使用 FastEmbed 准备本地检索嵌入
Generate dense, sparse, image, and reranking embeddings locally before writing vectors into Qdrant or another retrieval stack for agent memory and RAG workflows.
Prerequisites
Python, FastEmbed, optional qdrant-client and Qdrant vector database
Installation
Use the upstream install or setup path that matches your environment:
- pip install fastembed
- pip install fastembed-gpu
- pip install qdrant-client[fastembed]
- pip install qdrant-client[fastembed-gpu]
Requirements and caveats from upstream:
- FastEmbed is a lightweight, fast, Python library built for embedding generation. We support popular text models. Please [open a GitHub issue](https://gi...
- Light: FastEmbed is a lightweight library with few external dependencies. We don't require a GPU and don't download GBs of PyTorch dependencies, and instead use the ONNX Runtime. This makes it a great candidate for se...
- python
Basic usage or getting-started notes:
- The default text embedding (TextEmbedding) model is Flag Embedding, presented in the MTEB leaderboard. It supports "query" and "passage" prefixes for the input text. H...
- Check our example for detailed instructions, CUDA 12.x support and troubleshooting of the common issues.
- or
- Source: https://github.com/1991513ccie-png/skills
- Extracted from upstream docs: https://raw.githubusercontent.com/qdrant/fastembed/HEAD/README.md
Documentation
- https://qdrant.github.io/fastembed/
