nomic-embed-text-v1.5 - GGUF

Original model: nomic-embed-text-v1.5

Usage

Embedding text with nomic-embed-text requires task instruction prefixes at the beginning of each string.

For example, the code below shows how to use the search_query prefix to embed user questions, e.g. in a RAG application.

To see the full set of task instructions available & how they are designed to be used, visit the model card for nomic-embed-text-v1.5.

Description

This repo contains llama.cpp-compatible files for nomic-embed-text-v1.5 in GGUF format.

llama.cpp will default to 2048 tokens of context with these files. For the full 8192 token context length, you will have to choose a context extension method. The 🤗 Transformers model uses Dynamic NTK-Aware RoPE scaling, but that is not currently available in llama.cpp.

Example llama.cpp Command

Compute a single embedding:

./embedding -ngl 99 -m nomic-embed-text-v1.5.f16.gguf -c 8192 -b 8192 --rope-scaling yarn --rope-freq-scale .75 -p 'search_query: What is TSNE?'

You can also submit a batch of texts to embed, as long as the total number of tokens does not exceed the context length. Only the first three embeddings are shown by the embedding example.

texts.txt:

search_query: What is TSNE?
search_query: Who is Laurens Van der Maaten?

Compute multiple embeddings:

./embedding -ngl 99 -m nomic-embed-text-v1.5.f16.gguf -c 8192 -b 8192 --rope-scaling yarn --rope-freq-scale .75 -f texts.txt

Compatibility

These files are compatible with llama.cpp as of commit 4524290e8 from 2/15/2024.

Provided Files

The below table shows the mean squared error of the embeddings produced by these quantizations of Nomic Embed relative to the Sentence Transformers implementation.