Experiment with and compare different tokenizers
Read research updates on model compression and AI tools
Classify text prompts as safe or unsafe
How well can you prompt a small AI?
Every tiny LM, same eval harness, transparent benchmarks