소스 검색

ollamarunner: Quiet debug logging and panic on unimplemented features

Debug logging of every token has previously caused test timeouts
on slower machines.
Jesse Gross 1 개월 전
부모
커밋
0daaaef8c9
2개의 변경된 파일8개의 추가작업 그리고 6개의 파일을 삭제
  1. 0 4
      model/process_text.go
  2. 8 2
      runner/ollamarunner/runner.go

+ 0 - 4
model/process_text.go

@@ -177,7 +177,6 @@ func (bpe BytePairEncoding) Encode(s string, addSpecial bool) ([]int32, error) {
 	for _, frag := range fragments {
 	for _, frag := range fragments {
 		if len(frag.ids) > 0 {
 		if len(frag.ids) > 0 {
 			ids = append(ids, frag.ids...)
 			ids = append(ids, frag.ids...)
-			slog.Debug("encoded", "text", frag.value, "ids", frag.ids, "special", true)
 			continue
 			continue
 		}
 		}
 
 
@@ -201,7 +200,6 @@ func (bpe BytePairEncoding) Encode(s string, addSpecial bool) ([]int32, error) {
 			// short circuit if the fragment is in the vocabulary
 			// short circuit if the fragment is in the vocabulary
 			if id := bpe.vocab.Encode(sb.String()); id >= 0 {
 			if id := bpe.vocab.Encode(sb.String()); id >= 0 {
 				ids = append(ids, id)
 				ids = append(ids, id)
-				slog.Debug("encoded", "text", sb.String(), "ids", []int32{id})
 				continue
 				continue
 			}
 			}
 
 
@@ -275,7 +273,6 @@ func (bpe BytePairEncoding) Encode(s string, addSpecial bool) ([]int32, error) {
 					// TODO: handle the edge case where the rune isn't in the vocabulary
 					// TODO: handle the edge case where the rune isn't in the vocabulary
 					if id := bpe.vocab.Encode(string(merge.runes)); id >= 0 {
 					if id := bpe.vocab.Encode(string(merge.runes)); id >= 0 {
 						ids = append(ids, id)
 						ids = append(ids, id)
-						slog.Debug("encoded", "text", string(merge.runes), "ids", []int32{id})
 					}
 					}
 				}
 				}
 			}
 			}
@@ -329,6 +326,5 @@ func (bpe BytePairEncoding) Decode(ids []int32) (string, error) {
 		}
 		}
 	}
 	}
 
 
-	slog.Debug("decoded", "ids", ids, "text", sb.String())
 	return sb.String(), nil
 	return sb.String(), nil
 }
 }

+ 8 - 2
runner/ollamarunner/runner.go

@@ -436,8 +436,10 @@ func (s *Server) processBatch() error {
 		// if done processing the prompt, generate an embedding and return
 		// if done processing the prompt, generate an embedding and return
 		if seq.embeddingOnly {
 		if seq.embeddingOnly {
 			// TODO(jessegross): Embedding support
 			// TODO(jessegross): Embedding support
-			s.removeSequence(i, "")
-			continue
+			// s.removeSequence(i, "")
+			// continue
+
+			panic("generation of embedding outputs not yet supported")
 		}
 		}
 
 
 		// sample a token
 		// sample a token
@@ -597,6 +599,10 @@ func (s *Server) completion(w http.ResponseWriter, r *http.Request) {
 		req.Seed,
 		req.Seed,
 	)
 	)
 
 
+	if req.Grammar != "" {
+		panic("grammars are not yet supported")
+	}
+
 	seq, err := s.NewSequence(req.Prompt, req.Images, NewSequenceParams{
 	seq, err := s.NewSequence(req.Prompt, req.Images, NewSequenceParams{
 		numPredict: req.NumPredict,
 		numPredict: req.NumPredict,
 		stop:       req.Stop,
 		stop:       req.Stop,