Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
server: fix ollama ps showing configured instead of actual context length
Jesse Grossd11fbd2c603a
diff --git a/llm/server.go b/llm/server.goindex 8fedc8468..c09f52c6d 100644--- a/llm/server.go+++ b/llm/server.go@@ -80,6 +80,7 @@ type LlamaServer interface {GetPort() intGetDeviceInfos(ctx context.Context) []ml.DeviceInfoHasExited() bool+ ContextLength() int}// llmServer is an instance of a runner hosting a single model@@ -1901,6 +1902,10 @@ func (s *llmServer) VRAMByGPU(id ml.DeviceID) uint64 {return 0}+func (s *llmServer) ContextLength() int {+ return s.options.NumCtx+}+func (s *ollamaServer) GetDeviceInfos(ctx context.Context) []ml.DeviceInfo {devices, err := ml.GetDevicesFromRunner(ctx, s)if err != nil {diff --git a/server/routes.go b/server/routes.goindex d6c1cbe16..e28eb7798 100644--- a/server/routes.go+++ b/server/routes.go@@ -1897,8 +1897,8 @@ func (s *Server) PsHandler(c *gin.Context) {Details: modelDetails,ExpiresAt: v.expiresAt,}- if v.Options != nil {- mr.ContextLength = v.Options.NumCtx+ if v.llama != nil {+ mr.ContextLength = v.llama.ContextLength()}// The scheduler waits to set expiresAt, so if a model is loading it's// possible that it will be set to the unix epoch. For those cases, justdiff --git a/server/sched_test.go b/server/sched_test.goindex ebf9d7695..7eaf4a9f9 100644--- a/server/sched_test.go+++ b/server/sched_test.go@@ -804,6 +804,7 @@ func (s *mockLlm) GetPort() int { return -func (s *mockLlm) GetDeviceInfos(ctx context.Context) []ml.DeviceInfo { return nil }func (s *mockLlm) HasExited() bool { return false }func (s *mockLlm) GetActiveDeviceIDs() []ml.DeviceID { return nil }+func (s *mockLlm) ContextLength() int { return 0 }// TestImageGenRunnerCanBeEvicted verifies that an image generation model// loaded in the scheduler can be evicted when idle.diff --git a/x/imagegen/server.go b/x/imagegen/server.goindex ae13f5ad7..ca9367694 100644--- a/x/imagegen/server.go+++ b/x/imagegen/server.go@@ -347,6 +347,11 @@ func (s *Server) VRAMByGPU(id ml.DeviceID) uint64 {return s.vramSize}+// Context length is not applicable for image generation.+func (s *Server) ContextLength() int {+ return 0+}+func (s *Server) Embedding(ctx context.Context, input string) ([]float32, int, error) {return nil, 0, errors.New("not supported")}