Rehangmu1710/hmne-qwen3-8b-grpo-h200
0
HMNE GRPO Adapter
- Warm-start model/adapter:
Rehangmu1710/hmne-qwen3-8b-sft - Curriculum:
easy,mixed,hard,deceptive - Stage steps:
25,30,30,20 - Prompts per stage:
128 - Group size:
8 - KL beta:
0.04 - Learning rate:
5e-06 - Seed:
3407
