| license: apache-2.0 | |
| base_model: allenai/Olmo-3-1025-7B | |
| tags: [grpo, rlvr, reasoning-registers, lora] | |
| # arm4 — register forced, no truncation penalty | |
| GRPO (TRL) LoRA adapters (r=64, alpha=128) on `allenai/Olmo-3-1025-7B`, RL-Zero prompt. | |
| One cell of the 2x2 factorial {register forced} x {truncation penalty}. Folders `step100/ ... step300/`. | |
| Paired eval rollouts: `ReasoningRegisters/arm4-eval-rollouts` (splits `step{k}_{benchmark}`). | |