mirrors/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-05-11 02:16:36 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

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
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 (
cudaMinimumMemory = 256 * format.MebiByte
rocmMinimumMemory = 256 * format.MebiByte
cudaMinimumMemory = 457 * format.MebiByte
rocmMinimumMemory = 457 * format.MebiByte
)
var gpuMutex sync.Mutex