default rope params to 0 for new models (#968)

This commit is contained in:
Jeffrey Morgan
2023-11-02 08:41:30 -07:00
committed by GitHub
parent 527f9a7975
commit 2e53704685
2 changed files with 12 additions and 3 deletions

View File

@@ -85,7 +85,10 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
switch ggml.Name() {
case "gguf":
opts.NumGQA = 0 // TODO: remove this when llama.cpp runners differ enough to need separate newLlama functions
// TODO: gguf will load these options automatically from the model binary
opts.NumGQA = 0
opts.RopeFrequencyBase = 0.0
opts.RopeFrequencyScale = 0.0
return newLlama(model, adapters, chooseRunners(workDir, "gguf"), ggml.NumLayers(), opts)
case "ggml", "ggmf", "ggjt", "ggla":
return newLlama(model, adapters, chooseRunners(workDir, "ggml"), ggml.NumLayers(), opts)