FU

Fugaku-LLM-13B-instruct

FugakuGGUF by mmngaUpdated 2024-05-12
Download on Hugging Face
Parameters13.4B
Context2K
Layers40
Architecturegpt2
Smallest quant6.2 GB
Quick answer

Fugaku-LLM-13B-instruct needs about 16.9 GB of memory at Q4_K_M (a 9.6 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 24 quants, from 6.2 GB (IQ1_S) to 14.3 GB (Q8_0).

Fugaku‑LLM‑13B‑instruct – 13.4B GPT‑2 style instruction model

Fugaku‑LLM‑13B‑instruct is a 13.4B parameter GPT‑2 architecture model converted to GGUF format. It supports up to 2048 context tokens and is available in multiple quantizations. The model is released under a non‑specified license that requires agreement to terms of use.

Good at

  • Instruction following
  • Japanese language support
  • Medium‑size context handling

Limits

  • No vision capability
  • Requires agreement to terms of use
  • Context limited to 2048 tokens

Best for

  • Chatbot development
  • Text summarization
  • Code generation

Who it’s for. Developers needing a Japanese‑focused instruction model with a moderate context window. Suitable for local deployment on machines with sufficient VRAM to load the chosen quantization.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S6.2 GB 8.4 GB
Too big Get file →
IQ1_M6.3 GB 8.5 GB
Too big Get file →
IQ2_XXS6.4 GB 8.6 GB
Too big Get file →
IQ2_XS6.5 GB 8.7 GB
Too big Get file →
IQ2_S6.6 GB 8.8 GB
Too big Get file →
IQ2_M6.7 GB 8.9 GB
Too big Get file →
IQ3_XXS6.9 GB 9.1 GB
Too big Get file →
Q2_K7.1 GB 9.3 GB
Too big Get file →
Q3_K_S7.1 GB 9.3 GB
Too big Get file →
IQ3_S7.1 GB 9.3 GB
Too big Get file →
IQ3_XS7.1 GB 9.3 GB
Too big Get file →
IQ4_XS7.6 GB 9.8 GB
Too big Get file →
Q4_07.7 GB 9.9 GB
Too big Get file →
IQ4_NL7.7 GB 9.9 GB
Too big Get file →
IQ3_M7.7 GB 9.9 GB
Too big Get file →
Q3_K_M8.3 GB 10.5 GB
Too big Get file →
Q3_K_L8.8 GB 11.0 GB
Too big Get file →
Q4_K_S8.9 GB 11.1 GB
Too big Get file →
Q5_09.3 GB 11.5 GB
Too big Get file →
Q4_K_M9.6 GB 11.8 GB
Too big Get file →
Q5_K_S9.9 GB 12.1 GB
Too big Get file →
Q5_K_M11.2 GB 13.4 GB
Too big Get file →
Q6_K13.2 GB 15.4 GB
Too big Get file →
Q8_014.3 GB 16.5 GB
Too big Get file →

For 8 GB VRAM, use IQ1_S (6.2 GB file, ~13.5 GB memory). For 12 GB VRAM, IQ4_XS (7.57 GB file, ~14.9 GB memory). For 16 GB VRAM, Q4_K_S (8.88 GB file, ~16.2 GB memory). Estimates are approximate.

What GPU do you need for Fugaku-LLM-13B-instruct?

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: the best quant that fits is Q3_K_M (8.3 GB file, about 15.6 GB in use).
  • 24 GB VRAM: the best quant that fits is Q8_0 (14.3 GB file, about 21.6 GB in use).

Which Fugaku-LLM-13B-instruct quant should you choose?

Q4_K_M is the usual starting point for Fugaku-LLM-13B-instruct: it balances file size and quality well. Here it is a 9.6 GB file.

The Q8_0 file is 14.3 GB, about 1.5 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 IQ1_S at 6.2 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

Architecturegpt2
Parameters13.4B
Layers40
Trained context2,048 tokens
KV cache (fp16)810 KB per token

Source

Quantized bymmnga
FamilyFugaku
Licenseother
Downloads (HF)3,034
Last updated2024-05-12

Frequently asked questions about Fugaku-LLM-13B-instruct

How much VRAM does Fugaku-LLM-13B-instruct need?

About 16.9 GB at Q4_K_M with an 8K context window. That is the 9.6 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 Fugaku-LLM-13B-instruct on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 13.5 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Fugaku-LLM-13B-instruct GGUF?

It depends on the quant: IQ1_S is 6.2 GB, IQ1_M is 6.3 GB, IQ2_XXS is 6.4 GB, IQ2_XS is 6.5 GB, IQ2_S is 6.6 GB, IQ2_M is 6.7 GB and 18 more.

What context length does Fugaku-LLM-13B-instruct support?

The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 6.8 GB at 8K and 1.7 GB at 2K.

Where can I download Fugaku-LLM-13B-instruct GGUF?

The files are on Hugging Face in the mmnga/Fugaku-LLM-13B-instruct-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 Fugaku-LLM-13B-instruct use?

Hugging Face lists the license as other. 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 Fugaku‑LLM‑13B‑instruct on 8 GB VRAM?

Yes, using the IQ1_S quantization which requires about 13.5 GB of VRAM at 8k context. The file size is 6.2 GB.

Does the model support image inputs?

No, the model has no vision capability.

What license applies to this model?

The model is released under a non‑specified license that requires agreement to the terms of use before use.

Similar models

Browse all →

More from mmnga

All by mmnga →

More Fugaku models

All Fugaku →

Other medium models

All medium models →