fixed refusing to quantize some models

YellowRoseCx · Jul 5, 2023 · ea79e54 · ea79e54
1 parent 69add28
commit ea79e54
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/llama.cpp b/llama.cpp
@@ -2404,9 +2404,9 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
  int ny = tensor.ne.at(1);
  if (nx % QK_K != 0 || ny % QK_K != 0) {
  fprintf(stderr, "\n\n========================= Tensor sizes %d x %d are not divisible by %d\n",nx,ny,QK_K);
- fprintf(stderr, "This is required to be able to use k-quants for now!\n");
+ fprintf(stderr, "Verify before using\n");
  fprintf(stderr, "========================================================================================\n\n");
- throw std::runtime_error("Unsupported tensor size encountered\n");
+ // throw std::runtime_error("Unsupported tensor size encountered\n");
  }
  }
  if (tensor.name == "output.weight") {