Wanabi Gemma4 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 29 quants, from 8.7 GB (IQ2_XXS) to 61.4 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 128.8 GB.
Wanabi Gemma4 31B – a 30.7B‑parameter Instruct model optimized for Project Wannabe and general Japanese creative tasks
Wanabi Gemma4 31B is a 30.7B‑parameter model based on Google Gemma 4 31B‑IT. It is fine‑tuned for Project Wannabe’s structured prompts and includes Japanese CoT data for natural Japanese output. The model supports vision and is available in GGUF format.
Good at
- Vision‑enabled Japanese creative writing
- Structured prompt handling for Project Wannabe
- Instruct‑style general chat in Japanese
Limits
- Large context size requires high VRAM
- No explicit support for non‑Japanese languages
- License is Apache‑2.0
Best for
- Project Wannabe writing workflow
- OpenWebUI or SillyTavern chat interfaces
- Japanese role‑play and creative writing
Who it’s for. Developers and writers who use Project Wannabe or other chat UIs and need a Japanese‑centric creative model with vision support. The model is suitable for those who can run GGUF files with llama.cpp or similar engines.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ2_XXS | 8.7 GB | 41.4 GB | Too big | Get file → | |
| IQ2_XS | 9.5 GB | 42.2 GB | Too big | Get file → | |
| IQ2_M | 10.9 GB | 43.6 GB | Too big | Get file → | |
| Q2_K_S | 11.0 GB | 43.7 GB | Too big | Get file → | |
| Q2_K | 11.9 GB | 44.6 GB | Too big | Get file → | |
| IQ3_XS | 13.1 GB | 45.8 GB | Too big | Get file → | |
| Q3_K_S | 13.8 GB | 46.5 GB | Too big | Get file → | |
| IQ3_S | 13.8 GB | 46.5 GB | Too big | Get file → | |
| IQ3_M | 14.4 GB | 47.1 GB | Too big | Get file → | |
| Q3_K_M | 15.3 GB | 48.0 GB | Too big | Get file → | |
| Q3_K_L | 16.6 GB | 49.3 GB | Too big | Get file → | |
| IQ4_XS | 16.7 GB | 49.4 GB | Too big | Get file → | |
| Q4_0 | 17.7 GB | 50.4 GB | Too big | Get file → | |
| IQ4_NL | 17.7 GB | 50.4 GB | Too big | Get file → | |
| Q4_K_S | 17.8 GB | 50.5 GB | Too big | Get file → | |
| 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 → | |
| Q4_K_M | 18.7 GB | 51.4 GB | Too big | Get file → | |
| Q4_1 | 19.5 GB | 52.2 GB | Too big | Get file → | |
| Q5_0 | 21.3 GB | 54.0 GB | Too big | Get file → | |
| Q5_K_S | 21.3 GB | 54.0 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 → | |
| Q5_K_M | 21.8 GB | 54.6 GB | Too big | Get file → | |
| Q5_1 | 23.1 GB | 55.9 GB | Too big | Get file → | |
| Q6_K | 25.2 GB | 57.9 GB | Too big | Get file → | |
| Q6_K | 25.2 GB | 57.9 GB | Too big | Get file → | |
| Q8_0 | 32.6 GB | 65.3 GB | Too big | Get file → | |
| BF16 | 61.4 GB (2 parts) | 94.1 GB | Too big | Get file → |
For 8 GB VRAM, use IQ2_XXS (8.67 GB file, 41.4 GB memory at 8 k context). For 12 GB VRAM, IQ2_XS (9.53 GB, 42.2 GB). For 24 GB VRAM, Q4_K_M (18.69 GB, 51.4 GB). Higher precision quants such as Q6_K or BF16 require 57.9 GB or 94.1 GB memory respectively and are suitable only for large‑memory setups.
What GPU do you need for Wanabi Gemma4 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: the best quant that fits is Q3_K_M (15.3 GB file, about 48.0 GB in use).
Which Wanabi Gemma4 31B quant should you choose?
Q4_K_M is the usual starting point for Wanabi Gemma4 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 IQ2_XXS at 8.7 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Wanabi Gemma4 31B
How much VRAM does Wanabi Gemma4 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 Wanabi Gemma4 31B on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 41.4 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Wanabi Gemma4 31B GGUF?
It depends on the quant: IQ2_XXS is 8.7 GB, IQ2_XS is 9.5 GB, IQ2_M is 10.9 GB, Q2_K_S is 11.0 GB, Q2_K is 11.9 GB, IQ3_XS is 13.1 GB and 23 more.
What context length does Wanabi Gemma4 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 Wanabi Gemma4 31B GGUF?
The files are on Hugging Face in the kawaimasa/Wanabi-Gemma4-31B-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 Wanabi Gemma4 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 Wanabi Gemma4 31B on 8 GB VRAM?
Use the IQ2_XXS quantized file (8.67 GB). It estimates 41.4 GB memory at 8 k context, which is within 8 GB VRAM limits when using efficient batching.
Does the model support vision tasks?
Yes, the model has detected vision ability and can process image inputs when used with compatible front‑ends.
What license applies to this model?
The model is released under the Apache‑2.0 license, allowing free use and modification.