Josh Yan 9 months ago
parent
commit
64405525b4
2 changed files with 2 additions and 5 deletions
  1. 1 2
      llm/gguf.go
  2. 1 3
      server/model.go

+ 1 - 2
llm/gguf.go

@@ -252,8 +252,7 @@ func (llm *gguf) Decode(rs io.ReadSeeker) error {
 		return fmt.Errorf("failed to get current offset: %w", err)
 	}
 
-	// ADD PADDING
-
+	// align to next 32-byte boundary
 	llm.offset = offset + llm.padding(offset, int64(alignment))
 
 	for _, tensor := range llm.tensors {

+ 1 - 3
server/model.go

@@ -260,7 +260,6 @@ func parseFromFile(ctx context.Context, file *os.File, digest string, fn func(ap
 					Shape: shape,
 
 					WriterTo: &llm.TensorWriter{
-						// This needs offset + tensors.Offset int64(tensor.Offset) to be correct
 						Reader: io.NewSectionReader(file, offset+ggmlTensors.Offset+int64(tensor.Offset), int64(tensor.Size())),
 					},
 				})
@@ -268,10 +267,9 @@ func parseFromFile(ctx context.Context, file *os.File, digest string, fn func(ap
 
 			reader = &llm.GGUFWriter{
 				KV: ggml.KV(),
-				// Update .Tensors
 				Tensors: llm.Tensors{
 					Items:  tensors,
-					Offset: ggml.Tensors().Offset,
+					Offset: ggmlTensors.Offset,
 				},
 			}
 		}