🎭 What happens when a 1B model has only 1,000 tokens of memory?

Meet Thousand-Token Theater, an AI improv theater built by developer @jhahimanshu653 with tiny models. MiniCPM5-1B writes every actor's line while VoxCPM2 gives each character a distinct voice, enabling AI actors to perform live with a strict 1,000-token memory limit.

Beyond the models themselves, the project also showcases a complete small-model application pipeline: 🎭 Interactive Gradio-based theater interface 🧠 Real-time token streaming generation 📉 Visual memory meter showing context consumption ⭐ Saveable highlights that survive after memories fade 🚀 Fully powered by open-source models without external cloud APIs

Built for the #BuildSmall Hackathon, this project demonstrates how tiny models can go beyond simple conversations and power creative, autonomous experiences.

🔗 Try the demo: https://huggingface.co/spaces/build-small-hackathon/thousand-token-theater

🤗 Models: MiniCPM5-1B: https://huggingface.co/openbmb/MiniCPM5-1B VoxCPM2: https://huggingface.co/openbmb/VoxCPM2