[Kernel] Add GPTQv2 format support for low-bit or asymmetric quantization, by adapting gptq_gemm (#26092)

csrc/torch_bindings.cpp

@@ -557,7 +557,8 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
   // to prevent the meta function registry.
   ops.def(
       "gptq_gemm(Tensor a, Tensor b_q_weight, Tensor b_gptq_qzeros, "
-      "Tensor b_gptq_scales, Tensor b_g_idx, bool use_exllama, int bit) "
+      "Tensor b_gptq_scales, Tensor b_g_idx, bool use_exllama, bool "
+      "use_v2_format, int bit) "
       "-> Tensor",
       {stride_tag});
   ops.impl("gptq_gemm", torch::kCUDA, &gptq_gemm);