Split 31B needs about 51.4 GB of memory at Q4_K_M (a 18.7 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 10 quants, from 17.8 GB (Q4_K_S) to 36.8 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 128.8 GB.
Split 31B – a 30.7B‑parameter Gemma model with vision and roleplay abilities
Split 31B is a 30.7B‑parameter Gemma model derived from Ateron/Gemma-4-Dark-Thoughts-V2-31B. It supports vision and roleplay tasks and is released under the apache-2.0 license. The model was trained on 262144 context tokens.
Good at
- Vision tasks
- Roleplay
- Long‑context dialogue
Limits
- No vision support in llama.cpp
- Large VRAM requirement
- No tool‑calling support
Best for
- Chatbots with visual input
- Creative storytelling
- Long‑form roleplay
Who it’s for. Developers building chat or creative applications that need vision and roleplay capabilities. The model is suitable for users with sufficient GPU memory to run large quantizations.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_S | 17.8 GB | 50.5 GB | Too big | Get file → | |
| Q4_K_M | 18.7 GB | 51.4 GB | Too big | Get file → | |
| Q5_K_S | 21.3 GB | 54.0 GB | Too big | Get file → | |
| Q5_K_M | 21.8 GB | 54.6 GB | Too big | Get file → | |
| Q4_K_M | 23.1 GB | 55.8 GB | Too big | Get file → | |
| Q6_K | 25.2 GB | 57.9 GB | Too big | Get file → | |
| Q5_K_M | 25.3 GB | 58.0 GB | Too big | Get file → | |
| Q6_K | 27.6 GB | 60.3 GB | Too big | Get file → | |
| Q8_0 | 32.6 GB | 65.3 GB | Too big | Get file → | |
| Q8_0 | 36.8 GB | 69.6 GB | Too big | Get file → |
For 8 GB VRAM use Q4_K_S (17.76 GB, 50.5 GB memory at 8k context). For 16 GB VRAM use Q5_K_S (21.31 GB, 54 GB memory). For 24 GB VRAM use Q6_K (25.2 GB, 57.9 GB memory).
What GPU do you need for Split 31B?
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: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which Split 31B quant should you choose?
Q4_K_M is the usual starting point for Split 31B: it balances file size and quality well. Here it is a 18.7 GB file.
The Q8_0 file is 32.6 GB, about 1.7 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.
If memory is tight, the smallest file is Q4_K_S at 17.8 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 Split 31B
How much VRAM does Split 31B need?
About 51.4 GB at Q4_K_M with an 8K context window. That is the 18.7 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 Split 31B on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 50.5 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Split 31B GGUF?
It depends on the quant: Q4_K_S is 17.8 GB, Q4_K_M is 18.7 GB, Q5_K_S is 21.3 GB, Q5_K_M is 21.8 GB, Q4_K_M is 23.1 GB, Q6_K is 25.2 GB and 4 more.
What context length does Split 31B support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 32.2 GB at 8K and 128.8 GB at 32K.
Where can I download Split 31B GGUF?
The files are on Hugging Face in the Nimbz/Split-31B repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Split 31B 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 Split 31B on 8 GB VRAM?
Use the Q4_K_S quantization; it requires about 50.5 GB memory at 8k context, so 8 GB VRAM is insufficient.
Does Split 31B support vision in llama.cpp?
No, the model has vision capability but llama.cpp does not support vision for this model.