mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

Use runners for GPU discovery (#12090)

Browse Source 
This revamps how we discover GPUs in the system by leveraging the Ollama
runner.  This should eliminate inconsistency between our GPU discovery and the
runners capabilities at runtime, particularly for cases where we try to filter
out unsupported GPUs.  Now the runner does that implicitly based on the actual
device list.  In some cases free VRAM reporting can be unreliable which can
leaad to scheduling mistakes, so this also includes a patch to leverage more
reliable VRAM reporting libraries if available.

Automatic workarounds have been removed as only one GPU leveraged this, which
is now documented. This GPU will soon fall off the support matrix with the next
ROCm bump.

Additional cleanup of the scheduler and discovery packages can be done in the
future once we have switched on the new memory management code, and removed
support for the llama runner.
This commit is contained in:
Daniel Hiltgen
2025-10-01 15:12:32 -07:00
committed by GitHub
parent 6b50f2b9cd
commit bc8909fb38
57 changed files with 3288 additions and 3819 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
runner/ollamarunner/runner.go
View File

@@ -28,6 +28,7 @@ import (
"github.com/ollama/ollama/api"
"github.com/ollama/ollama/envconfig"
"github.com/ollama/ollama/fs/ggml"
"github.com/ollama/ollama/llm"
"github.com/ollama/ollama/logutil"
"github.com/ollama/ollama/ml"
@@ -1235,6 +1236,52 @@ func (s *Server) load(w http.ResponseWriter, r *http.Request) {
}
}
// info is the handler called by the Ollama server to report information
// about the GPU devices in use by this runner
func (s *Server) info(w http.ResponseWriter, r *http.Request) {
s.loadMu.Lock()
defer s.loadMu.Unlock()
w.Header().Set("Content-Type", "application/json")
m := s.model
if m == nil {
startLoad := time.Now()
// Dummy load to get the backend wired up
f, err := os.CreateTemp("", "*.bin")
if err != nil {
http.Error(w, fmt.Sprintf("failed to initialize baackend: %v", err), http.StatusInternalServerError)
return
}
defer f.Close()
defer os.Remove(f.Name())
if err := ggml.WriteGGUF(f, ggml.KV{
"general.architecture": "llama",
"tokenizer.ggml.model": "gpt2",
}, nil); err != nil {
http.Error(w, fmt.Sprintf("failed to initialize baackend: %v", err), http.StatusInternalServerError)
return
}
m, err = model.New(f.Name(), ml.BackendParams{NumThreads: runtime.NumCPU(), AllocMemory: false, GPULayers: ml.GPULayersList{{}}})
if err != nil {
http.Error(w, fmt.Sprintf("failed to initialize baackend: %v", err), http.StatusInternalServerError)
return
}
slog.Debug("dummy model load took", "duration", time.Since(startLoad))
}
startDevices := time.Now()
infos := m.Backend().BackendDevices()
slog.Debug("gathering device infos took", "duration", time.Since(startDevices))
if err := json.NewEncoder(w).Encode(&infos); err != nil {
http.Error(w, fmt.Sprintf("failed to encode response: %v", err), http.StatusInternalServerError)
}
}
func Execute(args []string) error {
fs := flag.NewFlagSet("runner", flag.ExitOnError)
mpath := fs.String("model", "", "Path to model binary file")
@@ -1275,6 +1322,7 @@ func Execute(args []string) error {
mux := http.NewServeMux()
// TODO: support embeddings
mux.HandleFunc("GET /info", server.info)
mux.HandleFunc("POST /load", server.load)
mux.HandleFunc("POST /embedding", server.embeddings)
mux.HandleFunc("POST /completion", server.completion)