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

revert cuda overhead to 20%

Browse source
This commit is contained in:
Jeffrey Morgan 2024-01-09 00:54:25 -05:00
parent f387e9631b
commit 6164f378f2
1 changed files with 2 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
gpu/gpu.go
View file

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