bge-base-en-v1.5

bge-base-en-v1.5

Turns English text into number vectors for search and retrieval. A 109M-parameter embedding model from BAAI, for developers who build semantic search or retrieval for LLM apps through sentence-transformers.

Made by BAAI 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: bge-base-en-v1.5 is an English text embedding model from BAAI. It reads a piece of text and returns a vector of numbers, which you compare with other vectors to find related texts. It has 109M parameters and is used through the sentence-transformers library. The 1.5 release was made to ease the similarity distribution issue and to retrieve better without an instruction.

Who it suits: Developers who build search over their own documents fit best, along with people who give an LLM relevant passages before it answers. Researchers who fine-tune embedding models can start from it too.

Industry | Who uses it | How they use it | Example
Software | Developers building search | Embed documents and queries, then match them by vector closeness | Finding the help pages closest to a user's question
AI apps | LLM app builders | Fetch passages to hand to a chat model as context | Pulling product notes into a reply
Research | NLP researchers | Fine-tune or compare embedding models from a known base | Starting a domain-specific version from this model

You can skip it if your texts are not in English.

Where to use it: It runs in Python through sentence-transformers.

• Windows, macOS or Linux, where the files download through Python
• Python scripts and notebooks that load models with sentence-transformers
• Langchain, which the card says integrates the BGE models
• Search setups that pair an embedding model with a reranker

Use case | Who | What they get back
Semantic search | Developers | Vectors to compare, so close meanings rank first
Retrieval for LLMs | App builders | Relevant passages to place in a prompt
Fine-tuning base | Researchers | A small English embedding model to adapt

How to set it up: You need Python and a little disk space.

• Python on Windows, macOS or Linux
• The sentence-transformers library
• About 438 MB of disk for the F32 weights, or about 55 MB at 4-bit, which is an estimate

1. Install the Hugging Face hub library.
2. Download the model files with the hf command.
3. Install sentence-transformers and load the model from the downloaded folder.

pip install -U huggingface_hub
hf download BAAI/bge-base-en-v1.5

How to use it: The part of the card available here shows no run command. In practice you load the model with sentence-transformers, embed your texts, and compare the vectors. The table lists related pieces the card names.

Tool | What it does
bge-m3 | A sibling model that covers other languages, long inputs and extra retrieval modes
BAAI/bge-reranker-base | A cross-encoder that re-ranks the top documents an embedding model returns
BAAI/bge-reranker-large | The larger cross-encoder reranker for the same job
Langchain | A framework the BGE models are integrated into

Example: You turn a few support articles and one customer question into vectors with this model. You compare them and keep the articles whose vectors sit closest to the question.

For newcomers and experts: An embedding is a list of numbers that stands for the meaning of a text, and texts with similar meaning get lists that sit close together. The first small thing to try is embedding two short sentences and comparing them.

This model is a bi-encoder, so you embed documents once and compare queries against them. The card recommends adding a cross-encoder reranker over the top results when you want better ordering. At 109M parameters it is small enough to run on modest hardware, and the card's v1.5 notes say retrieval works without a query instruction. The record counts 508 fine-tuned versions by other people.

Download the files with hf
Load the model in sentence-transformers
Embed your documents and the query
Compare the vectors for closeness
Re-rank the top hits with a reranker

Good to know: The model covers English only, and the card sets it inside the larger FlagEmbedding project.

• Other versions: the record counts 31 quantized and 508 fine-tuned versions made by other people.
• Reranking: the card suggests re-ranking the top documents with its cross-encoder models.
• Task: the record lists feature extraction.

Title: Weight size by precision
F32: 438 MB
4-bit (estimate): 55 MB

• License: MIT.
• Cost to run: none named, no API key, account or paid service appears in the source.
• Data it reaches: it downloads files from Hugging Face, and the only text it reads is the text you embed.
• Not for: other languages, long inputs and extra retrieval modes, which the card hands to bge-m3.

On Hugging Face, 13 November 2026

• Task: feature extraction
• Library: sentence-transformers
• Parameters: 109M
• Weights: about 438 MB in F32; about 55 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 8,508,165
• Likes: 542
• License: mit
• Languages: en
• Quantized versions: 31
• Fine-tunes: 508
• Tags: sentence-transformers, pytorch, onnx, safetensors, bert, feature-extraction, sentence-similarity, transformers, mteb, model-index, text-embeddings-inference

Links

• Model: https://huggingface.co/BAAI/bge-base-en-v1.5
• Files: https://huggingface.co/BAAI/bge-base-en-v1.5/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download BAAI/bge-base-en-v1.5

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