KA

Kaiju A 57B

KaijuGGUF by mradermacherUpdated 2024-11-22
Download on Hugging Face
Parameters57.3B
Context195K
Layers101
Architecturellama
Smallest quant21.2 GB
Quick answer

Kaiju A 57B needs about 38.2 GB of memory at Q4_K_M (a 34.3 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 10 quants, from 21.2 GB (Q2_K) to 47.0 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 13.6 GB.

Kaiju A 57B is a 57.3B-parameter Kaiju model in GGUF format, quantized by mradermacher. Its trained context length is 195K tokens. The repository has 10 quant files, from 21.2 GB (Q2_K) to 47.0 GB (Q6_K).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K21.2 GB 25.1 GB
Too big Get file →
Q3_K_S24.8 GB 28.7 GB
Too big Get file →
Q3_K_M27.7 GB 31.5 GB
Too big Get file →
Q3_K_L30.1 GB 34.0 GB
Too big Get file →
IQ4_XS31.0 GB 34.8 GB
Too big Get file →
Q4_K_S32.6 GB 36.4 GB
Too big Get file →
Q4_K_M34.3 GB 38.2 GB
Too big Get file →
Q5_K_S39.4 GB 43.3 GB
Too big Get file →
Q5_K_M40.5 GB 44.3 GB
Too big Get file →
Q6_K47.0 GB 50.9 GB
Too big Get file →

What GPU do you need for Kaiju A 57B?

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: the best quant that fits is Q3_K_M (27.7 GB file, about 31.5 GB in use).
  • 48 GB VRAM: the best quant that fits is Q5_K_M (40.5 GB file, about 44.3 GB in use).

Which Kaiju A 57B quant should you choose?

Q4_K_M is the usual starting point for Kaiju A 57B: it balances file size and quality well. Here it is a 34.3 GB file.

If memory is tight, the smallest file is Q2_K at 21.2 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

Architecturellama
Parameters57.3B
Layers101
Trained context200,000 tokens
KV cache (fp16)404 KB per token

Source

Quantized bymradermacher
FamilyKaiju
Licenseother
Downloads (HF)39
Last updated2024-11-22

Frequently asked questions about Kaiju A 57B

How much VRAM does Kaiju A 57B need?

About 38.2 GB at Q4_K_M with an 8K context window. That is the 34.3 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 Kaiju A 57B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 25.1 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Kaiju A 57B GGUF?

It depends on the quant: Q2_K is 21.2 GB, Q3_K_S is 24.8 GB, Q3_K_M is 27.7 GB, Q3_K_L is 30.1 GB, IQ4_XS is 31.0 GB, Q4_K_S is 32.6 GB and 4 more.

What context length does Kaiju A 57B support?

The GGUF header lists a trained context length of 200,000 tokens (195K). The KV cache grows with the context you use: about 3.4 GB at 8K and 13.6 GB at 32K.

Where can I download Kaiju A 57B GGUF?

The files are on Hugging Face in the mradermacher/Kaiju-A-57B-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 Kaiju A 57B 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.

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Kaiju models

All Kaiju →

Other large models

All large models →