Sakura Gemma4 E2B EmbeddingGemma2 DualMode needs about 4.7 GB of memory at Q5_K_M (a 3.6 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 2 quants, from 3.6 GB (Q5_K_M) to 3.6 GB (Q5_K_M). Longer context needs more: at 32K tokens the KV cache alone adds about 2.3 GB.
Gemma4 E2B DualMode: generation and 768‑d embeddings in one GGUF
Sakura Gemma4 E2B DualMode is a 4.65 billion‑parameter model that supports both chat generation and 768‑dimensional embeddings. It uses the Gemma4 architecture and is quantized to Q5_K_M. The model requires a small llama.cpp patch for embed mode.
Good at
- Chat generation
- 768‑d embeddings
- Bidirectional attention for retrieval
Limits
- Requires patched llama.cpp for embed mode
- No vision support
- Context window up to 256K tokens
Best for
- Local RAG pipelines
- Chatbots needing embeddings
- Edge devices with 8‑GB VRAM
Who it’s for. Developers building local LLM applications that need both generation and embedding capabilities in a single file. Ideal for mobile or laptop deployments where memory is limited.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q5_K_M | 3.6 GB | 4.7 GB | Fits | Get file → | |
| Q5_K_M | 3.6 GB | 4.7 GB | Fits | Get file → |
The Q5_K_M quantization produces a 3.63 GB file and needs about 4.7 GB of RAM at 8 k context. For 8 GB VRAM devices, this quant is suitable. If more memory is available, the same quant can be used without changes.
What GPU do you need for Sakura Gemma4 E2B EmbeddingGemma2 DualMode?
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: no quant fits at an 8K context.
- 6 GB VRAM: the best quant that fits is Q5_K_M (3.6 GB file, about 4.7 GB in use).
Which Sakura Gemma4 E2B EmbeddingGemma2 DualMode quant should you choose?
Q5_K_M (3.6 GB) is the middle option in this repository. Closer to the original model than 4-bit, for a moderate increase in size. A good pick when you have memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Sakura Gemma4 E2B EmbeddingGemma2 DualMode
How much VRAM does Sakura Gemma4 E2B EmbeddingGemma2 DualMode need?
About 4.7 GB at Q5_K_M with an 8K context window. That is the 3.6 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 Sakura Gemma4 E2B EmbeddingGemma2 DualMode on an 8 GB GPU?
Yes, with Q5_K_M: the 3.6 GB file needs about 4.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Sakura Gemma4 E2B EmbeddingGemma2 DualMode GGUF?
It depends on the quant: Q5_K_M is 3.6 GB, Q5_K_M is 3.6 GB.
What context length does Sakura Gemma4 E2B EmbeddingGemma2 DualMode support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.6 GB at 8K and 2.3 GB at 32K.
Where can I download Sakura Gemma4 E2B EmbeddingGemma2 DualMode GGUF?
The files are on Hugging Face in the webmp3/Sakura-Gemma4-E2B-EmbeddingGemma2-DualMode-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 Sakura Gemma4 E2B EmbeddingGemma2 DualMode use?
Hugging Face lists the license as gemma. 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 this on 8 GB VRAM?
Yes, the Q5_K_M quant uses about 4.7 GB of RAM at 8 k context, fitting within 8 GB VRAM with headroom.
Do I need a special version of llama.cpp?
Yes, a small llama.cpp patch is required for embed mode; stock llama.cpp cannot load the file.
Is the model multimodal?
No, this version supports only text generation and embeddings; it has no vision or audio capability.