Llama-3-8B-shisa-2x8B needs about 9.9 GB of memory at Q4_K_M (a 8.3 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 14 quants, from 5.2 GB (Q2_K) to 14.5 GB (Q8_0).
Mixture of Experts Llama‑3 model with enhanced Japanese support
Llama-3-8B-shisa-2x8B is a Mixture of Experts language model created with MergeKit. It builds on Meta‑Llama‑3‑8B‑Instruct and shisa‑ai/shisa‑v1‑llama3‑8b to improve Japanese language capabilities while retaining the base model’s abilities. The model is available in multiple GGUF quantizations.
Good at
- Japanese language generation
- Large context handling up to 8192 tokens
- Efficient inference with low‑bit quantizations
Limits
- Requires 8k context for memory estimates
- No vision support
- License is llama3
Best for
- Japanese chatbots
- Content creation in Japanese
- Fine‑tuning Japanese language models
Who it’s for. Developers and researchers needing a Japanese‑capable Llama‑3 model with Mixture of Experts architecture. Suitable for those who can run GGUF models on GPU or CPU with sufficient memory.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 5.2 GB | 6.8 GB | Fits | Get file → | |
| IQ3_XS | 5.8 GB | 7.4 GB | Tight | Get file → | |
| Q3_K_S | 6.1 GB | 7.7 GB | Tight | Get file → | |
| IQ3_S | 6.1 GB | 7.7 GB | Tight | Get file → | |
| IQ3_M | 6.2 GB | 7.8 GB | Tight | Get file → | |
| Q3_K_M | 6.7 GB | 8.3 GB | Too big | Get file → | |
| Q3_K_L | 7.2 GB | 8.8 GB | Too big | Get file → | |
| IQ4_XS | 7.5 GB | 9.1 GB | Too big | Get file → | |
| Q4_K_S | 7.9 GB | 9.5 GB | Too big | Get file → | |
| Q4_K_M | 8.3 GB | 9.9 GB | Too big | Get file → | |
| Q5_K_S | 9.5 GB | 11.0 GB | Too big | Get file → | |
| Q5_K_M | 9.7 GB | 11.3 GB | Too big | Get file → | |
| Q6_K | 11.2 GB | 12.8 GB | Too big | Get file → | |
| Q8_0 | 14.5 GB | 16.1 GB | Too big | Get file → |
For 12 GB VRAM, Q4_K_M (8.33 GB file, ~9.9 GB memory at 8k context) is recommended. With 8 GB VRAM, Q3_K_S (6.09 GB file, ~7.7 GB memory) is a good balance. For CPU‑only use, Q2_K (5.22 GB file, ~6.8 GB memory) provides the lowest memory footprint.
What GPU do you need for Llama-3-8B-shisa-2x8B?
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: the best quant that fits is IQ3_M (6.2 GB file, about 7.8 GB in use).
- 12 GB VRAM: the best quant that fits is Q5_K_M (9.7 GB file, about 11.3 GB in use).
- 16 GB VRAM: the best quant that fits is Q6_K (11.2 GB file, about 12.8 GB in use).
- 24 GB VRAM: the best quant that fits is Q8_0 (14.5 GB file, about 16.1 GB in use).
Which Llama-3-8B-shisa-2x8B quant should you choose?
Q4_K_M is the usual starting point for Llama-3-8B-shisa-2x8B: it balances file size and quality well. Here it is a 8.3 GB file.
The Q8_0 file is 14.5 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 Q2_K at 5.2 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 Llama-3-8B-shisa-2x8B
How much VRAM does Llama-3-8B-shisa-2x8B need?
About 9.9 GB at Q4_K_M with an 8K context window. That is the 8.3 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 Llama-3-8B-shisa-2x8B on an 8 GB GPU?
Yes, with IQ3_M: the 6.2 GB file needs about 7.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Llama-3-8B-shisa-2x8B GGUF?
It depends on the quant: Q2_K is 5.2 GB, IQ3_XS is 5.8 GB, Q3_K_S is 6.1 GB, IQ3_S is 6.1 GB, IQ3_M is 6.2 GB, Q3_K_M is 6.7 GB and 8 more.
What context length does Llama-3-8B-shisa-2x8B support?
The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 1.1 GB at 8K and 1.1 GB at 8K.
Where can I download Llama-3-8B-shisa-2x8B GGUF?
The files are on Hugging Face in the mradermacher/Llama-3-8B-shisa-2x8B-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 Llama-3-8B-shisa-2x8B use?
Hugging Face lists the license as llama3. 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 Q4_K_M on a 12 GB GPU?
Yes, Q4_K_M (8.33 GB file) can be fully loaded on a 12 GB GPU such as the RTX 3060.
What is the memory requirement for Q3_K_S?
Q3_K_S requires approximately 7.7 GB of VRAM at 8k context, based on the provided estimates.
Does this model support vision tasks?
No, the model does not include vision capabilities.