mirror of
https://github.com/ollama/ollama.git
synced 2025-07-28 16:13:35 +02:00
Bump VRAM buffer back up
Under stress scenarios we're seeing OOMs so this should help stabilize the allocations under heavy concurrency stress.
This commit is contained in:
@@ -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
|
||||
|
@@ -15,7 +15,7 @@ import (
|
||||
)
|
||||
|
||||
const (
|
||||
metalMinimumMemory = 384 * format.MebiByte
|
||||
metalMinimumMemory = 512 * format.MebiByte
|
||||
)
|
||||
|
||||
func GetGPUInfo() GpuInfoList {
|
||||
|
Reference in New Issue
Block a user