Bladeren bron

runner: default to greedy sampler for performance (#9407)

As are adding support for weighted sampling we have seen some performance
regressions, bypassing the sampler logic for now and defaulting to greedy
until we can benchmark the new sampler logic.
Bruce MacDonald 2 maanden geleden
bovenliggende
commit
0c1041ad85
1 gewijzigde bestanden met toevoegingen van 1 en 13 verwijderingen
  1. 1 13
      runner/ollamarunner/runner.go

+ 1 - 13
runner/ollamarunner/runner.go

@@ -575,23 +575,11 @@ func (s *Server) completion(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
-	sampler, err := sample.NewSampler(
-		req.Temperature,
-		req.TopK,
-		req.TopP,
-		req.MinP,
-		req.Seed,
-	)
-	if err != nil {
-		http.Error(w, fmt.Sprintf("Failed to create sampler: %v", err), http.StatusInternalServerError)
-		return
-	}
-
 	seq, err := s.NewSequence(req.Prompt, req.Images, NewSequenceParams{
 		numPredict: req.NumPredict,
 		stop:       req.Stop,
 		numKeep:    int32(req.NumKeep),
-		sampler:    sampler,
+		sampler:    sample.Greedy(), // TODO: add support for different samplers when performance is optimized
 		embedding:  false,
 	})
 	if err != nil {