[Kernel][FP8] Initial support with dynamic per-tensor scaling (#4118)

Provide an initial support to FP8 computation. This PR is inspired by HuggingFace TGI: huggingface/text-generation-inference#1726

This feature can be enabled with --quantization fp8 or -q fp8 when launching an engine.

Algorithm:
We still load a model checkpoint in FP16/BF16. After the weights are loaded, Fp8LinearMethod calculates the per-tensor scaling factor of weights and quantizes the weights accordingly. The scaling factor will then be stored for future use. Meanwhile, the per-tensor scaling factor for activations is calculated in every forward pass.

Initial Results:
Currently tested Mistral-7B on 1xH100. With prompt length ~5 and decoding length 128:

BF16: 1.47s
FP8: 1.66s
I'll try to use larger models and try to find more performance bottleneck. Meanwhile, you're welcome to try this code.

vllm/model_executor/layers/quantization/__init__.py

@@ -3,12 +3,14 @@ from typing import Type
 from vllm.model_executor.layers.quantization.awq import AWQConfig
 from vllm.model_executor.layers.quantization.base_config import (
     QuantizationConfig)
+from vllm.model_executor.layers.quantization.fp8 import FP8Config
 from vllm.model_executor.layers.quantization.gptq import GPTQConfig
 from vllm.model_executor.layers.quantization.marlin import MarlinConfig
 from vllm.model_executor.layers.quantization.squeezellm import SqueezeLLMConfig
 
 QUANTIZATION_METHODS = {
     "awq": AWQConfig,
+    "fp8": FP8Config,
     "gptq": GPTQConfig,
     "squeezellm": SqueezeLLMConfig,
     "marlin": MarlinConfig,