AMD, in partnership with Lablab.ai, is hosting the AMD Developer Hackathon: ACT III, a competitive engineering sprint focused on accelerating large language models, computer vision pipelines, and agentic workflows on AMD hardware and the open-source ROCm ecosystem. Offering over $11,000 in prizes, hardware allocations, and cloud computing credits, the event challenges developers to demonstrate cost efficiency and high throughput with open model stacks.
The competition provides registered participants with direct access to the AMD Developer Cloud and dedicated resources through the AMD AI Developer Program (ADP), allowing engineers to train, fine-tune, and benchmark open-weight models without local VRAM limitations.
About the event
- Format: Global online build with collaborative on-site community nodes (hybrid model).
- Online build phase: October 12 – October 17, 2026.
- Final showcase and deadline: October 18, 2026.
- Core technologies: ROCm open software stack, AMD-optimized PyTorch, vLLM, and AMD Cloud GPU instances.
- Developer incentive: $100 in complimentary cloud compute credits upon registration via the AMD AI Developer Program.
Prizes
- 1st Place Overall — $5,000 cash + Premium AMD hardware package
- 2nd Place — $3,000 cash + AMD Developer Cloud credits
- 3rd Place — $1,500 cash + Acceleration lab mentorship
- Open Innovation Track (Evolus Track) — Specialized partner prizes and developer software subscriptions valued at $1,500+
Priority engineering tracks
Judges prioritize submissions that tackle genuine bottlenecks in distributed inference and autonomous agent architectures:
- Open inference optimization: Deploying frontier open models (Llama 3, DeepSeek, or Qwen) with custom Triton/ROCm kernels to maximize tokens-per-second throughput.
- Low-latency multi-agent systems: Local autonomous tool-calling pipelines taking full advantage of compute concurrency on AMD Instinct GPUs.
- Multimodal vision agents: Edge-compatible or server-scale multimodal agents performing real-time video telemetry with native FP8/INT4 quantization.
Submission guidelines
Teams must register and submit through Lablab.ai. Each submission requires a public repository with reproducible orchestration scripts, explicit documentation of the ROCm configuration, a concise technical pitch video, and a working demo prototype.
References
- Lablab.ai Official Challenge Page — Challenge timeline, rules, and entry portal. Date: 2026-10-07. Type: Official competition.
- AMD AI Developer Program (ADP) — ROCm documentation, PyTorch porting guides, and AMD Cloud access. Date: 2026-10-07. Type: Official documentation.
