The Gittensor, SN74 Roadmap
Where we are now:
The subnet is currently incentivizing miners to contribute to select open source repositories. While the first pieces of the roadmap are in development, we plan to continue raising Gittensor’s floor by fine tuning scoring to incentivize more frequent code contributions, higher quality code contributions & to mitigate scoring exploits. Additionally, by reducing the repository lists, selecting more specific high quality projects, and re-weighting repositories to steer the direction of contributions.
In order to unlock the full potential of what our miners can code, we need to align them to common goals. This can be through more precise repository lists (less targets), competing to solve issues first, or even just heavily weighted repositories that are attractive for miners to contribute to in order to keep a competitive score. For these reasons and the consideration of bringing early value to our alpha token, we find these roadmap milestones in our future planning
Issue bounties
- One of the initial ways we’ll provide value to our alpha token is enabling people to pay, in ALPHA, for specific github issues to be solved.
- Major score boosts will be rewarded to miners who close paid issues with their merged PRs.
- The issue boost multiplier will directly correlate to the amount of alpha paid
- Unpaid issue resolution will still boost miner score, but not as much relative to a paid issue
- There’s more evolution left for this such as issue payment being circulated to miners reward (in alpha), dynamically allocating emissions to supply additional bounty reward for miners, whatever we may find necessary
Market based repository weights
- Another exciting way we’ll provide value to our alpha token is by having market determined repository weights
- Users can stake their ALPHA as collateral into a smart contract and choose which repositories they want it to count toward.
- A short lockup window (TBD) will prevent people from instantly unstaking to game scoring.
- Each repository’s weight will directly reflect how much ALPHA has been committed to it
Repository Registration
- We will enable repositories to be registered/added into the master repository list using alpha as the payment
- LIke subnet/neuron registration there will be repository deregistrations, repository immunity, and registration cooldown
- This allows for a fully democratized, market-driven repository list
⚡️Ex Top miner on Subnet 14 has now created a optimization competition on Gittensor for the exact thing they were best in the world at... Inference Speed
Nvidia 4060, 8gb vram/64gb ram, I play around with gemma4 26B and qwen3.6 35B
Ofc with my hardware there's a lot of spillage to ram so gets nerfed to optimize space and a usable speed. (I see about 10-12 tok/s) I really like to just experiment and see if I can use it for basic agents/coding. It's actually pretty capable but takes forever still tbh. I can let experiments sit for 20+ min and come back pleasantly suprised at the results