qwen3/model.go
View File

@@ -200,7 +200,6 @@ func New(c fs.Config) (model.Model, error) {
m := Model{
BytePairEncoding: model.NewBytePairEncoding(
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
&model.Vocabulary{
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
@@ -213,6 +212,7 @@ func New(c fs.Config) (model.Model, error) {
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
`(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+`,
),
Layers: layers,
Options: &Options{