Base model: Qwen/Qwen2.5-0.5B-Instruct. Environment: 4x4 FrozenLake, p=0.8 safe-tile probability, no slipping, grid and coordinates. PPO/GAE, no rollout filtering. at most 2 actions per turn, response limit 128 tokens. Actor LR 1e-6, critic LR 1e-5, entropy coefficient 0.001. See results/config.yaml for all settings. Validation: fixed 64 seeds (123 through 186), greedy decoding, before training and every 10 steps. Training maps start at seed 10000. One training run only. W&B: https://wandb.ai/xiaoyiwang2027-northwestern-university/ragenpilotfrozenlake/runs/8baf50af33eb Best saved TRAINED checkpoint by validation success, then reward: step0030 Each completed HF checkpoint is under…
Publisher
Xiaoyou77527
xiaoyou77527
Models in Library1
Datasets in Library0
Models on Hugging Face—
Followers—
Compare the published evidence
This profile connects the 1 model and dataset record shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Reinforcement learning.
Reported licenses: apache-2.0. Open a record to review its specific license and conditions.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.