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

add TODO for cuda overhead

Browse source
This commit is contained in:
Jeffrey Morgan 2024-01-09 00:28:03 -05:00
parent 37708931fb
commit 6566387ae3
1 changed files with 3 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
gpu/gpu.go
View file

@ -131,8 +131,9 @@ func getCPUMem() (memInfo, error) {
func CheckVRAM() (int64, error) {
gpuInfo := GetGPUInfo()
if gpuInfo.FreeMemory > 0 && (gpuInfo.Library == "cuda" || gpuInfo.Library == "rocm") {
// leave 20% of VRAM free for overhead
return int64(gpuInfo.FreeMemory * 4 / 5), nil
// leave 25% of VRAM free for overhead
// TODO: improve cuda memory allocation to avoid needing this large of an amount
return int64(gpuInfo.FreeMemory * 3 / 4), nil
}
return 0, fmt.Errorf("no GPU detected") // TODO - better handling of CPU based memory determiniation