Qwen Local Hackathon 2026
Take a small Qwen model, make it better at one thing, make it smaller, and run it on your own machine. One week, fully online, free to join.
The Qwen Local Hackathon is organized by Qwen Pakistan for developers, students, and anyone curious about running open models locally. You don't need a big GPU. A laptop, an Apple Silicon Mac, or a free Colab or Kaggle notebook is enough to take part.
Join the Qwen Pakistan WhatsApp community: Join WhatsApp Group
What you'll build
Pick a small open-weight Qwen model and do at least one of these:
Fine-tune it for a specific task with LoRA or QLoRA
Quantize it so it fits and runs well on everyday hardware
Run it locally with tools like llama.cpp, Ollama, MLX, or vLLM
Then build a small demo that shows it doing something useful. A coding helper, a support bot for a niche domain, a study assistant, a tool for your own workflow. If it runs on your machine and solves a real problem, it counts.
We're not training models from scratch. The goal is adapting existing Qwen models and showing what they can do on modest hardware.
How it runs
Kickoff on Google Meet. Rules, resources, starter guides, and Q&A
WhatsApp. Join the Qwen Pakistan community using the link above. Registered participants get added to a private hackathon group inside it before kickoff, where announcements and help happen
Help all week. The organizer is available throughout the build week to answer questions and help you get unstuck
Timeline (Pakistan time)
Fri 20 Nov, 11:59 PM: Registration closes
Mon 23 Nov, 7:00 PM: Live kickoff on Google Meet
Mon 23 to Sat 28 Nov: Build week, with help on WhatsApp
Sat 28 Nov, 11:59 PM: Submissions close
Sun 29 Nov, 8:00 PM: Results announced
What to submit
Your model. The fine-tuned weights or adapter, and the quantized version if you made one, published on ModelScope or Hugging Face
A public GitHub repo with a README covering the base model, your data, your method, the hardware you used, and before-and-after results
A 2 to 3 minute demo video of the model running locally
How projects are judged
Usefulness: does it solve a real problem?
Technical work: is the fine-tuning or quantization done properly?
Measured results: did you show before-and-after numbers, not just vibes?
Runs locally: how well does it perform on modest hardware?
Write-up and demo: could someone else understand and reproduce it?
Rewards
Top projects win Qwen API credits and official Qwen swag. Exact prizes are announced at kickoff. Winners will need a free ModelScope account to receive their credits.
Rules
Solo or teams of up to 3
Use any small open-weight Qwen model, and follow its license
Build during the hackathon week. Older projects don't qualify
Only use datasets you're allowed to use, and never include private or personal data
Judges' decisions are final
Useful tools
Unsloth, LLaMA-Factory, and MLX for fine-tuning. llama.cpp, Ollama, and MLX for running models locally. Free GPUs on Google Colab and Kaggle if you don't have your own.
Questions? Reach out to Qwen Pakistan on LinkedIn or in the WhatsApp community.