The mixer may benefit from seeing a good match on its port, so I'd be inclined to put at least some attenuation between the filter and the mixer.
The Si5351 claims to be able to drive a 50 ohm load directly from its outputs, however, be aware that an LC filter, even if designed to be 50 ohms in the passband, is not 50 ohms at all frequencies. Typically, most people implement a lowpass filter with shunt capacitor at each port to ground, as capacitors are cheaper than inductors. This will draw excessive current from a squarewave output.
You could put some of the attenuation before the filter to help with this, or put a diplex filter in front of the LC filter to improve its broadband match, or you could design the LC filter with a series L instead of a shunt C input so it goes high impedance instead of short at high frequency.
a few of the options (there are others)
- Si5351, 5dB, LC filter, 10dB, mixer
- Si5351, diplex filter, LC filter, 15dB, mixer
- Si5351, L-input LC filter, 15dB, mixer
In the first option, there's no need to make the 5dB attenuator a full '50 ohm on each port' attenuator. It would be better, and kinder to your 5351, to make it an L-pad, driving the filter with 50 ohms, but presenting a higher resistance to your driver.