The SolanaLM blog
Engineering notes on building a decentralized inference and federated-learning network. Most posts are written by the people running the gateway in production.
Running a SolanaLM node: real numbers from week one
Field notes from running an inference node for a week. Hardware, hours, payouts, and the operational reality that nobody puts in marketing copy.
Read post →The threat model for onion-routed LLM inference
Privacy-preserving inference sounds great in marketing copy. The honest engineering question is: what attacks does the 3-hop onion circuit actually stop, and what does it not?
Read post →OpenAI API compatibility is a distribution strategy, not an integration
Every decentralized inference network has to answer: how does an existing application start using us? Speaking OpenAI's schema is the cheapest, most honest answer.
Read post →FedAvg, FedProx, FedAdam, SCAFFOLD: what actually breaks in production federated learning
Four algorithms ship in the SolanaLM training runtime. They are not interchangeable. Here is what each one is for, what breaks when you pick wrong, and how to choose.
Read post →Why we picked Solana for per-request AI inference payments
A 400ms finality target and ~$0.00025 per-transaction fee are not nice-to-haves for a decentralized inference market. They are load-bearing. Here is the engineering case.
Read post →SolanaLM vs Bittensor: a deep dive on settlement, schemas, and subnet doctrine
Bittensor pioneered decentralized AI markets. SolanaLM bets on a different chain, a different schema, and a different definition of done. Here is the honest engineering comparison.
Read post →