plclmezboss/opd-power-diff-student-qwen3-0.6b-teacher-qwen3-4b-alpha1-checkpoint-1500

plclmezboss/opd-power-diff-student-qwen3-0.6b-teacher-qwen3-4b-alpha1-checkpoint-1500

Final checkpoint from an OPD Reward Shaping run.

Run Metadata

• —Student model: Qwen3-0.6B-Base
• —Teacher model: Qwen3-4B
• —Reward mode: sampled_power_diff
• —Power reward alpha: 1
• —Terminal stop target: mask
• —Max completion length: 1024
• —Checkpoint: checkpoint-1500
• —Run tag: 20260518_112528
• —Source checkpoint path: /workspace/anhao/OPD-Reward-Shaping/experiments/power_diff_reward_sampled_power_diff_studentQwen3-0.6B-Base_teacherQwen3-4B/non_thinking/maxlen1024/terminal_mask/alpha1_loss_window_token_mean_reward_none/20260518_112528/checkpoints/checkpoint-1500

Contents

This repository contains the inference-ready model files only: model weights, config, tokenizer files, and chat template. Training-state artifacts such as optimizer, scheduler, RNG, and trainer state are omitted.