Modelpublic

Bifrost-AI/Qwen3-Bifrost-SOL-4B-GGUF

sourceHugging Facemitupdated 1y agoView on Hugging Face

1likes97downloads

Model Card

Qwen3 Bifrost SOL 4B

This fine-tuned variant of the Qwen3 4B model was supervised fine-tuned on blockchain-specific datasets(Bifrost-AI/Solana-Vanguard-Challenge), optimized for downstream tasks in blockchain coding and smart contract development on the Solana ecosystem.

The Solana Vanguard Challenge dataset, comprising 1,000 diverse and in-depth questions, offers full-spectrum coverage of the Solana ecosystem. It spans fundamental blockchain concepts, advanced on-chain programming in Rust and the Anchor framework, client-side integration in TypeScript, detailed security strategies, and performance as well as regulatory considerations.

Qwen3 Bifrost SOL 4B is in active development with additional fine-tuning sessions, & benchmark statistics coming soon!

Provided Quants

Link	Type	Size/GB	Notes
GGUF	IQ1_S	1.1	very low quality
GGUF	IQ1_M	1.2	very low quality
GGUF	TQ1_0	1.2	very low quality
GGUF	IQ2_S	1.4	fast, lower quality
GGUF	Q2_K	1.6	fast, lower quality
GGUF	Q4KM	2.5	fast, recommended
GGUF	Q4KS	2.3	fast, recommended
GGUF	Q4_0	2.3	fast, recommended
GGUF	Q5KS	2.7
GGUF	Q5KM	2.8
GGUF	Q6_K	3.1	very good quality
GGUF	Q8_0	4.0	fast, best quality
GGUF	F16	7.7	16 bpw, highest quality

Training Session:

—Time: 11 hours & 22 minutes
—GPU: NVIDIA GeForce RTX 3090
—Batches: 1000
—Context-Size: 2043
—Batch-size: 1
—Learning-rate: 2e-5
—Training-loss: 1.06
—Eval-loss: 0.81

Dataset Composition

—Total Questions: 1,000
—Languages Covered:
—Rust: On-chain smart contract development, security best practices, advanced state management, CPIs, PDAs, and more.
—TypeScript: Client-side integration using @solana/web3.js, wallet adapters, Metaplex for NFT protocols, dynamic transaction composition, and front-end dApp development.
—Planned Extensions:
—C# (Solnet): To be integrated later for .NET ecosystem coverage.

Disclaimer

We do not recommend using Qwen3 Bifrost SOL 4B in commercial or real-world applications without further testing and development. This current model(v1) is intended for research and development purposes. While efforts have been made to align it using SFT and DPO, it may still produce outputs that are unexpected, biased, or inaccurate. Please use responsibly.