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

Bump VRAM buffer back up

Browse source 
Under stress scenarios we're seeing OOMs so this should help stabilize
the allocations under heavy concurrency stress.
This commit is contained in:
Daniel Hiltgen 2024-05-10 09:15:28 -07:00
parent 200a18820e
commit 30a7d7096c
2 changed files with 3 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
gpu/gpu.go
View file

@ -31,8 +31,8 @@ type handles struct {
} }
const ( const (
cudaMinimumMemory = 256 * format.MebiByte cudaMinimumMemory = 457 * format.MebiByte
rocmMinimumMemory = 256 * format.MebiByte rocmMinimumMemory = 457 * format.MebiByte
) )
var gpuMutex sync.Mutex var gpuMutex sync.Mutex

2
gpu/gpu_darwin.go
View file

@ -15,7 +15,7 @@ import (
) )
const ( const (
metalMinimumMemory = 384 * format.MebiByte metalMinimumMemory = 512 * format.MebiByte
) )
func GetGPUInfo() GpuInfoList { func GetGPUInfo() GpuInfoList {