.png&w=256&q=75)
🤝 Top Collaborators
.png&w=640&q=75)
🤓 Latest Submissions
.png&w=828&q=75)
RK Routing Agent
# AMD Developer Hackathon — Track 1: General-Purpose AI Agent A token-efficient AI agent built for AMD Developer Hackathon Track 1. It handles natural language tasks across 8 capability categories — factual knowledge, mathematical reasoning, sentiment classification, text summarisation, named entity recognition, code debugging, logical/deductive reasoning, and code generation — while minimizing paid API usage. ## How it works A DistilBERT-based "Traffic Cop" router classifies each incoming prompt by category and difficulty, then decides whether to answer it locally (free) or route it to a Fireworks AI model (counted toward the leaderboard's token cost). - **Local inference**: Llama 3.2 3B (4-bit quantized GGUF), run in-process via `llama-cpp-python` - **Remote inference**: routed through `FIREWORKS_BASE_URL`, using only models listed in `ALLOWED_MODELS` - **Routing model**: a fine-tuned DistilBERT multi-task head (category + difficulty), bundled locally in the image — no external downloads at runtime ## Usage The container reads tasks from `/input/tasks.json` and writes results to `/output/results.json`: ```bash docker run --rm \ -e RUNNING_IN_DOCKER=true \ -e FIREWORKS_API_KEY=your_key \ -e FIREWORKS_BASE_URL=https://api.fireworks.ai/inference/v1 \ -e ALLOWED_MODELS="model-id-1,model-id-2" \ -v $(pwd)/input:/input \ -v $(pwd)/output:/output \ yourdockerhubusername/amd-hackathon-agent:latest ``` ## Requirements - `linux/amd64` platform - Runs within a 4GB RAM / 2 vCPU budget - Starts within 60 seconds, completes within 10 minutes
13 Jul 2026
