Tülu 3.1 8B
Tülu 3.1 8B is Ai2's instruction adaptation of Meta's Llama 3.1.
What the release provides
This update changes the final reinforcement-learning stage from PPO to GRPO and tunes its hyperparameters. Earlier stages come from the Tülu 3 8B line. The Llama 3.1 Community License applies.Tülu 3.1 8B official model card
How to use it well
Practical guidance. Keep Meta's base-model terms with the Ai2 release. Compare the final checkpoint with its training predecessors only when the experiment needs that distinction.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | allenai/Llama-3.1-Tulu-3.1-8B |