Low bit quantization type

[oovm]

As LLM becomes larger and larger, some highly quantized types have emerged in recent years, such as FP8, Q4.

Should tensor-type add these definitions, or provide custom(string) enumeration