Countdown on Countdown-4 (test)
0.65RewardA-HPO
Evaluation Results
| Method | Links | |
|---|---|---|
| A-HPOModel=Qwen2.5-7B-Instruct, Training Steps=1000 steps2026.05 | 0.65 | |
| A-HPOModel=Qwen2.5-7B-Instruct, Training Steps=200 steps2026.05 | 0.63 | |
| GRPOModel=Qwen2.5-7B-Instruct, Training Steps=200 steps2026.05 | 0.58 | |
| GRPOModel=Qwen2.5-7B-Instruct, Training Steps=1000 steps2026.05 | 0.58 | |
| A-HPOModel=Llama3.1-3B-Instruct, Training Steps=1000 steps2026.05 | 0.58 | |
| GRPOModel=Llama3.1-3B-Instruct, Training Steps=1000 steps2026.05 | 0.58 | |
| A-HPOModel=Llama3.1-3B-Instruct, Training Steps=200 steps2026.05 | 0.54 | |
| GRPOModel=Llama3.1-3B-Instruct, Training Steps=200 steps2026.05 | 0.4 | |
| A-HPOModel=Qwen2.5-1.5B-Instruct, Training Steps=1000 steps2026.05 | 0.23 | |
| GRPOModel=Qwen2.5-1.5B-Instruct, Training Steps=1000 steps2026.05 | 0.2 | |
| GRPOModel=Qwen2.5-1.5B-Instruct, Training Steps=200 steps2026.05 | 0.19 | |
| A-HPOModel=Qwen2.5-1.5B-Instruct, Training Steps=200 steps2026.05 | 0.17 |