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

no rope parameters

Browse source
This commit is contained in:
Michael Yang 2024-04-05 17:59:58 -07:00
parent fc8e108642
commit be517e491c
4 changed files with 0 additions and 14 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
api/types.go
View file

@ -121,8 +121,6 @@ type Runner struct {
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"`
RopeFrequencyBase float32 `json:"rope_frequency_base,omitempty"`
RopeFrequencyScale float32 `json:"rope_frequency_scale,omitempty"`
NumThread int `json:"num_thread,omitempty"` NumThread int `json:"num_thread,omitempty"`
} }
@ -383,8 +381,6 @@ func DefaultOptions() Options {
Runner: Runner{ Runner: Runner{
// options set when the model is loaded // options set when the model is loaded
NumCtx: 2048, NumCtx: 2048,
RopeFrequencyBase: 10000.0,
RopeFrequencyScale: 1.0,
NumBatch: 512, NumBatch: 512,
NumGPU: -1, // -1 here indicates that NumGPU should be set dynamically NumGPU: -1, // -1 here indicates that NumGPU should be set dynamically
NumGQA: 1, NumGQA: 1,

1
convert/convert.go
View file

@ -32,7 +32,6 @@ type Params struct {
AttentionHeads int `json:"num_attention_heads"` // n_head AttentionHeads int `json:"num_attention_heads"` // n_head
KeyValHeads int `json:"num_key_value_heads"` KeyValHeads int `json:"num_key_value_heads"`
NormEPS float64 `json:"rms_norm_eps"` NormEPS float64 `json:"rms_norm_eps"`
RopeFreqBase float64 `json:"rope_theta"`
BoSTokenID int `json:"bos_token_id"` BoSTokenID int `json:"bos_token_id"`
EoSTokenID int `json:"eos_token_id"` EoSTokenID int `json:"eos_token_id"`
HeadDimension int `json:"head_dim"` HeadDimension int `json:"head_dim"`

1
convert/mistral.go
View file

@ -144,7 +144,6 @@ func (m *MistralModel) WriteGGUF() (string, error) {
"llama.attention.head_count": uint32(m.Params.AttentionHeads), "llama.attention.head_count": uint32(m.Params.AttentionHeads),
"llama.attention.head_count_kv": uint32(m.Params.KeyValHeads), "llama.attention.head_count_kv": uint32(m.Params.KeyValHeads),
"llama.attention.layer_norm_rms_epsilon": float32(m.Params.NormEPS), "llama.attention.layer_norm_rms_epsilon": float32(m.Params.NormEPS),
"llama.rope.freq_base": float32(m.Params.RopeFreqBase),
"general.file_type": uint32(1), "general.file_type": uint32(1),
"tokenizer.ggml.model": "llama", "tokenizer.ggml.model": "llama",

8
llm/server.go
View file

@ -172,14 +172,6 @@ func NewLlamaServer(model string, adapters, projectors []string, opts api.Option
params = append(params, "--main-gpu", fmt.Sprintf("%d", opts.MainGPU)) params = append(params, "--main-gpu", fmt.Sprintf("%d", opts.MainGPU))
} }
if opts.RopeFrequencyBase > 0 {
params = append(params, "--rope-freq-base", fmt.Sprintf("%f", opts.RopeFrequencyBase))
}
if opts.RopeFrequencyScale > 0 {
params = append(params, "--rope-freq-scale", fmt.Sprintf("%f", opts.RopeFrequencyScale))
}
if len(adapters) > 0 { if len(adapters) > 0 {
// TODO: applying multiple adapters is not supported by the llama.cpp server yet // TODO: applying multiple adapters is not supported by the llama.cpp server yet
params = append(params, "--lora", adapters[0]) params = append(params, "--lora", adapters[0])