Poro 2 8B Long Math Reasoning RL Preview
This preview adds reinforcement learning to the Poro 2 Long mathematical SFT checkpoint. It trains mathematical reasoning in Finnish and English through separate language-specific environments.
Preview
Model design and training
The release follows mathematical SFT and reports 900 reinforcement-learning steps. It uses separate Finnish and English problem environments to retain bilingual reasoning. The card still limits its intended scope to mathematics and labels it a preview. Its lineage runs through Poro 2 Long to Llama 3.1 8B.Llama-Poro-2-8B-Long-Math-Reasoning-RL-Preview: official model card
Practical assessment
Practical guidance. Assess mathematical correctness separately from the language and length of the explanation. Compare the RL checkpoint with SFT on the same unseen Finnish problems. A persuasive chain of reasoning can still contain an incorrect intermediate result.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | Downloadable |
| Model identifiers | Llama-Poro-2-8B-Long-Math-Reasoning-RL-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-RL-Preview: official model card |
| Model size | 8.03 billion parameters.Llama-Poro-2-8B-Long-Math-Reasoning-RL-Preview: official model card |
| License | Llama 3.1 Community License.Llama-Poro-2-8B-Long-Math-Reasoning-RL-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 SFT Preview
LumiOpen · Poro 2 Long
A supervised Poro 2 Long preview trained for mathematical reasoning in Finnish and English.
Launch date not confirmedPreviewDownloadable