fuse_fx

class torch.ao.quantization.quantize_fx.fuse_fx(model, fuse_custom_config=None, backend_config=None)

Fuse modules like conv+bn, conv+bn+relu etc, model must be in eval mode. Fusion rules are defined in torch.ao.quantization.fx.fusion_pattern.py

Parameters

• model (*) – a torch.nn.Module model

• fuse_custom_config (*) – custom configurations for fuse_fx. See FuseCustomConfig for more details

Return type

GraphModule

Example:

from torch.ao.quantization import fuse_fx
m = Model().eval()
m = fuse_fx(m)