OSSForks/ollama
1
0
Fork 0
mirror of https://github.com/zebrajr/ollama.git synced 2025-12-06 00:19:51 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

llm: Enable new memory estimates by default

Browse Source 
New memory estimates (see #11090 for more information) are now
enabled automatically for all models running on the Ollama engine,
improving both stability and performance through more accurate sizing
and allocation. Models running on the llama engine will continue to
use the original style of memory estimation.
This commit is contained in:
Jesse Gross 2025-09-11 10:30:18 -07:00 committed by Jesse Gross
parent feb18cd710
commit eb10390de9
2 changed files with 1 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
envconfig/config.go
View File

@ -185,8 +185,6 @@ var (
ContextLength = Uint("OLLAMA_CONTEXT_LENGTH", 4096)
// Auth enables authentication between the Ollama client and server
UseAuth = Bool("OLLAMA_AUTH")
// Enable the new memory estimation logic
NewMemoryEstimates = Bool("OLLAMA_NEW_ESTIMATES")
)
func String(s string) func() string {
@ -272,7 +270,6 @@ func AsMap() map[string]EnvVar {
"OLLAMA_MULTIUSER_CACHE": {"OLLAMA_MULTIUSER_CACHE", MultiUserCache(), "Optimize prompt caching for multi-user scenarios"},
"OLLAMA_CONTEXT_LENGTH": {"OLLAMA_CONTEXT_LENGTH", ContextLength(), "Context length to use unless otherwise specified (default: 4096)"},
"OLLAMA_NEW_ENGINE": {"OLLAMA_NEW_ENGINE", NewEngine(), "Enable the new Ollama engine"},
"OLLAMA_NEW_ESTIMATES": {"OLLAMA_NEW_ESTIMATES", NewMemoryEstimates(), "Enable the new memory estimation logic"},
// Informational
"HTTP_PROXY": {"HTTP_PROXY", String("HTTP_PROXY")(), "HTTP proxy"},

7
llm/server.go
View File

@ -162,11 +162,6 @@ func NewLlamaServer(gpus discover.GpuInfoList, modelPath string, f *ggml.GGML, a
}
}
newEstimates := textProcessor != nil && envconfig.NewMemoryEstimates()
if newEstimates {
slog.Info("enabling new memory estimates")
}
// Verify the requested context size is <= the model training size
trainCtx := f.KV().ContextLength()
if opts.NumCtx > int(trainCtx) && trainCtx > 0 {
@ -434,7 +429,7 @@ func NewLlamaServer(gpus discover.GpuInfoList, modelPath string, f *ggml.GGML, a
}
}()
if newEstimates {
if textProcessor != nil {
return &ollamaServer{llmServer: s}, nil
} else {
return &llamaServer{llmServer: s, ggml: f}, nil