matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-12 00:37:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

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

Browse Source
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
Download Patch File Download Diff File Expand all files Collapse all files

10
llm/llama.go
View File

@@ -306,13 +306,19 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
params := []string{
"--model", model,
"--ctx-size", fmt.Sprintf("%d", opts.NumCtx),
"--rope-freq-base", fmt.Sprintf("%f", opts.RopeFrequencyBase),
"--rope-freq-scale", fmt.Sprintf("%f", opts.RopeFrequencyScale),
"--batch-size", fmt.Sprintf("%d", opts.NumBatch),
"--n-gpu-layers", fmt.Sprintf("%d", numGPU),
"--embedding",
}
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 opts.NumGQA > 0 {
params = append(params, "--gqa", fmt.Sprintf("%d", opts.NumGQA))
}