bge m3 needs about 1.7 GB of memory at Q4_K_M (a 0.4 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 5 quants, from 0.4 GB (Q4_K_M) to 2.3 GB (F32).
BGE‑M3: a 0.567B BERT model for dense, multi‑vector, and sparse embeddings
BGE‑M3 is a 0.567B parameter BERT model trained on 8192‑token contexts. It supports dense, multi‑vector, and sparse retrieval and can process inputs from short sentences to long documents. The model is released under the MIT license.
Good at
- Dense retrieval
- Multi‑vector retrieval
- Sparse retrieval
Limits
- No vision support
- Requires 8192‑token context
- MIT license only
Best for
- Embedding generation for RAG pipelines
- Hybrid retrieval with BM25
- Long‑document similarity search
Who it’s for. Developers building retrieval‑augmented generation systems or large‑scale semantic search. The model is suitable for applications needing multilingual embeddings and multi‑granularity support.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 0.4 GB | 1.7 GB | Fits | Get file → | |
| Q6_K | 0.5 GB | 1.8 GB | Fits | Get file → | |
| Q8_0 | 0.6 GB | 1.9 GB | Fits | Get file → | |
| — | 1.2 GB | 2.5 GB | Fits | Get file → | |
| F32 | 2.3 GB | 3.6 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K_M (0.44 GB, ~1.7 GB memory) or Q6_K (0.5 GB, ~1.8 GB memory). If 12 GB VRAM is available, Q8_0 (0.63 GB, ~1.9 GB memory) offers higher precision. The F32 variant (2.27 GB, ~3.6 GB memory) is best for maximum accuracy when memory permits.
What GPU do you need for bge m3?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is F32 (2.3 GB file, about 3.6 GB in use).
Which bge m3 quant should you choose?
Q4_K_M is the usual starting point for bge m3: it balances file size and quality well. Here it is a 0.4 GB file.
The Q8_0 file is 0.6 GB, about 1.4 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about bge m3
How much VRAM does bge m3 need?
About 1.7 GB at Q4_K_M with an 8K context window. That is the 0.4 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run bge m3 on an 8 GB GPU?
Yes, with F32: the 2.3 GB file needs about 3.6 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of bge m3 GGUF?
It depends on the quant: Q4_K_M is 0.4 GB, Q6_K is 0.5 GB, Q8_0 is 0.6 GB, file is 1.2 GB, F32 is 2.3 GB.
What context length does bge m3 support?
The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 0.8 GB at 8K and 0.8 GB at 8K.
Where can I download bge m3 GGUF?
The files are on Hugging Face in the KimChen/bge-m3-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does bge m3 use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run BGE‑M3 on 8 GB VRAM?
Yes, the Q4_K_M quantized model is 0.44 GB and requires about 1.7 GB VRAM at 8192 context.
Does BGE‑M3 support multilingual embeddings?
Yes, it supports more than 100 working languages as stated in its parent model card.
Is BGE‑M3 suitable for hybrid retrieval?
Yes, it can perform both embedding and sparse retrieval, enabling hybrid pipelines with BM25.