Nomic nomic-embed-text-v1.5
FreeLong context, Apache 2.0 license
FreeFree tier
Inputs: text
About Nomic nomic-embed-text-v1.5
nomic-embed-text-v1.5 is a resizable production embedding model developed by Nomic AI, utilizing Matryoshka Representation Learning to allow flexible dimensionality. It supports four task-specific instruction prefixes: search_document, search_query, clustering, and classification, enabling optimized embeddings for retrieval-augmented generation (RAG), clustering, and classification tasks. The model is now multimodal when combined with nomic-embed-vision-v1.5, aligning vision embeddings to the same space. It is released under the Apache 2.0 license and can be used via Sentence Transformers, Transformers, or the Nomic API.
Key Features
Task-specific instruction prefixes (search_document, search_query, clustering, classification)
Resizable embedding dimensions via Matryoshka Representation Learning
Multimodal alignment with nomic-embed-vision-v1.5
Supports Sentence Transformers, Transformers, and Transformers.js
Available via Nomic API and Infinity serving
Apache 2.0 license, free and open source
Pros & Cons
Pros
- Free and open source under Apache 2.0 license
- Flexible with task-specific prefixes for optimized performance
- Resizable embedding dimensions allow trade-off between storage and accuracy
- Multimodal capability through alignment with vision model
- Easy integration with popular libraries like Sentence Transformers and Transformers
Cons
- Requires precise task instruction prefix for each use case, adding complexity
- Older versions of Transformers may require trust_remote_code=True
- Documentation on training and fine-tuning is limited on the main page
Best For
Retrieval-Augmented Generation (RAG) with search_document and search_query prefixesClustering and topic discovery using the clustering prefixClassification tasks using the classification prefixMultimodal search and retrieval when paired with vision model
FAQ
What task instruction prefixes does nomic-embed-text-v1.5 support?
It supports four prefixes: search_document, search_query, clustering, and classification. Each is designed for a specific task (e.g., search_document for indexing documents in RAG).
Is nomic-embed-text-v1.5 multimodal?
Yes, it is now multimodal when used with nomic-embed-vision-v1.5, which is aligned to the same embedding space, enabling joint text and image embeddings.
What license is nomic-embed-text-v1.5 released under?
It is released under the Apache 2.0 license, making it free for commercial and research use.
How do I use nomic-embed-text-v1.5 with Sentence Transformers?
Load the model with `SentenceTransformer('nomic-ai/nomic-embed-text-v1.5')` and prepend the appropriate task prefix to each input text.