server: use tiered VRAM-based default context length

Replace binary low VRAM mode with tiered VRAM thresholds that set
default context lengths for all models:

- < 24 GiB VRAM: 4,096 context
- 24-48 GiB VRAM: 32,768 context
- >= 48 GiB VRAM: 262,144 context
This commit is contained in:
Jesse Gross
2026-01-27 16:12:17 -08:00
parent d11fbd2c60
commit 0334ffa625
7 changed files with 157 additions and 24 deletions

View File

@@ -201,7 +201,7 @@ var (
// Enable the new Ollama engine
NewEngine = Bool("OLLAMA_NEW_ENGINE")
// ContextLength sets the default context length
ContextLength = Uint("OLLAMA_CONTEXT_LENGTH", 4096)
ContextLength = Uint("OLLAMA_CONTEXT_LENGTH", 0)
// Auth enables authentication between the Ollama client and server
UseAuth = Bool("OLLAMA_AUTH")
// Enable Vulkan backend
@@ -290,7 +290,7 @@ func AsMap() map[string]EnvVar {
"OLLAMA_ORIGINS": {"OLLAMA_ORIGINS", AllowedOrigins(), "A comma separated list of allowed origins"},
"OLLAMA_SCHED_SPREAD": {"OLLAMA_SCHED_SPREAD", SchedSpread(), "Always schedule model across all GPUs"},
"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_CONTEXT_LENGTH": {"OLLAMA_CONTEXT_LENGTH", ContextLength(), "Context length to use unless otherwise specified (default: 4k/32k/256k based on VRAM)"},
"OLLAMA_NEW_ENGINE": {"OLLAMA_NEW_ENGINE", NewEngine(), "Enable the new Ollama engine"},
"OLLAMA_REMOTES": {"OLLAMA_REMOTES", Remotes(), "Allowed hosts for remote models (default \"ollama.com\")"},

View File

@@ -282,7 +282,7 @@ func TestVar(t *testing.T) {
func TestContextLength(t *testing.T) {
cases := map[string]uint{
"": 4096,
"": 0,
"2048": 2048,
}