Partnering to Harmonise Fine-tuning for Bittensor
SN37 is Bittensor's new subnet for fine-tuning
Fine-tuning delivers the 'final mile' of AI model development, playing a critical role in whether or not an AI application actually meets the needs of its users.
Setting new standards for performance
By learning from the experience of SN6, the previous fine-tuning subnet, we can build on existing knowledge to set a new standard for fine tuning on Bittensor.
Overcoming technical hurdles together
Operating a fine-tuning Subnet demands the right pretrained base model, the right dataset, the right evaluation mechanism and sufficient compute - all aligned with Bittensor's competitive, open-source community. With Taoverse's experience and our engineering skill, we can overcome these challenges.
How SN37 Harmonises
Fine-Tuning
Enhancing performance from the start
Our core improvements will refine the codebase, removing inefficiencies and strengthening the subnet's design to increase competitiveness. We're also improving front-end experience by designing and launching a real-time leaderboard.
Driving continuous improvement as standard
We'll also harden the incentive structure, by deploying incentive structures that prohibit miners from deploying models resistant to further improvements. This ensures all miners can use the leading model as a basis for further training, helping to maintain continuous improvement.
Deeper integration for Bittensor
By layering through SN1's inference, SN9's pretraining, and SN13's data scraping, we can develop a feedback loop where data is sourced, refined, trained, and shared entirely within the Bittensor ecosystem. This can unlock entirely new use cases for the whole community.