baalajimaestro/ollama
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

more free

Browse source
This commit is contained in:
Michael Yang 2023-07-06 11:24:54 -07:00 committed by Jeffrey Morgan
parent d436d51c78
commit 2c80eddd71
1 changed files with 24 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

32
llama/llama.go
View file

@ -36,7 +36,7 @@ import (
) )
type LLama struct { type LLama struct {
ctx unsafe.Pointer ctx unsafe.Pointer
embeddings bool embeddings bool
contextSize int contextSize int
} }
@ -68,6 +68,7 @@ func (l *LLama) Eval(text string, opts ...PredictOption) error {
if po.Tokens == 0 { if po.Tokens == 0 {
po.Tokens = 99999999 po.Tokens = 99999999
} }
defer C.free(unsafe.Pointer(input))
reverseCount := len(po.StopPrompts) reverseCount := len(po.StopPrompts)
reversePrompt := make([]*C.char, reverseCount) reversePrompt := make([]*C.char, reverseCount)
@ -76,6 +77,7 @@ func (l *LLama) Eval(text string, opts ...PredictOption) error {
cs := C.CString(s) cs := C.CString(s)
reversePrompt[i] = cs reversePrompt[i] = cs
pass = &reversePrompt[0] pass = &reversePrompt[0]
defer C.free(unsafe.Pointer(cs))
} }
params := C.llama_allocate_params(input, C.int(po.Seed), C.int(po.Threads), C.int(po.Tokens), C.int(po.TopK), params := C.llama_allocate_params(input, C.int(po.Seed), C.int(po.Threads), C.int(po.Tokens), C.int(po.TopK),
@ -88,13 +90,13 @@ func (l *LLama) Eval(text string, opts ...PredictOption) error {
C.CString(po.MainGPU), C.CString(po.TensorSplit), C.CString(po.MainGPU), C.CString(po.TensorSplit),
C.bool(po.PromptCacheRO), C.bool(po.PromptCacheRO),
) )
defer C.llama_free_params(params)
ret := C.eval(params, l.ctx, input) ret := C.eval(params, l.ctx, input)
if ret != 0 { if ret != 0 {
return fmt.Errorf("inference failed") return fmt.Errorf("inference failed")
} }
C.llama_free_params(params)
return nil return nil
} }
@ -109,6 +111,8 @@ func (l *LLama) Predict(text string, opts ...PredictOption) (string, error) {
if po.Tokens == 0 { if po.Tokens == 0 {
po.Tokens = 99999999 po.Tokens = 99999999
} }
defer C.free(unsafe.Pointer(input))
out := make([]byte, po.Tokens) out := make([]byte, po.Tokens)
reverseCount := len(po.StopPrompts) reverseCount := len(po.StopPrompts)
@ -118,18 +122,32 @@ func (l *LLama) Predict(text string, opts ...PredictOption) (string, error) {
cs := C.CString(s) cs := C.CString(s)
reversePrompt[i] = cs reversePrompt[i] = cs
pass = &reversePrompt[0] pass = &reversePrompt[0]
defer C.free(unsafe.Pointer(cs))
} }
cLogitBias := C.CString(po.LogitBias)
defer C.free(unsafe.Pointer(cLogitBias))
cPathPromptCache := C.CString(po.PathPromptCache)
defer C.free(unsafe.Pointer(cPathPromptCache))
cMainGPU := C.CString(po.MainGPU)
defer C.free(unsafe.Pointer(cMainGPU))
cTensorSplit := C.CString(po.TensorSplit)
defer C.free(unsafe.Pointer(cTensorSplit))
params := C.llama_allocate_params(input, C.int(po.Seed), C.int(po.Threads), C.int(po.Tokens), C.int(po.TopK), params := C.llama_allocate_params(input, C.int(po.Seed), C.int(po.Threads), C.int(po.Tokens), C.int(po.TopK),
C.float(po.TopP), C.float(po.Temperature), C.float(po.Penalty), C.int(po.Repeat), C.float(po.TopP), C.float(po.Temperature), C.float(po.Penalty), C.int(po.Repeat),
C.bool(po.IgnoreEOS), C.bool(po.F16KV), C.bool(po.IgnoreEOS), C.bool(po.F16KV),
C.int(po.Batch), C.int(po.NKeep), pass, C.int(reverseCount), C.int(po.Batch), C.int(po.NKeep), pass, C.int(reverseCount),
C.float(po.TailFreeSamplingZ), C.float(po.TypicalP), C.float(po.FrequencyPenalty), C.float(po.PresencePenalty), C.float(po.TailFreeSamplingZ), C.float(po.TypicalP), C.float(po.FrequencyPenalty), C.float(po.PresencePenalty),
C.int(po.Mirostat), C.float(po.MirostatETA), C.float(po.MirostatTAU), C.bool(po.PenalizeNL), C.CString(po.LogitBias), C.int(po.Mirostat), C.float(po.MirostatETA), C.float(po.MirostatTAU), C.bool(po.PenalizeNL), cLogitBias,
C.CString(po.PathPromptCache), C.bool(po.PromptCacheAll), C.bool(po.MLock), C.bool(po.MMap), cPathPromptCache, C.bool(po.PromptCacheAll), C.bool(po.MLock), C.bool(po.MMap),
C.CString(po.MainGPU), C.CString(po.TensorSplit), cMainGPU, cTensorSplit,
C.bool(po.PromptCacheRO), C.bool(po.PromptCacheRO),
) )
defer C.llama_free_params(params)
ret := C.llama_predict(params, l.ctx, (*C.char)(unsafe.Pointer(&out[0])), C.bool(po.DebugMode)) ret := C.llama_predict(params, l.ctx, (*C.char)(unsafe.Pointer(&out[0])), C.bool(po.DebugMode))
if ret != 0 { if ret != 0 {
@ -145,8 +163,6 @@ func (l *LLama) Predict(text string, opts ...PredictOption) (string, error) {
res = strings.TrimRight(res, s) res = strings.TrimRight(res, s)
} }
C.llama_free_params(params)
if po.TokenCallback != nil { if po.TokenCallback != nil {
setCallback(l.ctx, nil) setCallback(l.ctx, nil)
} }