Fireworks 用不到 10 美元教会你微调 Qwen3-Embedding-8B，三个实验最高 nDCG 涨 36%，做 RAG 的值得看。
Fireworks 展示了如何用对比微调将 Qwen3-Embedding-8B 适配到特定领域，成本不到 10 美元。文章给出 3 个实验，其中最好成绩 nDCG 提升 36%。作者强调，检索不到文档的重排器无法生效，RAG 也会漏掉失败案例。完整步骤和结果见 Fireworks 官方博客。
Fine-tune your own embedding model for the price of a coffee. A great reranker can't surface a doc ...
Fine-tune your own embedding model for the price of a coffee. A great reranker can't surface a doc that was never retrieved. A RAG pipeline cannot cite a case it failed to retrieve. See how contrastive fine-tuning adapts Qwen3-Embedding-8B to specific domains for under $10 on Fireworks. Here are 3 straightforward experiments and results, e.g. +36% nDCG. Read the breakdown → fireworks.ai/blog/fine-tuni… 💬 0 🔄 1 ❤️ 2 👀 387 📊 1 ⚡