matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 08:17:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

load projectors

Browse Source
This commit is contained in:
Michael Yang
2023-11-30 10:30:23 -08:00
parent 409bb9674e
commit b9495ea162
4 changed files with 23 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
llm/llama.go
View File

@@ -325,7 +325,7 @@ func (w *StatusWriter) Write(b []byte) (int, error) {
return os.Stderr.Write(b)
}
func newLlama(model string, adapters []string, runners []ModelRunner, numLayers int64, opts api.Options) (*llama, error) {
func newLlama(model string, adapters, projectors []string, runners []ModelRunner, numLayers int64, opts api.Options) (*llama, error) {
fileInfo, err := os.Stat(model)
if err != nil {
return nil, err
@@ -365,6 +365,11 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
params = append(params, "--lora", adapters[0])
}
if len(projectors) > 0 {
// TODO: applying multiple projectors is not supported by the llama.cpp server yet
params = append(params, "--mmproj", projectors[0])
}
if opts.NumThread > 0 {
params = append(params, "--threads", fmt.Sprintf("%d", opts.NumThread))
}

6
llm/llm.go
View File

@@ -23,7 +23,7 @@ type LLM interface {
Ping(context.Context) error
}
func New(workDir, model string, adapters []string, opts api.Options) (LLM, error) {
func New(workDir, model string, adapters, projectors []string, opts api.Options) (LLM, error) {
if _, err := os.Stat(model); err != nil {
return nil, err
}
@@ -82,9 +82,9 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
opts.NumGQA = 0
opts.RopeFrequencyBase = 0.0
opts.RopeFrequencyScale = 0.0
return newLlama(model, adapters, chooseRunners(workDir, "gguf"), ggml.NumLayers(), opts)
return newLlama(model, adapters, projectors, chooseRunners(workDir, "gguf"), ggml.NumLayers(), opts)
case "ggml", "ggmf", "ggjt", "ggla":
return newLlama(model, adapters, chooseRunners(workDir, "ggml"), ggml.NumLayers(), opts)
return newLlama(model, adapters, projectors, chooseRunners(workDir, "ggml"), ggml.NumLayers(), opts)
default:
return nil, fmt.Errorf("unknown ggml type: %s", ggml.ModelFamily())
}