You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I’m currently working on a project using FINN to deploy a quantized neural network on an FPGA. My goal is to modify the hardware to introduce a random switch in the weight block, which randomly selects between the original weight and zero.
I’ve explored some of the hardware modules in the FINN-rtllib, specifically mvu_4sx4u.sv and memstream.sv, where I’ve seen weight-related signals. However, I’m uncertain whether these are the best places to introduce this modification, and I would appreciate some guidance. Or should I focus on transformation or customOPs parts? Or should I use the file under the external HLS library?
Could you please point me to the most appropriate location within the FINN hardware flow to:
Modify the weight block during matrix-vector multiplication (or similar operation).
Introduce custom logic for switching between weights and zero at runtime.
Additionally, if there are any examples or documentation related to modifying hardware modules like weights or activations, that would be extremely helpful.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hi FINN Team,
I’m currently working on a project using FINN to deploy a quantized neural network on an FPGA. My goal is to modify the hardware to introduce a random switch in the weight block, which randomly selects between the original weight and zero.
I’ve explored some of the hardware modules in the FINN-rtllib, specifically mvu_4sx4u.sv and memstream.sv, where I’ve seen weight-related signals. However, I’m uncertain whether these are the best places to introduce this modification, and I would appreciate some guidance. Or should I focus on transformation or customOPs parts? Or should I use the file under the external HLS library?
Could you please point me to the most appropriate location within the FINN hardware flow to:
Modify the weight block during matrix-vector multiplication (or similar operation).
Introduce custom logic for switching between weights and zero at runtime.
Additionally, if there are any examples or documentation related to modifying hardware modules like weights or activations, that would be extremely helpful.
Thanks in advance for your assistance!
Regards,
prospersu
Beta Was this translation helpful? Give feedback.
All reactions