Build Small Hackathon

2026-06-05 · Source: HuggingFace · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Intermediate, extended

Summary

The "Build Small Hackathon" kickoff event introduced a competition focused on developing applications with models under 32 billion parameters, aiming to revive an era of tinkerable, fine-tunable AI. Participants can choose between two tracks: "Backyard AI" for practical, useful applications and "Thousand Token Wood" for creative, whimsical projects. Submissions must be Gradio apps hosted on the Hugging Face hackathon organization, include a demo video, and proof of a social media post. The hackathon offers approximately \$48,000 in cash prizes, \$20,000 in model credits, two Nvidia RTX 580s, and a year of J GBD Pro, distributed across 29 categories. Sponsors like Black Forest Labs, OpenBMB, OpenAI, Nvidia, Modal, JetBrains, and Cohere Labs are providing models, credits, and specific prize categories, encouraging innovation with smaller, efficient AI.

Key takeaway

For AI Engineers and Machine Learning Engineers seeking to innovate with efficient, smaller models, this hackathon provides a structured environment and significant resources. You should explore the diverse sponsor models like Nvidia NeMo Guardrails or JetBrains Milum 2, focusing on their specific strengths for your chosen track. Consider using Gradio Workflow for rapid prototyping and ensure your project adheres to the 32 billion parameter limit and submission requirements to maximize your chances for prizes and credits.

Key insights

The hackathon promotes building practical or creative AI applications using models under 32 billion parameters.

Principles

• Small models enable broad accessibility and local deployment.
• Fine-tuning and customization enhance model utility.
• Community engagement drives innovation in AI development.

Method

Develop AI pipelines without code using Gradio Workflow's drag-and-drop interface, integrating Hugging Face models, datasets, and custom Python functions.

In practice

• Utilize Flux 2 Client (4B/9B) for text-to-image and editing.
• Deploy Mini CPM models (1B-4.18B) for local-first text or multimodal apps.
• Leverage OpenAI Codex for code orchestration and fine-tuning models.

Topics

• AI Hackathon
• Small Language Models
• Gradio
• Hugging Face
• Model Fine-tuning
• On-device AI
• Multimodal AI

Best for: NLP Engineer, AI Student, Machine Learning Engineer, AI Engineer

Related on AIssential

• HackerNoon Projects of the Week: Pantry Pilot, RecomendeMe & Charmpay — Real-world utility is the primary metric for evaluating projects in HackerNoon's "Proof of Usefulness" hackathon.
• Nanochat Can Now Train a GPT-2 Level Model in Just 2 Hours — Autonomous AI agents and optimized pipelines are rapidly accelerating model training and self-improvement in open-source projects.
• How To Win Humanity's Last Hackathon - The hardest agent contest in AI. — Optimizing AI kernels for specific hardware, especially Mac Metal, requires advanced agent context engineering.
• One-Shot Any Web App with Gradio's gr.HTML — Gradio's `gr.HTML` enables single-file, interactive web components with custom HTML, CSS, and JavaScript, ideal for LLM-driven development.
• Build Small with Modal — Modal offers purpose-built, developer-friendly infrastructure for scaling data and compute-intensive AI workloads with fast GPU orchestration and sandboxing.
• Engaging the AI community through building, research, and shared learning — Community engagement through hackathons and academic challenges accelerates AI innovation and addresses real-world problems.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by HuggingFace.