At the end of September, SN 5 underwent a major update. This article will outline the key details of the update, helping the market reassess and reevaluate the significance and potential of SN 5.
Subnet 5 Open Kaito
Emission：0.88%（2024–10–10）
Github：https://github.com/OpenKaito/openkaito
What is SN5?
The core goal of Bittensor Subnet 5 is to develop the world’s best-performing and most general-purpose text embedding models. It enables decentralized model training, evaluation, and service, while making these models accessible via APIs to support a wide range of applications.
SN 5 leverages the decentralized node network of Bittensor for dynamic evaluation and continuous model improvement.
Miners in SN 5 are responsible for training models using a continuously updated large-scale text corpus. They commit to delivering models with low latency and high throughput to meet the demands of downstream applications.
SN 5 validators rigorously evaluate models using multiplebenchmarks, constantly selecting better ones. Miners’ models are compared to existing SOTA text embedding models, ensuring SN 5 stays competitive and continuously improves.
How are Miners Evaluated?
In SN 5, Miners receive batches of text and use their models to generate text embeddings. Validators then assess the quality of these embeddings using a contrastive learning loss function.
The function is:
Here, c is the target embedding, x is the positive sample, and x’ is the negative sample. This process aims to maximize the mutual information between the positive sample x and the target embedding c.
Miners with higher-quality embeddings receive better scores, and their models are considered superior.
What Has SN 5 Achieved?
Under this incentive mechanism, SN 5 has made notable progress:
As of October 9, the text embedding models provided by Miners have shown improvement compared to the OpenAI baseline.
As SN 5 continues to enhance the quality of its models, users will gain access to highly general-purpose text embeddings that outperform the existing SOTA models. These models will be made publicly available through the SN 5 Validator API, enabling wide adoption and integration into various applications.
How to Evaluate the New SN 5?
SN 5 leverages decentralized optimization and training for text embedding models, contributing significantly to the Bittensor ecosystem and decentralized AI:
- Advancing the evolution of widely applicable models: SN 5 aims to develop the world’s best-performing, most versatile text embedding models. These models will be evaluated against an infinitely large and dynamic dataset, ensuring maximum domain generalization.
- Leveraging decentralized networks for dynamic evaluation and continuous improvement: By utilizing the Bittensor network, SN 5 avoids the limitations of centralized AI model development, enhancing transparency and resistance to censorship. Validators’ dynamic evaluations constantly push Miners to improve their models, ensuring that SN 5 not only surpasses existing SOTA models but also adapts to the latest real-world knowledge, maintaining competitiveness and pushing industry boundaries.
