[Feature]: Vectorize `scaled_int8_quant`

[mgoin]

🚀 The feature, motivation and pitch

Similar to the recent discoveries in #18844, vectorizing our quantization methods can have a huge impact on e2e performance.

Currently we only use vectorization.h in csrc/quantization/fp8/common.cuh and csrc/quantization/fused_kernels/layernorm_utils.cuh, so we should expand this to more implementations like csrc/quantization/compressed_tensors/int8_quant_kernels.cu for faster INT8 activation quantization.

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[Drice1999]

I’d like to work on this issue

[jeejeelee]

@Drice1999 Thank you , I have assigned this issue to you

[yewentao256]

Try to fix this issue throughly, will have a pr soon.