Cody Rushing's repository to hold exercises and open ended projects from the 2022 Virtual ARENA program
This is my collection of activities I did from October to December 2022 to strengthen AI engineering skills.
Some highlights of my work include:
- Implemented SARSA and Q-Learning to solve CliffWalking and custom environments
- Used Deep Q-Netowrks to solve the CartPole Gym environment
cartpole.mp4
Week six:
- Learned the fundamentals of Reinforcement Learning
- Trained agents through iterative Policy Improvement and Policy Extraction
[Small haitus]
Week three:
- Learned about, and built, basic optimizers
- Examined scaling laws, and replicated them - report here!
Week two:
- Learned how to use Wandb for hyperparameter sweeps and optimization. Also played around with reports here
- Recreated a... rather broken version of GPT-2. Not working as expected, as of 11/8
Week one:
- Implementing a decoder-only transformer from scratch (so far it can reverse numbers!)
- Used the baseline transformer, coupled with decoder methods, to produce my own Shakespeare (~25 min of training with free Colab GPUs!); play with the transformere here!
Week zero:
- Implementing CNNs from scratch
- Building resnet34 from scratch
- Learning some fundamental libraries: einops, ensum