Automate end-to-end on-chain AI model deployment with Frostbite Cauldron on Solana.
npx skills add https://github.com/reflow-research/Cauldron --skill frostbite-cauldron
Please help me install this Agent Skill. Skill: frostbite-cauldron Source: https://github.com/reflow-research/Cauldron/tree/main Command: npx skills add https://github.com/reflow-research/Cauldron --skill frostbite-cauldron
This Skill streamlines the end-to-end Frostbite/Cauldron workflow on Solana, including manifest authoring, guest builds, seeded deterministic accounts, weight upload, and on-chain inference.
Initialize a project with cauldron init, prepare weights with convert, build the guest, set up seeded accounts, upload weights, load the guest, and invoke to get on-chain results.
High-intent search queries and answers about installing and using this skill.
The Frostbite Cauldron workflow automates on-chain AI model deployment on Solana by handling manifest validation, weight conversion, guest builds, and deterministic seeded account creation before executing on-chain inference.
Supported Frostbite templates for on-chain inference include linear, mlp, cnn1d, tiny_cnn, and tree architectures, allowing diverse neural network designs to be packaged and deployed on Solana.
Seeded deterministic accounts use v3 addressing and wallet integration to guarantee multi-user safety, creating predictable on-chain addresses for weight storage and program interaction during deployment.
The deployment process follows cauldron init, weight conversion, guest build, seeded account initialization, weight upload, program load, input-write, and on-chain invoke to retrieve inference results.
Yes, the workflow depends on torch and safetensors, enabling conversion of PyTorch model weights into the packaged format required for on-chain upload and Solana inference.
On-chain deployment is constrained to supported Frostbite templates like linear, mlp, cnn1d, tiny_cnn, and tree architectures, requiring deterministic seeded accounts, which limits deployment of large-scale models.
Simon-Copilot-StudioRemove refusal directions from open-weight LLM weights using mechanistic interpretability techniques.
SterdamSend USDC to named agents on Solana devnet via the SolClaw API.
tarotmansaManage Solana wallets, transfers, swaps, and token launches via CLI scripts.
ritual-foundationCall the ONNX precompile from smart contracts for deterministic on-chain ML inference.
