model.LightExpert Class Reference

Inheritance diagram for model.LightExpert:

Collaboration diagram for model.LightExpert:

Public Member Functions
	__init__ (self, mixer_channels=64, bottleneck_channels=32, hidden_dim=128, expert_pool="flat")
	forward (self, x)

Public Attributes
str	expert_pool = expert_pool
	head_conv = nn.Conv2d(mixer_channels, bottleneck_channels, kernel_size=1)
	head_act = nn.ReLU(inplace=True)
	head_hidden = nn.Linear(hidden_in, hidden_dim)
	head_act2 = nn.ReLU(inplace=True)
	head_wdl = nn.Linear(hidden_dim, 3)

Detailed Description

The ultra-fast Dense Expert. 
It compresses the massive Mixer output down to a small bottleneck before flattening.

Constructor & Destructor Documentation

__init__()

model.LightExpert.__init__	(		self,
			mixer_channels = 64,
			bottleneck_channels = 32,
			hidden_dim = 128,
			expert_pool = "flat" )

Here is the call graph for this function:

Here is the caller graph for this function:

Member Function Documentation

forward()

model.LightExpert.forward	(		self,
			x )

Member Data Documentation

expert_pool

str model.LightExpert.expert_pool = expert_pool

head_act

model.LightExpert.head_act = nn.ReLU(inplace=True)

head_act2

model.LightExpert.head_act2 = nn.ReLU(inplace=True)

head_conv

model.LightExpert.head_conv = nn.Conv2d(mixer_channels, bottleneck_channels, kernel_size=1)

head_hidden

model.LightExpert.head_hidden = nn.Linear(hidden_in, hidden_dim)

head_wdl

model.LightExpert.head_wdl = nn.Linear(hidden_dim, 3)

---

The documentation for this class was generated from the following file:

• training/model.py