Meeting notes 27/09/24

[iboB]

The topic was the post-MVP focus and the road to Beta

• Create public roadmap and mission statement
• Local SDK:
  • More features: Potential API features #47 is a start, but we need a more focused feature process. Possibly the initial user cohort will help. We should discuss and at least somewhat formalize feature procedures
    • RAG and LORA adapters are the only confirmed ones
  • Continue development of asset deduplication: #96 Investigate using Android's download manager as an asset source #63 Investigate using IOS download manager as an asset source #64 will likely leak to post MVP. Asset management should be a primary focus of Beta
  • Performance benchmarks and automated benchmarking on multiple platforms and hardware must be available here
  • Add support for more inference classes
  • Add support for more programming languages
  • Take advantage of mobile GPUs on Android
• Server (Remote provider and Server #75)
  • We must have server support for Beta
  • The basic server support will evolve in our private cloud offerings:
    • Running CUDA on TEE, remote attestation
    • Seamless pairing between the user devices for federated inference (vpn, private cloud, personal cloud)
• Router and remotes. Some of thee items below may not be complete for Beta, but we must have initial specs for all of
  • Create Model Manifest #91 Model sources #61 Model manifest structure and queries RnD #105
  • Clarify model sources, model database
  • Benchmarking SDK
  • Permissionless model uploads
  • Fine tuned models
  • Revenue sharing options
  • Routing strategy. Basic routing should be available for Beta

Not for Beta. Will be discussed post beta:

• Training utilities:
  • Federated Learning
  • Reinforcement Learning from Human Feedback
  • Direct Preference Optimization

[iboB]

Some items moved to #169

[iboB]

Other items moved to private discussion. Closing