configurable rope frequency parameters
This commit is contained in:
parent
e3fb1fd3f1
commit
b9f4d67554
2 changed files with 27 additions and 21 deletions
46
api/types.go
46
api/types.go
|
@ -147,19 +147,21 @@ type Options struct {
|
||||||
UseNUMA bool `json:"numa,omitempty"`
|
UseNUMA bool `json:"numa,omitempty"`
|
||||||
|
|
||||||
// Model options
|
// Model options
|
||||||
NumCtx int `json:"num_ctx,omitempty"`
|
NumCtx int `json:"num_ctx,omitempty"`
|
||||||
NumKeep int `json:"num_keep,omitempty"`
|
NumKeep int `json:"num_keep,omitempty"`
|
||||||
NumBatch int `json:"num_batch,omitempty"`
|
NumBatch int `json:"num_batch,omitempty"`
|
||||||
NumGQA int `json:"num_gqa,omitempty"`
|
NumGQA int `json:"num_gqa,omitempty"`
|
||||||
NumGPU int `json:"num_gpu,omitempty"`
|
NumGPU int `json:"num_gpu,omitempty"`
|
||||||
MainGPU int `json:"main_gpu,omitempty"`
|
MainGPU int `json:"main_gpu,omitempty"`
|
||||||
LowVRAM bool `json:"low_vram,omitempty"`
|
LowVRAM bool `json:"low_vram,omitempty"`
|
||||||
F16KV bool `json:"f16_kv,omitempty"`
|
F16KV bool `json:"f16_kv,omitempty"`
|
||||||
LogitsAll bool `json:"logits_all,omitempty"`
|
LogitsAll bool `json:"logits_all,omitempty"`
|
||||||
VocabOnly bool `json:"vocab_only,omitempty"`
|
VocabOnly bool `json:"vocab_only,omitempty"`
|
||||||
UseMMap bool `json:"use_mmap,omitempty"`
|
UseMMap bool `json:"use_mmap,omitempty"`
|
||||||
UseMLock bool `json:"use_mlock,omitempty"`
|
UseMLock bool `json:"use_mlock,omitempty"`
|
||||||
EmbeddingOnly bool `json:"embedding_only,omitempty"`
|
EmbeddingOnly bool `json:"embedding_only,omitempty"`
|
||||||
|
RopeFrequencyBase float32 `json:"rope_frequency_base,omitempty"`
|
||||||
|
RopeFrequencyScale float32 `json:"rope_frequency_scale,omitempty"`
|
||||||
|
|
||||||
// Predict options
|
// Predict options
|
||||||
RepeatLastN int `json:"repeat_last_n,omitempty"`
|
RepeatLastN int `json:"repeat_last_n,omitempty"`
|
||||||
|
@ -261,14 +263,16 @@ func DefaultOptions() Options {
|
||||||
|
|
||||||
UseNUMA: false,
|
UseNUMA: false,
|
||||||
|
|
||||||
NumCtx: 2048,
|
NumCtx: 2048,
|
||||||
NumBatch: 512,
|
NumBatch: 512,
|
||||||
NumGPU: 1,
|
NumGPU: 1,
|
||||||
NumGQA: 1,
|
NumGQA: 1,
|
||||||
LowVRAM: false,
|
LowVRAM: false,
|
||||||
F16KV: true,
|
F16KV: true,
|
||||||
UseMMap: true,
|
UseMMap: true,
|
||||||
UseMLock: false,
|
UseMLock: false,
|
||||||
|
RopeFrequencyBase: 10000.0,
|
||||||
|
RopeFrequencyScale: 1.0,
|
||||||
|
|
||||||
RepeatLastN: 64,
|
RepeatLastN: 64,
|
||||||
RepeatPenalty: 1.1,
|
RepeatPenalty: 1.1,
|
||||||
|
|
|
@ -142,6 +142,8 @@ func New(model string, opts api.Options) (*LLM, error) {
|
||||||
params.use_mmap = C.bool(llm.UseMMap)
|
params.use_mmap = C.bool(llm.UseMMap)
|
||||||
params.use_mlock = C.bool(llm.UseMLock)
|
params.use_mlock = C.bool(llm.UseMLock)
|
||||||
params.embedding = C.bool(llm.EmbeddingOnly)
|
params.embedding = C.bool(llm.EmbeddingOnly)
|
||||||
|
params.rope_freq_base = C.float(llm.RopeFrequencyBase)
|
||||||
|
params.rope_freq_scale = C.float(llm.RopeFrequencyScale)
|
||||||
llm.params = ¶ms
|
llm.params = ¶ms
|
||||||
|
|
||||||
cModel := C.CString(model)
|
cModel := C.CString(model)
|
||||||
|
|
Loading…
Reference in a new issue