Poro 2 8B Long Instruct
Poro 2 Long Instruct adds Finnish and English instruction following to the context-extended Poro base. It is trained separately from the mathematical reasoning previews.
Model design and training
Instruction tuning uses packed 32,768-token sequences. The team then merges 80% of the SFT checkpoint with 20% of the Long Base checkpoint to recover long-context behavior. This release has no DPO stage. Its documented context reaches 128K tokens; the underlying model descends from Llama 3.1 8B.Llama-Poro-2-8B-Long-Instruct: official model card
Practical assessment
Practical guidance. Use this release for long Finnish documents that require a response to instructions. Check whether the output distinguishes facts from separate files. The merge recipe addresses a training trade-off, but it does not guarantee correct attribution across a large collection.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Llama-Poro-2-8B-Long-Instruct |
| Context window | 131,072 tokens Sequence length reported by this release's model card; this is not a guaranteed useful retrieval length. Llama-Poro-2-8B-Long-Instruct: official model card |
| License | Llama 3.3 Community License.Llama-Poro-2-8B-Long-Instruct: official model card |
Related models
- Poro 2 8B Instruct
LumiOpen · Poro 2
An 8B Finnish and English conversation model trained with supervised instructions and Direct Preference Optimization.
Launch date not confirmedDownloadable - Poro 2 8B Long Base
LumiOpen · Poro 2 Long
A Finnish and English Poro 2 8B base model extended from 8K to 128K context.
Launch date not confirmedDownloadable