nomic-embed-text-v1.5

nomic-embed-text-v1.5

Turns English text into embedding vectors for search, clustering and classification, and the vectors can be cut to a smaller size. It suits developers building retrieval apps.

Made by nomic-ai on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: nomic-embed-text-v1.5 is a text embedding model with 137M parameters. It turns English text into vectors, so you can compare meaning, search documents or group similar texts. Its vectors support Matryoshka resizing, which lets you keep only the first part of each one.

Who it suits: It suits developers and data teams who need text vectors for retrieval, grouping or classification. You need some Python comfort to run it.

Industry | Who uses it | How they use it | Example
Software | Developers of retrieval apps | Embed documents and user questions with different prefixes | A search box over a pile of internal notes
Data science | Analysts | Embed texts to cluster them, find topics or remove near duplicates | Grouping short feedback comments by theme
Machine learning | ML engineers | Use the vectors as input features for a classifier | Training a classifier on top of the embeddings

People who need text in other languages can skip it, since the record lists English only.

Where to use it:

• Python on Windows, macOS or Linux
• The sentence-transformers library
• The transformers library, with the pooling code from the card
• A RAG pipeline that needs document and query vectors

Use case | Who | What they get back
RAG indexing and querying | App developers | Document vectors and matching question vectors
Clustering and deduplication | Analysts | Vectors that group similar texts and expose repeats
Classification features | ML engineers | Vectors a classifier can train on

How to set it up:

• Python on your computer
• The sentence-transformers library and PyTorch
• About 547 MB of disk for the F32 weights

1. Install the Hugging Face hub package.
2. Download the model files.
3. Load the model by name in Python.

pip install -U huggingface_hub

hf download nomic-ai/nomic-embed-text-v1.5

How to use it: Every text you embed must start with a task prefix that tells the model what the vector is for. Add the document prefix to texts you index and the query prefix to questions. Then call encode.

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("nomic-ai/nomic-embed-text-v1.5")
sentences = ['search_query: Who is Laurens van Der Maaten?']
embeddings = model.encode(sentences)
print(embeddings)

To shrink the vectors, the card normalizes them, keeps the first 512 values and normalizes again.

import torch.nn.functional as F
from sentence_transformers import SentenceTransformer

matryoshka_dim = 512

model = SentenceTransformer("nomic-ai/nomic-embed-text-v1.5")
sentences = ['search_query: What is TSNE?', 'search_query: Who is Laurens van der Maaten?']
embeddings = model.encode(sentences, convert_to_tensor=True)
embeddings = F.layer_norm(embeddings, normalized_shape=(embeddings.shape[1],))
embeddings = embeddings[:, :matryoshka_dim]
embeddings = F.normalize(embeddings, p=2, dim=1)
print(embeddings)

Tool | What it does
search_document | Prefix for texts embedded as documents, such as a RAG index
search_query | Prefix for texts embedded as questions that documents could answer
clustering | Prefix for texts you want to group, find topics in or deduplicate
classification | Prefix for texts whose vectors feed a classifier

Example: You build a question answering app over your own files. You put the document prefix in front of each text chunk and the query prefix in front of each user question, then encode both. Each text comes back as a vector you can compare.

For newcomers and experts: An embedding turns a sentence into a list of numbers, and similar sentences get similar lists. A good first try is to encode one query with its prefix and print the result. Forgetting the prefix is the easiest mistake.

Underneath, the model was trained with Matryoshka representation learning, so the first dimensions carry most of the meaning and you can truncate. The card says it supports sequence lengths past 2048 tokens. The card also names a vision model from the same team that shares this model's vector space.

Add a task prefix to each text
Encode the text with the model
Optionally cut vectors to 512 values
Normalize and compare the vectors

Good to know:

• License: Apache 2.0.
• Cost to run: none named. The steps need no API key or paid service.
• Data it reaches: it downloads files from the Hugging Face Hub. Your texts are encoded on your own machine.
• Not for: texts without a task prefix, which the card says the model must have.
• Newer releases of transformers and sentence-transformers no longer need the trust_remote_code flag for this model.

Title: Weight size in megabytes
F32: 547 MB
4-bit estimate: 68 MB

On Hugging Face, 13 November 2026

• Task: sentence similarity
• Library: sentence-transformers
• Parameters: 137M
• Weights: about 547 MB in F32; about 68 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 11,809,681
• Likes: 994
• License: apache-2.0
• Languages: en
• Quantized versions: 40
• Fine-tunes: 42
• Tags: sentence-transformers, onnx, safetensors, nomic_bert, feature-extraction, sentence-similarity, mteb, transformers, transformers.js, model-index, text-embeddings-inference

Links

• Model: https://huggingface.co/nomic-ai/nomic-embed-text-v1.5
• Files: https://huggingface.co/nomic-ai/nomic-embed-text-v1.5/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download nomic-ai/nomic-embed-text-v1.5

Then follow the usage in its model card. Nothing in this kit runs by itself.