Viking 13B
Viking 13B is a base model pretrained on Finnish, English, Swedish, Danish, Norwegian, Icelandic and code. It uses a LLaMA-like architecture, not a claimed adaptation of Meta's released weights.
Model design and training
40 decoder layers. The card describes a two-trillion-token target and reports 1.3 trillion tokens for the published release. Viking uses rotary position embeddings and FlashAttention. The downloadable base is intended for further tuning. Training checkpoints and later packaging are grouped with the same size-specific release.Viking-13B: official model card
Practical assessment
Practical guidance. The model name rounds its size differently from the architecture table. Use the reported parameter count when planning memory. Preserve the repository revision when comparing it with Viking 7B because the cards report different training progress.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Viking-13B |
| Context window | 4,096 tokens Sequence length reported by this release's model card; this is not a guaranteed useful retrieval length. Viking-13B: official model card |
| Model size | 14 billion parameters in the model table.Viking-13B: official model card |
| License | Apache 2.0Viking-13B: official model card |
Related models
- Viking 33B
LumiOpen · Viking
A 33B base model pretrained on Finnish, English, Swedish, Danish, Norwegian, Icelandic and code.
Launch date not confirmedDownloadable A 7B base model pretrained on Finnish, English, Swedish, Danish, Norwegian, Icelandic and code.
Launch date not confirmedDownloadable