benja263 / rl-cc

RL-CC model artifact

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Implementing Reinforcement Learning Datacenter Congestion Control in NVIDIA NICs

RL-CC is an RDMA congestion control algorithm based on reinforcement learning. This repository contains a trained RL-CC neural network model and a script to distill it into an ensemble of decision trees using synthetic data. The RL-CC model was trained using a proprietary NVIDIA congestion control simulator and the synthetic data is designed to approximate the data distribution that is created when running the model within the simulator.

To distill the model:

1. pip install -r requirements.txt
2. python3 distill_network.py

About

RL-CC model artifact


Languages

Language:Python 100.0%