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.
Refactor the tensor parallelism, quantization, and weight-loading codes.
Summary of the new features enabled by this PR:
- **All models** are able to be quantized with AWQ and SqueezeLLM, and [soon GPTQ](https://github.com/vllm-project/vllm/pull/1580).
- Model loading code became much simpler.
- Support model parallelism for all MQA/GQA models when the number of key/value heads is smaller than the tensor parallel size.
Robert Shaw