Function load_and_dequantize

convert.py:319–324

(weight_name: str, scale_name: str)

  tensors = {}

  def load_and_dequantize(weight_name: str, scale_name: str) -> torch.Tensor:
    t = weights[weight_name]
    if scale_name in weights:
      scale = weights[scale_name]
      t = blockwise_dequantize(t, scale, dequant_block_size)
    return t
  
  def quantize(t: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
    if dtype not in [torch.float32, torch.float16]: