highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-11-10 20:37:52 +01:00
Code Issues Packages Projects Releases Wiki Activity

llm: New memory management

Browse Source 
This changes the memory allocation strategy from upfront estimation to
tracking actual allocations done by the engine and reacting to that. The
goal is avoid issues caused by both under-estimation (crashing) and
over-estimation (low performance due to under-utilized GPUs).

It is currently opt-in and can be enabled for models running on the
Ollama engine by setting OLLAMA_NEW_ESTIMATES=1. Behavior in other
cases is unchanged and will continue to use the existing estimates.
This commit is contained in:
Jesse Gross
2025-05-29 12:21:48 -07:00
committed by Jesse Gross
parent ef7d26ba2c
commit d5a0d8d904
26 changed files with 1860 additions and 900 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
server/routes_generate_test.go
View File

@@ -77,12 +77,13 @@ func TestGenerateChat(t *testing.T) {
getGpuFn: discover.GetGPUInfo,
getCpuFn: discover.GetCPUInfo,
reschedDelay: 250 * time.Millisecond,
loadFn: func(req *LlmRequest, _ *ggml.GGML, _ discover.GpuInfoList, _ int) {
loadFn: func(req *LlmRequest, _ *ggml.GGML, _ discover.GpuInfoList, _ bool) bool {
// add small delay to simulate loading
time.Sleep(time.Millisecond)
req.successCh <- &runnerRef{
llama: &mock,
}
return false
},
},
}
@@ -620,12 +621,13 @@ func TestGenerate(t *testing.T) {
getGpuFn: discover.GetGPUInfo,
getCpuFn: discover.GetCPUInfo,
reschedDelay: 250 * time.Millisecond,
loadFn: func(req *LlmRequest, _ *ggml.GGML, _ discover.GpuInfoList, _ int) {
loadFn: func(req *LlmRequest, _ *ggml.GGML, _ discover.GpuInfoList, _ bool) bool {
// add small delay to simulate loading
time.Sleep(time.Millisecond)
req.successCh <- &runnerRef{
llama: &mock,
}
return false
},
},
}