Function gptq_loop

Quantize weight tensor with GPTQ algorithm weight: (C, R), transposed weight tensor to quantize, modified in-place and returned hessian_inv: (C, C), inverse of Hessian matrix scale: (C, R), transposed scale tensor for quantization qzero: (C, R), transposed zero-point tensor for

(
    weight: torch.Tensor,
    hessian_inv: torch.Tensor,
    scale: torch.Tensor,
    qzero: torch.Tensor,
    maxq: torch.Tensor,
    dtype: torch.dtype,
    gptq_block_size: int = 128,
)