| # Build Small Hackathon: Complete Resource & Rule Guide |
|
|
| This document serves as a comprehensive summary and structured guide for the "Build Small" hackathon hosted by Hugging Face. |
|
|
| The event aims to redirect focus from large API-based models back to the era of tinkerable, locally tunable open-weight models. |
|
|
| ## 1. Hackathon General Rules |
|
|
| - Model Size Limit: All models used in the hackathon must be strictly under 32 billion parameters. You may use a combination of multiple models, but each individual model must have a total parameter count (not just active parameters) below 32 billion. |
| - Platform & Submission: Submissions must be deployed as a Gradio App and hosted within the official Hugging Face hackathon organization. Docker-based environments are permitted as long as the underlying interface relies on a Gradio Space. |
| - Demo Video Requirement: Participants must submit a demo video demonstrating the app's functionality. This ensures the app can be evaluated even if API limitations or GPU exhaustion prevent judges from running it live. |
| - Social Proof Requirement: Participants are required to create a social media post showcasing their application and include the link to this post in the README file of their Gradio Space. |
| - App Limitations: Participants may submit as many applications as they want; however, if relying on the provided Zero GPU resources, participants are limited to a maximum of 10 Zero GPU applications per user. |
|
|
| ## 2. Tracks and Themes |
| The hackathon features two primary tracks aimed at stimulating different forms of creativity and utility: |
| |
| ### Backyard AI |
|
|
| Practical, problem-solving applications built to improve daily life for the creator or someone close to them. |
|
|
| eg. A custom storybook generator for a child, a personal study tutor, or a receipt/bill parser. |
|
|
|
|
| ### Thousand Token Wood |
|
|
| Whimsical, delightful, and highly creative AI-native applications that push the boundaries of fun and engagement. |
|
|
| eg. Interactive AI games, out-of-the-box entertainment tools, desktop pets. |
|
|
| ## 3. Prizes and Awards Structure |
|
|
| With a total of 29 different ways to win, the prize pool includes a robust mix of cash, credits, and hardware. |
|
|
| - Total Cash Pool: $48,000 USD distributed across multiple sponsor-specific categories. |
| - Compute & Credits: 20,000 Modal credits. |
| - Hardware: Two NVIDIA RTX GPUs. |
| - Subscriptions: 1-Year ChatGPT Pro subscriptions. |
|
|
| ### General |
|
|
| Each of the 2 tracks has the following prizes: |
|
|
| - 1st place: $4000 |
| - 2nd place: $2500 |
| - 3rd place: $1500 |
| - 4th place: $1000 |
| - Community choice (most upvotes): $2000 |
|
|
| No additional criteria |
|
|
| ### OpenBMB |
|
|
| $5k dedicated to Backyard AI and $5k to Thousand Token Wood. \ |
|
|
| - 1st place: $2500 |
| - 2nd place $1500 |
| - 3rd place: $1000 |
|
|
| Participants must build with MiniCPM models. |
|
|
| ### OpenAI |
|
|
| $10000 across all submissions |
|
|
| - 1st place: $5000 |
| - 2nd place: $3000 |
| - 3rd place: $1000 |
|
|
| Requires Codex-attributed commits in the connected GitHub repository or space. |
|
|
| Using Codex holistically (e.g. fine-tuning models, complex agents) ranks higher. |
|
|
| ### nvidia |
|
|
| 2x RTX 5080 GPUs across all submissions. |
|
|
| - Best space (as judged by nvidia team) |
| - Community engagement (likes/ interactions) |
|
|
| Participants must use Nemotron models. |
|
|
| ### modal |
|
|
| $20000 modal credits across all submissions for best use of Modal’s platform (as judged by the modal team): |
|
|
| - 1st place: $10000 credits |
| - 2nd place: $7000 credits |
| - 3rd place: $3000 credits |
|
|
| Participants must use the modal platform for either the development or runtime of their application and note this in the space README. |
|
|
|
|
| ### bonus awards |
|
|
| Additional criteria that participants can optionally meet to be eligible for extra prizes. |
|
|
| #### Off brand |
|
|
| $1500 |
|
|
| The best custom UI pushing past the default Gradio look. gr.Server is your friend. |
|
|
| #### tiny titan |
|
|
| $1500 |
|
|
| The best app built on a genuinely tiny model, less than ≤4B parameters. |
|
|
| #### best demo |
|
|
| $1000 |
|
|
| The full package: great app, great demo video, great social post. |
|
|
| #### best agent |
|
|
| $1000 |
|
|
| The best agentic app, under the 32B cap, of course. |
|
|
| #### Bonus Quest Champion |
|
|
| $2000 |
|
|
| The most bonus criteria met! |
|
|
| If there is a tie then the award will go to the most ambitious, highest quality submissions, despite the extra constraints! |
|
|
|
|
| #### judges wildcard |
|
|
| $1000 |
|
|
| For the entry that's amazing but fits no category. We know you're out there. |
|
|
|
|
| ## submitting |
|
|
| - ensure the appropriate criteria are met! |
| - join the build small hackathon organisation |
| - upload your submission as a space to the build small hackathon organisation. |
| - create a demo video demonstrating your space and why it is great! no humility here, sell it! upload the video to youtube, upload it as a file to the space, or put it somewhere else that is public (so we can watch it). |
| - post about it on social media! |
| - Update the README of your space with a link to the social media post and the demo video. |
| - Update the README frontmatter with tags detailing which tracks and badges you wish to be considered for. |
| - Add a short description to the readme explaining the idea, how you built it, the tech you used. |
|
|
