mirrors/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-05-11 18:36:41 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

llama: update llama.cpp vendor code to commit d7cfe1ff (#9356)

Browse source
This commit is contained in:
Jeffrey Morgan 2025-02-26 20:34:44 -08:00 • committed by GitHub
parent 2db96c18e7
commit d7d7e99662
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
149 changed files with 18215 additions and 11009 deletions
Download patch file Download diff file Expand all files Collapse all files

23
llama/llama.go
View file

@ -14,6 +14,7 @@ package llama
#include "llama.h"
#include "clip.h"
#include "llava.h"
#include "gguf.h"
#include "mllama.h"
#include "sampling_ext.h"
@ -293,29 +294,29 @@ func NewContextWithModel(model *Model, params ContextParams) (*Context, error) {
}
func (m *Model) NumVocab() int {
return int(C.llama_n_vocab(m.c))
return int(C.llama_n_vocab(m.Vocab()))
}
func (m *Model) TokenIsEog(token int) bool {
return bool(C.llama_token_is_eog(m.c, C.llama_token(token)))
return bool(C.llama_token_is_eog(m.Vocab(), C.llama_token(token)))
}
func (m *Model) AddBOSToken() bool {
return bool(C.llama_add_bos_token(m.c))
return bool(C.llama_add_bos_token(m.Vocab()))
}
func (m *Model) ApplyLoraFromFile(context *Context, loraPath string, scale float32, threads int) error {
cLoraPath := C.CString(loraPath)
defer C.free(unsafe.Pointer(cLoraPath))
loraAdapter := C.llama_lora_adapter_init(m.c, cLoraPath)
loraAdapter := C.llama_adapter_lora_init(m.c, cLoraPath)
if loraAdapter == nil {
return errors.New("unable to load lora")
}
err := -1
if loraAdapter != nil {
err = int(C.llama_lora_adapter_set(context.c, loraAdapter, C.float(scale)))
err = int(C.llama_set_adapter_lora(context.c, loraAdapter, C.float(scale)))
}
if err != 0 {
return errors.New("error applying lora from file")
@ -324,6 +325,10 @@ func (m *Model) ApplyLoraFromFile(context *Context, loraPath string, scale float
return nil
}
func (m *Model) Vocab() *C.struct_llama_vocab {
return C.llama_model_get_vocab(m.c)
}
type Batch struct {
c C.struct_llama_batch
batchSize int
@ -414,7 +419,7 @@ func (m *Model) TokenToPiece(token int) string {
tokenLen := 12
buf := make([]byte, tokenLen)
tokenLen = int(C.llama_token_to_piece(
m.c,
m.Vocab(),
C.int32_t(token),
(*C.char)(unsafe.Pointer(&buf[0])),
C.int32_t(tokenLen),
@ -426,7 +431,7 @@ func (m *Model) TokenToPiece(token int) string {
buf = make([]byte, tokenLen)
C.llama_token_to_piece(
m.c,
m.Vocab(),
C.int32_t(token),
(*C.char)(unsafe.Pointer(&buf[0])),
C.int32_t(tokenLen),
@ -444,7 +449,7 @@ func (m *Model) Tokenize(text string, addSpecial bool, parseSpecial bool) ([]int
defer C.free(unsafe.Pointer(cText))
result := C.llama_tokenize(
m.c,
m.Vocab(),
cText,
C.int32_t(len(text)),
&cTokens[0],
@ -458,7 +463,7 @@ func (m *Model) Tokenize(text string, addSpecial bool, parseSpecial bool) ([]int
maxTokens = int(-result)
cTokens = make([]C.llama_token, maxTokens)
result = C.llama_tokenize(
m.c,
m.Vocab(),
cText,
C.int32_t(len(text)),
&cTokens[0],