Function gptq_loop_graph

CUDA Graph wrapper for GPTQ loops

(
    weight: torch.Tensor,
    hessian_inv: torch.Tensor,
    scale: torch.Tensor,
    qzero: torch.Tensor,
    maxq: torch.Tensor,
    qweight: torch.Tensor = None,
    error_block: torch.Tensor = None,
    dtype: torch.dtype = None,
    gptq_block_size: int = 128,
    direct: bool = True,
)