Bifrost-AI/Phi-4-bifrost-sol-3.8B

TEXT GENERATIONPricing:Input $0.32 / Cached $0.016 / Output $1.4Concurrent Unit Cost:1Model Size:3.8BQuant:BF16Context Size:32kPublished:May 23, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Bifrost-AI/Phi-4-bifrost-sol-3.8B is a 3.8 billion parameter instruction-tuned causal language model developed by Bifrost-AI, based on Microsoft's Phi-4 Mini Instruct architecture. It is specifically fine-tuned on blockchain-specific datasets, including the Solana Vanguard Challenge, to excel in blockchain coding and smart contract development within the Solana ecosystem. This model is optimized for tasks involving Rust and TypeScript for on-chain and client-side Solana development, offering a specialized solution for blockchain engineers.

Loading preview...

Phi 4 Bifrost SOL 3.8B: Solana Blockchain Specialist

This model is a specialized 3.8 billion parameter instruction-tuned variant of Microsoft's Phi-4 Mini Instruct, developed by Bifrost-AI. It has been supervised fine-tuned using the Solana Vanguard Challenge dataset, which comprises 1,000 in-depth questions covering the Solana ecosystem. This extensive dataset includes fundamental blockchain concepts, advanced on-chain programming in Rust and the Anchor framework, and client-side integration in TypeScript.

Key Capabilities

  • Blockchain-Specific Expertise: Optimized for tasks related to blockchain coding and smart contract development on the Solana platform.
  • Language Proficiency: Strong understanding and generation capabilities in Rust (for smart contracts) and TypeScript (for client-side dApp development).
  • Solana Ecosystem Knowledge: Trained on a comprehensive dataset covering various aspects of Solana, from security strategies to performance considerations.

Training Details

The model underwent a 16-hour and 42-minute training session on an NVIDIA GeForce RTX 3090, with a context size of 4098 tokens. The base Phi-4-mini-instruct model, on which this variant is built, supports a 128K token context length and was trained on 5 trillion tokens, focusing on high-quality, reasoning-dense data.

Intended Use Cases

This model is ideal for developers and researchers working within the Solana blockchain space who require a specialized language model for:

  • Generating or understanding Solana smart contract code in Rust.
  • Assisting with client-side dApp development using TypeScript for Solana.
  • Answering complex questions related to the Solana ecosystem.

Important Considerations

As an actively developed model, it is currently intended for research and development purposes. Users should exercise caution and conduct thorough testing before deploying it in commercial or real-world applications, as it may still produce unexpected or inaccurate outputs.

Finetunes
2 models