Hi,
A proposal for a sponsored challenge rather than a donation.
Competitors on Zindi pay for their own compute, which quietly determines who can enter a challenge involving large models at all. That is a barrier the platform can remove for a specific competition.
Gonka is a decentralized inference network. Capacity comes from independent GPU operators coordinated by an open protocol, which is where the cost structure differs from a single vendor. DeepSeek Flash is available at a 1M token context window over an OpenAI-compatible endpoint.
The proposal: a challenge on African language tasks where every entrant gets API access at no cost, so the competition tests ideas rather than budgets. Whatever comes out of it, particularly the error analysis on where the model fails on African languages, should be published openly and belongs to the community rather than to Gonka.
Entrants can test it before any of that is agreed. The broker gives the first 100 million tokens free:
1. Get an API key at https://inference.dahl.global
2. Point any OpenAI-compatible client at https://inference.dahl.global/v1
3. Request the DeepSeek Flash model, key as a bearer token:
curl https://inference.dahl.global/v1/chat/completions \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json"
Discord is open for entrants who need setup help during a competition, and contributors answer directly:
Code: https://github.com/gonka-ai/gonka
Who handles challenge sponsorship?