Skip to content

vllm.model_executor.layers.quantization.utils.mxfp8_utils

logger module-attribute 

logger = init_logger(__name__)

mxfp8_e4m3_quantize 

mxfp8_e4m3_quantize(x: Tensor) -> tuple[Tensor, Tensor]
Source code in vllm/model_executor/layers/quantization/utils/mxfp8_utils.py 
11
12
13
14
15
16
17
18
19
20
21
22
23
24
def mxfp8_e4m3_quantize(x: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
    try:
        from flashinfer import mxfp8_quantize as mxfp8_e4m3_quantize
    except ImportError as err:
        raise ImportError(
            "The package `flashinfer` is required to do "
            "MX-FP8 quantization. Please install it with"
            "`pip install flashinfer`"
        ) from err

    x_q, x_scales = mxfp8_e4m3_quantize(x, is_sf_swizzled_layout=False)
    if x_scales.ndim == 1:
        x_scales = x_scales.view(x.size(0), -1)
    return x_q, x_scales