Record 089 · Models · 2023
67%23.89.L
TinyLlama
Zhang et al.
“A 1.1B Llama-architecture model that made 'always-on tiny local' real.”
TinyLlama showed a 1B-class model trained on 3T tokens could be useful on a Pi or a phone. Edge SLM before Phi was everywhere.
The cute-but-serious tiny local LLM.
- 1.1B Llama-arch
- High-token training
- Edge-friendly
1.1B, trained on a pile, fits anywhere.
