Why not set stream in `gemmv2_forward_cuda` and `dequantize_weights_cuda`

[grimoire]

Hi, thanks for the great work.

I am working on deploy model on distributed environment. Since dequantize_weights uses the default stream, Kernels would be overlap with NCCL all_reduce kernel, which leads to unexpected result.

AutoAWQ_kernels/awq_ext/quantization/gemm_cuda_gen.cu

Line 1162 in 83d1f4b

dequantize_weights<<<num_blocks, threads_per_block>>>(kernel, scaling_factors, zeros, de_kernel, G, in_c, out_c);

Is there any reason to use the default stream here?