Subnet 52
Dojo
Tensorplex
The Dojo Subnet by Tensorplex enables decentralized, community-driven generation of high-quality, multi-modal AI training data
SN52 : Dojo
| Subnet | Description | Category | Company |
|---|---|---|---|
| SN52 : Dojo | Human-generated datasets crowdsoursing | Data Pipeline | Tensorplex Labs |
Tensorplex has created the Dojo Subnet as a decentralized platform to meet the high demand for specialized, multi-modal AI data. As artificial intelligence rapidly evolves, driven by powerful computing, expansive model sizes, and the rise of multi-modal applications, high-quality, targeted data is essential. AI applications now stretch beyond traditional roles like chatbots into fields like robotics, further fueling the need for diverse, multi-modal datasets.
The open-sourcing of advanced language models by companies like Meta has empowered smaller developers and individual teams, allowing them to enter the AI space more easily. However, while this has democratized AI development, it also exposes a key hurdle: access to credible, high-quality training data at scale, especially for those working across various modalities, from text and audio to video and images.
In response, Dojo enables contributors to generate and share multi-modal, annotated, and labelled data, meeting the increasing demand for such resources in fine-tuning and model training. This project represents a significant step forward in making AI data more accessible, high-quality, and community-driven.
Tensorplex has designed the Dojo Subnet with robust, innovative features to ensure data quality and integrity in a decentralized ecosystem.
Key Features
Dojo offers several powerful mechanisms to maintain high standards:
- Synthetic Task Generation: Leveraging advanced Large Language Models (LLMs), Dojo generates unique tasks to gather human feedback data, which can refine open-source models.
- Synthetic Ground Truth Validation: Validators can synthetically generate partial ground truths to assess participant responses accurately, ensuring response quality.
- Obfuscation: Dojo incorporates obfuscation techniques to safeguard against sybil attacks and verify that contributions come from genuine human users.
Use Cases
The Dojo Subnet opens up a variety of applications:
- Synthetically Generated Tasks: These tasks provide a starting point for human participants and can be applied directly to model training and fine-tuning.
- Cross-subnet Validation: Validators assess response quality across other Bittensor subnets, motivating miners to enhance their outputs.
- External Data Acquisition: Entities outside of Bittensor can access Dojo’s high-quality human-generated data, allowing a broader user base to benefit from the subnet.
Through Dojo, they’re building a platform that tackles quality control, human verification, and sybil protection, driving more equitable AI development.
Participant Benefits
Dojo provides key advantages for those who contribute data through the subnet:
- Open Platform: Dojo is accessible to anyone capable of contributing, enabling wide participation and rich data diversity.
- Flexible Work Environment: Participants can complete tasks from anywhere, working at their convenience.
- Prompt Payment: Participants are consistently rewarded upon completing tasks within deadlines, provided the work is accepted by the subnet.
Subnet Mechanism
Miners’ Role
Miners in Dojo are tasked with gathering participants to complete tasks. They must curate pools of participants who are skilled in specific domains to excel.
Validators’ Role
Validators act as Instructors, Augmenters, Output Generators, and Obfuscators during task generation. They also manage scoring, set rewards, and establish trust levels for miners, ensuring system integrity and reward fairness.
Tensorplex is backed by a number of investors specialising in Artificial Intelligence, Blockchain, Consumer Applications, Decentralised Finance and Web3 Security. This list includes:
Collab Currency
Canonical Crypto
Digital Currency Group
Accomplice
Republic
Quantstamp
PetRock Capital
Mechanical Capital
Merit Circle
Athena Nodes
Amber
Hansa
Zellic