matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 07:46:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

model: handle multiple eos tokens (#10577)

Browse Source 
* get eos_token_id from generation_config.json

* refactor

* include both ids and strings in trace

* comments

* remove special case for gemma3 special vocab (#10743)
This commit is contained in:
Michael Yang
2025-05-16 13:40:23 -07:00
committed by GitHub
parent 27da2cddc5
commit 333e360422
18 changed files with 282 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
model/models/gemma2/model.go
View File

@@ -43,10 +43,13 @@ func New(c fs.Config) (model.Model, error) {
Values: c.Strings("tokenizer.ggml.tokens"),
Scores: c.Floats("tokenizer.ggml.scores"),
Types: c.Ints("tokenizer.ggml.token_type"),
BOS: int32(c.Uint("tokenizer.ggml.bos_token_id")),
EOS: int32(c.Uint("tokenizer.ggml.eos_token_id")),
// TODO: set EOT to EOS otherwise 0 will stop generation
EOT: int32(c.Uint("tokenizer.ggml.eos_token_id")),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
EOS: append(
[]int32{int32(c.Uint("tokenizer.ggml.eos_token_id"))},
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
),
Layers: make([]Layer, c.Uint("block_count")),

12
model/models/gemma3/model.go
View File

@@ -60,12 +60,16 @@ func New(c fs.Config) (model.Model, error) {
Values: c.Strings("tokenizer.ggml.tokens"),
Scores: c.Floats("tokenizer.ggml.scores"),
Types: c.Ints("tokenizer.ggml.token_type"),
BOS: int32(c.Uint("tokenizer.ggml.bos_token_id")),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
EOS: int32(1),
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
EOT: int32(106),
AddEOT: c.Bool("tokenizer.ggml.add_eot_token", false),
EOS: append(
[]int32{
int32(c.Uint("tokenizer.ggml.eos_token_id")),
int32(c.Uint("tokenizer.ggml.eot_token_id", 106)),
},
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
),
ImageProcessor: newImageProcessor(c),

10
model/models/llama/model.go
View File

@@ -43,13 +43,13 @@ func New(c fs.Config) (model.Model, error) {
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
Merges: c.Strings("tokenizer.ggml.merges"),
BOS: int32(c.Uint("tokenizer.ggml.bos_token_id")),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
EOS: int32(c.Uint("tokenizer.ggml.eos_token_id")),
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
// TODO: set EOT to EOS otherwise 0 will stop generation
EOT: int32(c.Uint("tokenizer.ggml.eos_token_id")),
AddEOT: c.Bool("tokenizer.ggml.add_eos_token", false),
EOS: append(
[]int32{int32(c.Uint("tokenizer.ggml.eos_token_id"))},
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
),
Layers: make([]Layer, c.Uint("block_count")),

10
model/models/llama4/model.go
View File

@@ -40,13 +40,13 @@ func New(c fs.Config) (model.Model, error) {
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
Merges: c.Strings("tokenizer.ggml.merges"),
BOS: int32(c.Uint("tokenizer.ggml.bos_token_id")),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
EOS: int32(c.Uint("tokenizer.ggml.eos_token_id")),
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
// TODO: set EOT to EOS otherwise 0 will stop generation
EOT: int32(c.Uint("tokenizer.ggml.eos_token_id")),
AddEOT: c.Bool("tokenizer.ggml.add_eos_token", false),
EOS: append(
[]int32{int32(c.Uint("tokenizer.ggml.eos_token_id"))},
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
),
ImageProcessor: newImageProcessor(c),

18
model/models/mistral3/model.go
View File

@@ -37,25 +37,25 @@ func New(c fs.Config) (model.Model, error) {
}
m := &Model{
TextModel: textModel,
VisionModel: newVisionModel(c),
ImageProcessor: newImageProcessor(c),
MultiModalProjector: newMultiModalProjector(c),
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"),
Merges: c.Strings("tokenizer.ggml.merges"),
BOS: int32(c.Uint("tokenizer.ggml.bos_token_id", 1)),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
EOS: int32(c.Uint("tokenizer.ggml.eos_token_id", 2)),
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
// TODO: set EOT to EOS otherwise 0 will stop generation
EOT: int32(c.Uint("tokenizer.ggml.eos_token_id")),
AddEOT: c.Bool("tokenizer.ggml.add_eos_token", false),
EOS: append(
[]int32{int32(c.Uint("tokenizer.ggml.eos_token_id"))},
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
),
TextModel: textModel,
VisionModel: newVisionModel(c),
ImageProcessor: newImageProcessor(c),
MultiModalProjector: newMultiModalProjector(c),
}
m.Cache = kvcache.NewCausalCache(m.TextModel.Shift)

10
model/models/mllama/model.go
View File

@@ -38,13 +38,13 @@ func New(c fs.Config) (model.Model, error) {
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
Merges: c.Strings("tokenizer.ggml.merges"),
BOS: int32(c.Uint("tokenizer.ggml.bos_token_id")),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
EOS: int32(c.Uint("tokenizer.ggml.eos_token_id")),
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
// TODO: set EOT to EOS otherwise 0 will stop generation
EOT: int32(c.Uint("tokenizer.ggml.eos_token_id")),
AddEOT: c.Bool("tokenizer.ggml.add_eos_token", false),
EOS: append(
[]int32{int32(c.Uint("tokenizer.ggml.eos_token_id"))},
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
),
ImageProcessor: newImageProcessor(c),

11
model/models/qwen25vl/model.go
View File

@@ -34,12 +34,13 @@ func New(c fs.Config) (model.Model, error) {
Values: c.Strings("tokenizer.ggml.tokens"),
Types: c.Ints("tokenizer.ggml.token_type"),
Merges: c.Strings("tokenizer.ggml.merges"),
BOS: int32(c.Uint("tokenizer.ggml.bos_token_id")),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", false),
EOS: int32(c.Uint("tokenizer.ggml.eos_token_id")),
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
EOT: int32(c.Uint("tokenizer.ggml.eos_token_id")),
AddEOT: c.Bool("tokenizer.ggml.add_eos_token", false),
EOS: append(
[]int32{int32(c.Uint("tokenizer.ggml.eos_token_id"))},
c.Ints("tokenizer.ggml.eos_token_ids")...,
),
},
),
TextModel: NewTextModel(c),