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

scale graph based on gpu count

Browse source
This commit is contained in:
Michael Yang 2024-04-16 14:44:13 -07:00
parent 7c9792a6e0
commit 26df674785
2 changed files with 4 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
gpu/gpu_darwin.go
View file

@ -55,6 +55,6 @@ func getCPUMem() (memInfo, error) {
return memInfo{
TotalMemory: uint64(C.getPhysicalMemory()),
FreeMemory: 0,
DeviceCount: 0,
DeviceCount: 1,
}, nil
}

3
llm/server.go
View file

@ -79,6 +79,9 @@ func NewLlamaServer(model string, adapters, projectors []string, opts api.Option
graphFullOffload = graphPartialOffload
}
graphFullOffload *= uint64(info.DeviceCount)
graphPartialOffload *= uint64(info.DeviceCount)
// memoryRequiredTotal represents the memory required for full GPU offloading (all layers)
memoryRequiredTotal := memoryMinimum + graphFullOffload