VI

Viking-33B

VikingGGUF by mradermacherUpdated 2024-08-02
Download on Hugging Face
Parameters33B
Context4K
Layers56
Architecturellama
Smallest quant7.5 GB
Quick answer

Viking-33B needs about 22.4 GB of memory at Q4_K_M (a 20.0 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 21 quants, from 7.5 GB (IQ1_S) to 27.2 GB (Q6_K).

Viking‑33B, a 33B‑parameter decoder‑only transformer for Finnish, English, and code.

Viking‑33B is a 33B‑parameter LLaMA‑style decoder trained on Finnish, English, Scandinavian languages, and code. It is released under Apache‑2.0 and available in multiple GGUF quantizations. The model is a base model and requires fine‑tuning for most applications.

Good at

  • Code generation and debugging
  • Multilingual text in Finnish and Scandinavian languages
  • Large‑context generation up to 4096 tokens

Limits

  • Requires 8‑k context for memory estimates
  • No vision or multimodal support
  • Large VRAM needed for higher‑quality quants

Best for

  • Developers needing Finnish code assistance
  • Researchers working on Scandinavian NLP
  • Users with 8‑k token context requirements

Who it’s for. The model is suited for developers, researchers, and hobbyists who need a large, multilingual language model with strong coding capabilities. It is ideal for those who can fine‑tune the base model for specific tasks.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S7.5 GB 9.0 GB
Too big Get file →
IQ1_M8.2 GB 9.6 GB
Too big Get file →
IQ2_XXS9.2 GB 10.7 GB
Too big Get file →
IQ2_XS10.1 GB 11.6 GB
Too big Get file →
IQ2_S10.7 GB 12.2 GB
Too big Get file →
IQ2_M11.6 GB 13.0 GB
Too big Get file →
Q2_K12.6 GB 14.0 GB
Too big Get file →
IQ3_XXS13.0 GB 14.5 GB
Too big Get file →
IQ3_XS13.9 GB 15.4 GB
Too big Get file →
Q3_K_S14.6 GB 16.0 GB
Too big Get file →
IQ3_S14.7 GB 16.1 GB
Too big Get file →
IQ3_M15.2 GB 16.6 GB
Too big Get file →
Q3_K_M16.2 GB 17.6 GB
Too big Get file →
Q3_K_L17.6 GB 19.0 GB
Too big Get file →
IQ4_XS17.9 GB 19.4 GB
Too big Get file →
Q4_018.9 GB 20.4 GB
Too big Get file →
Q4_K_S19.0 GB 20.5 GB
Too big Get file →
Q4_K_M20.0 GB 21.4 GB
Too big Get file →
Q5_K_S22.9 GB 24.3 GB
Too big Get file →
Q5_K_M23.5 GB 24.9 GB
Too big Get file →
Q6_K27.2 GB 28.6 GB
Too big Get file →

For 8‑GB VRAM, use IQ1_S (7.53 GB) or IQ1_M (8.16 GB). With 12‑GB VRAM, IQ2_S (10.75 GB) or Q2_K (12.56 GB) are suitable. For 24‑GB VRAM, Q4_K_M (19.99 GB) or Q5_K_M (23.47 GB) provide higher quality. Memory estimates are approximate.

What GPU do you need for Viking-33B?

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: the best quant that fits is IQ2_XXS (9.2 GB file, about 11.6 GB in use).
  • 16 GB VRAM: the best quant that fits is IQ3_XXS (13.0 GB file, about 15.4 GB in use).
  • 24 GB VRAM: the best quant that fits is Q4_K_M (20.0 GB file, about 22.4 GB in use).
  • 32 GB VRAM: the best quant that fits is Q6_K (27.2 GB file, about 29.6 GB in use).

Which Viking-33B quant should you choose?

Q4_K_M is the usual starting point for Viking-33B: it balances file size and quality well. Here it is a 20.0 GB file.

If memory is tight, the smallest file is IQ1_S at 7.5 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
Parameters33B
Layers56
Trained context4,096 tokens
KV cache (fp16)224 KB per token

Source

Quantized bymradermacher
FamilyViking
Licenseapache-2.0
Downloads (HF)4,481
Last updated2024-08-02

Frequently asked questions about Viking-33B

How much VRAM does Viking-33B need?

About 22.4 GB at Q4_K_M with an 8K context window. That is the 20.0 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 Viking-33B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 9.9 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Viking-33B GGUF?

It depends on the quant: IQ1_S is 7.5 GB, IQ1_M is 8.2 GB, IQ2_XXS is 9.2 GB, IQ2_XS is 10.1 GB, IQ2_S is 10.7 GB, IQ2_M is 11.6 GB and 15 more.

What context length does Viking-33B support?

The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 1.9 GB at 8K and 0.9 GB at 4K.

Where can I download Viking-33B GGUF?

The files are on Hugging Face in the mradermacher/Viking-33B-i1-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 Viking-33B 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 Viking‑33B on 8 GB VRAM?

Yes, the IQ1_S quant uses about 9.9 GB of VRAM at 8‑k context, which fits in an 8‑GB GPU with some headroom.

Does Viking‑33B support vision or multimodal inputs?

No, the model is a decoder‑only transformer and does not include vision or multimodal capabilities.

What license applies to Viking‑33B?

Viking‑33B is released under the Apache‑2.0 license, allowing free use and modification.

Other GGUFs of Viking-33B

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Viking models

All Viking →

Other medium models

All medium models →