RichardWooSJTU

Results 4 issues of RichardWooSJTU

### PR types Bug fixes ### PR changes Others ### Description Support csrc building on sm70

### PR types New features ### PR changes Others ### Description 5.2

1. Add auto export shell script and config file for GPT quant model 2. Fix GPT fp16 model exporting error

### PR types Bug fixes ### PR changes Others ### Description csrc building enable bfloat16 default, which is not supported when cuda arch < sm80. This PR add `CUDA_ENBALE_BF16` macro...

inference