vllm - 💡(How to fix) Fix [Feature]: W6A16 Support [1 participants]

# Update data type enumeration
from enum import Enum
class DataType(Enum):
    W4A16 = 1
    W6A16 = 2
    W8A16 = 3

# Modify quantization function
def quantize(data, data_type):
    if data_type == DataType.W6A16:
        # Implement W6A16 quantization logic
        return data >> 2
    elif data_type == DataType.W4A16:
        # Implement W4A16 quantization logic
        return data >> 4
    elif data_type == DataType.W8A16:
        # Implement W8A16 quantization logic
        return data

# Update cache storage
class Cache:
    def __init__(self, data_type):
        self.data_type = data_type
        self.cache = {}

    def store(self, key, value):
        if self.data_type == DataType.W6A16:
            # Store W6A16 data in cache
            self.cache[key] = value
        else:
            # Store other data types in cache
            self.cache[key] = value