Poro 2 8B Long Math Reasoning SFT Preview
This preview specializes Poro 2 Long Base for mathematical reasoning in Finnish and English. It is a supervised training stage, separate from the general Long Instruct model.
Preview
Model design and training
Training uses 2.2 million English mathematical reasoning samples from NVIDIA's Nemotron post-training dataset and 1.4 million Finnish translations. The SFT run uses packed sequences up to 32,768 tokens. The model inherits LongRoPE context extension but is not optimized for general conversation.Llama-Poro-2-8B-Long-Math-Reasoning-SFT-Preview: official model card
Practical assessment
Practical guidance. This intermediate model supports comparisons of supervised and reinforcement learning on bilingual mathematics. Use problems with independently known answers. Keep general conversation outside the evaluation scope unless you intend to measure how specialization changes that behavior.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | Downloadable |
| Model identifiers | Llama-Poro-2-8B-Long-Math-Reasoning-SFT-Preview |
| 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-Math-Reasoning-SFT-Preview: official model card |
| Model size | 8.03 billion parameters.Llama-Poro-2-8B-Long-Math-Reasoning-SFT-Preview: official model card |
| License | Llama 3.1 Community License.Llama-Poro-2-8B-Long-Math-Reasoning-SFT-Preview: official model card |
Related models
- 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 - Poro 2 8B Long Math Reasoning RL Preview
LumiOpen · Poro 2 Long
A mathematical reasoning preview that adds reinforcement learning to Poro 2 Long Math SFT.
Launch date not confirmedPreviewDownloadable