Views
No views yet
Qwen3 Bifrost SOL 4B is a specialized, fine-tuned variant of the Qwen3-4B base model crafted for blockchain coding and smart contract development within the Solana ecosystem. It was trained using the Solana Vanguard Challenge dataset, which comprises 1,000 in-depth questions covering a broad spectrum of topics: fundamental blockchain concepts, advanced on-chain programming in Rust (including security, state management, CPIs, and PDAs), as well as client-side integration in TypeScript with tools like @solana/web3.js, wallet adapters, and Metaplex for NFTs. This model was trained over 11 hours and 22 minutes using an NVIDIA GeForce RTX 3090, and features ongoing development with further fine-tuning, benchmarking, and future extensions planned (such as C# coverage via Solnet). Intended for research and development, Bifrost SOL 4B should not be deployed in production environments without thorough testing, as it may still produce unexpected or biased outputs despite alignment efforts using SFT and DPO.
| File Name | Quant Type | File Size |
|---|---|---|
| Qwen3-Bifrost-SOL-4B.BF16.gguf | BF16 | 8.05 GB |
| Qwen3-Bifrost-SOL-4B.F16.gguf | F16 | 8.05 GB |
| Qwen3-Bifrost-SOL-4B.F32.gguf | F32 | 16.1 GB |
| Qwen3-Bifrost-SOL-4B.Q2_K.gguf | Q2_K | 1.67 GB |
| Qwen3-Bifrost-SOL-4B.Q3_K_L.gguf | Q3_K_L | 2.24 GB |
| Qwen3-Bifrost-SOL-4B.Q3_K_M.gguf | Q3_K_M | 2.08 GB |
| Qwen3-Bifrost-SOL-4B.Q3_K_S.gguf | Q3_K_S | 1.89 GB |
| Qwen3-Bifrost-SOL-4B.Q4_K_M.gguf | Q4_K_M | 2.5 GB |
| Qwen3-Bifrost-SOL-4B.Q4_K_S.gguf | Q4_K_S | 2.38 GB |
| Qwen3-Bifrost-SOL-4B.Q5_K_M.gguf | Q5_K_M | 2.89 GB |
| Qwen3-Bifrost-SOL-4B.Q5_K_S.gguf | Q5_K_S | 2.82 GB |
| Qwen3-Bifrost-SOL-4B.Q6_K.gguf | Q6_K | 3.31 GB |
| Qwen3-Bifrost-SOL-4B.Q8_0.gguf | Q8_0 | 4.28 GB |
