Method forward

inference/modules/fused_mlp.py:41–42

(self, x)

        self.offset = 0x0F if self.w_bit == 4 else 0x03

    def forward(self, x):
        return self.down_proj(self.our_llama_mlp(x))

    def our_llama_mlp(self, x):
        out_shape = x.shape[:-1] + (self.intermediate_size,)