Embeddinggemma 2 needs about 2.4 GB of memory at Q6_K_XL (a 0.2 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 6 quants, from 0.2 GB (Q4_K_XL) to 0.6 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 6.4 GB.
EmbeddingGemma 2 – a 768‑dimensional multimodal embedding model
EmbeddingGemma 2 maps text, images, video, and audio into a unified 768‑dimensional vector space. It has 740 million parameters and supports a 8 K token context window. The model is designed for low‑latency on consumer hardware.
Good at
- Multimodal embeddings
- Fast on-device inference
- Task‑steered representations
Limits
- No vision support in some quant variants
- Requires 8 K context for full performance
- Apache 2.0 license only
Best for
- Search and retrieval‑augmented generation
- Classification and clustering tasks
- Multilingual code embeddings
Who it’s for. Developers building on‑device semantic search or retrieval systems that need text, image, video, and audio embeddings. The model is suitable for mobile and laptop deployments where latency and footprint matter.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_XL | 0.2 GB | 2.3 GB | Fits | Get file → | |
| Q5_K_XL | 0.2 GB | 2.3 GB | Fits | Get file → | |
| Q6_K_XL | 0.2 GB | 2.4 GB | Fits | Get file → | |
| Q8_0 | 0.3 GB | 2.4 GB | Fits | Get file → | |
| BF16 | 0.6 GB | 2.7 GB | Fits | Get file → | |
| F16 | 0.6 GB | 2.7 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K_XL (0.18 GB) or Q5_K_XL (0.21 GB). If 12 GB VRAM is available, Q6_K_XL (0.25 GB) or Q8_0 (0.31 GB) offer higher precision. Full‑precision BF16 or F16 (0.56 GB) require more memory but give the best accuracy.
What GPU do you need for Embeddinggemma 2?
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 BF16 (0.6 GB file, about 2.7 GB in use).
Which Embeddinggemma 2 quant should you choose?
Q6_K_XL (0.2 GB) is the middle option in this repository. Very close to the original quality for most uses, at a clearly larger size than 4-bit.
If memory is tight, the smallest file is Q4_K_XL at 0.2 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Embeddinggemma 2
How much VRAM does Embeddinggemma 2 need?
About 2.4 GB at Q6_K_XL with an 8K context window. That is the 0.2 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 Embeddinggemma 2 on an 8 GB GPU?
Yes, with BF16: the 0.6 GB file needs about 2.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Embeddinggemma 2 GGUF?
It depends on the quant: Q4_K_XL is 0.2 GB, Q5_K_XL is 0.2 GB, Q6_K_XL is 0.2 GB, Q8_0 is 0.3 GB, BF16 is 0.6 GB, F16 is 0.6 GB.
What context length does Embeddinggemma 2 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 1.6 GB at 8K and 6.4 GB at 32K.
Where can I download Embeddinggemma 2 GGUF?
The files are on Hugging Face in the unsloth/embeddinggemma-2-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 Embeddinggemma 2 use?
Hugging Face lists the license as apache-2.0. 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 EmbeddingGemma 2 on 8 GB VRAM?
Yes, the Q4_K_XL and Q5_K_XL quantizations fit within 8 GB VRAM with an estimated 2.3 GB memory usage at 8 K context.
Does the model support vision?
Yes, EmbeddingGemma 2 includes a 170 M vision encoder that can be loaded selectively.
What license applies to this model?
The model is released under the Apache 2.0 license.