| { |
| "best_global_step": 40, |
| "best_metric": -0.015412718057632446, |
| "best_model_checkpoint": "outputs/deepseek-ai/deepseek-coder-7b-instruct-v1.5/countdown/checkpoint-40", |
| "epoch": 3.0, |
| "eval_steps": 5, |
| "global_step": 75, |
| "is_hyper_param_search": false, |
| "is_local_process_zero": true, |
| "is_world_process_zero": true, |
| "log_history": [ |
| { |
| "epoch": 0.2, |
| "grad_norm": 0.04151817411184311, |
| "learning_rate": 9.466666666666666e-07, |
| "loss": -0.0153, |
| "step": 5 |
| }, |
| { |
| "epoch": 0.2, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 539.5, |
| "eval_completions/max_terminated_length": 539.5, |
| "eval_completions/mean_length": 323.3875, |
| "eval_completions/mean_terminated_length": 323.3875, |
| "eval_completions/min_length": 155.25, |
| "eval_completions/min_terminated_length": 155.25, |
| "eval_kl": 4.560351371765137e-05, |
| "eval_loss": 0.013234468176960945, |
| "eval_num_tokens": 33700.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0125, |
| "eval_rewards/equation_reward_func/std": 0.025, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2325.8168, |
| "eval_samples_per_second": 0.009, |
| "eval_steps_per_second": 0.002, |
| "step": 5 |
| }, |
| { |
| "epoch": 0.4, |
| "grad_norm": 0.0, |
| "learning_rate": 8.799999999999999e-07, |
| "loss": -0.0057, |
| "step": 10 |
| }, |
| { |
| "epoch": 0.4, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0375, |
| "eval_completions/max_length": 552.95, |
| "eval_completions/max_terminated_length": 449.75, |
| "eval_completions/mean_length": 327.275, |
| "eval_completions/mean_terminated_length": 297.87500152587893, |
| "eval_completions/min_length": 172.85, |
| "eval_completions/min_terminated_length": 172.85, |
| "eval_kl": 4.1404366493225096e-05, |
| "eval_loss": 0.016905201599001884, |
| "eval_num_tokens": 71178.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0125, |
| "eval_rewards/equation_reward_func/std": 0.025, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2390.7886, |
| "eval_samples_per_second": 0.008, |
| "eval_steps_per_second": 0.002, |
| "step": 10 |
| }, |
| { |
| "epoch": 0.6, |
| "grad_norm": 5.364913135963434e-07, |
| "learning_rate": 8.133333333333333e-07, |
| "loss": 0.0, |
| "step": 15 |
| }, |
| { |
| "epoch": 0.6, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 465.1, |
| "eval_completions/max_terminated_length": 465.1, |
| "eval_completions/mean_length": 311.7, |
| "eval_completions/mean_terminated_length": 311.7, |
| "eval_completions/min_length": 191.85, |
| "eval_completions/min_terminated_length": 191.85, |
| "eval_kl": 4.3448805809021e-05, |
| "eval_loss": 0.0106130326166749, |
| "eval_num_tokens": 108429.0, |
| "eval_reward": 0.025, |
| "eval_reward_std": 0.05, |
| "eval_rewards/equation_reward_func/mean": 0.025, |
| "eval_rewards/equation_reward_func/std": 0.05, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 1990.1115, |
| "eval_samples_per_second": 0.01, |
| "eval_steps_per_second": 0.003, |
| "step": 15 |
| }, |
| { |
| "epoch": 0.8, |
| "grad_norm": 0.0, |
| "learning_rate": 7.466666666666667e-07, |
| "loss": 0.0202, |
| "step": 20 |
| }, |
| { |
| "epoch": 0.8, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 440.2, |
| "eval_completions/max_terminated_length": 440.2, |
| "eval_completions/mean_length": 291.7375, |
| "eval_completions/mean_terminated_length": 291.7375, |
| "eval_completions/min_length": 169.5, |
| "eval_completions/min_terminated_length": 169.5, |
| "eval_kl": 4.6834349632263184e-05, |
| "eval_loss": 0.005631707608699799, |
| "eval_num_tokens": 144881.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0, |
| "eval_rewards/equation_reward_func/std": 0.0, |
| "eval_rewards/format_reward_func/mean": 0.0125, |
| "eval_rewards/format_reward_func/std": 0.025, |
| "eval_runtime": 1879.5768, |
| "eval_samples_per_second": 0.011, |
| "eval_steps_per_second": 0.003, |
| "step": 20 |
| }, |
| { |
| "epoch": 1.0, |
| "grad_norm": 7.261731411745131e-07, |
| "learning_rate": 6.800000000000001e-07, |
| "loss": 0.0, |
| "step": 25 |
| }, |
| { |
| "epoch": 1.0, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 506.95, |
| "eval_completions/max_terminated_length": 506.95, |
| "eval_completions/mean_length": 345.9, |
| "eval_completions/mean_terminated_length": 345.9, |
| "eval_completions/min_length": 199.95, |
| "eval_completions/min_terminated_length": 199.95, |
| "eval_kl": 4.442334175109863e-05, |
| "eval_loss": 0.015046780928969383, |
| "eval_num_tokens": 178858.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0, |
| "eval_rewards/equation_reward_func/std": 0.0, |
| "eval_rewards/format_reward_func/mean": 0.0125, |
| "eval_rewards/format_reward_func/std": 0.025, |
| "eval_runtime": 2176.219, |
| "eval_samples_per_second": 0.009, |
| "eval_steps_per_second": 0.002, |
| "step": 25 |
| }, |
| { |
| "epoch": 1.2, |
| "grad_norm": 0.0, |
| "learning_rate": 6.133333333333332e-07, |
| "loss": 0.0181, |
| "step": 30 |
| }, |
| { |
| "epoch": 1.2, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.05, |
| "eval_completions/max_length": 568.0, |
| "eval_completions/max_terminated_length": 447.0, |
| "eval_completions/mean_length": 329.3125, |
| "eval_completions/mean_terminated_length": 291.2708351135254, |
| "eval_completions/min_length": 161.75, |
| "eval_completions/min_terminated_length": 161.75, |
| "eval_kl": 4.176795482635498e-05, |
| "eval_loss": 1.6701305867172778e-06, |
| "eval_num_tokens": 214760.0, |
| "eval_reward": 0.0, |
| "eval_reward_std": 0.0, |
| "eval_rewards/equation_reward_func/mean": 0.0, |
| "eval_rewards/equation_reward_func/std": 0.0, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2459.6886, |
| "eval_samples_per_second": 0.008, |
| "eval_steps_per_second": 0.002, |
| "step": 30 |
| }, |
| { |
| "epoch": 1.4, |
| "grad_norm": 3.169759281718143e-07, |
| "learning_rate": 5.466666666666666e-07, |
| "loss": 0.0, |
| "step": 35 |
| }, |
| { |
| "epoch": 1.4, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 487.6, |
| "eval_completions/max_terminated_length": 487.6, |
| "eval_completions/mean_length": 318.45, |
| "eval_completions/mean_terminated_length": 318.45, |
| "eval_completions/min_length": 188.2, |
| "eval_completions/min_terminated_length": 188.2, |
| "eval_kl": 4.681050777435303e-05, |
| "eval_loss": 0.012316806241869926, |
| "eval_num_tokens": 251352.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0125, |
| "eval_rewards/equation_reward_func/std": 0.025, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2088.3713, |
| "eval_samples_per_second": 0.01, |
| "eval_steps_per_second": 0.002, |
| "step": 35 |
| }, |
| { |
| "epoch": 1.6, |
| "grad_norm": 6.91386446760589e-07, |
| "learning_rate": 4.8e-07, |
| "loss": 0.0018, |
| "step": 40 |
| }, |
| { |
| "epoch": 1.6, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 406.7, |
| "eval_completions/max_terminated_length": 406.7, |
| "eval_completions/mean_length": 278.6125, |
| "eval_completions/mean_terminated_length": 278.6125, |
| "eval_completions/min_length": 165.85, |
| "eval_completions/min_terminated_length": 165.85, |
| "eval_kl": 4.2584538459777835e-05, |
| "eval_loss": -0.015412718057632446, |
| "eval_num_tokens": 288126.0, |
| "eval_reward": 0.025, |
| "eval_reward_std": 0.05, |
| "eval_rewards/equation_reward_func/mean": 0.0125, |
| "eval_rewards/equation_reward_func/std": 0.025, |
| "eval_rewards/format_reward_func/mean": 0.0125, |
| "eval_rewards/format_reward_func/std": 0.025, |
| "eval_runtime": 1731.663, |
| "eval_samples_per_second": 0.012, |
| "eval_steps_per_second": 0.003, |
| "step": 40 |
| }, |
| { |
| "epoch": 1.8, |
| "grad_norm": 4.906948447569448e-07, |
| "learning_rate": 4.1333333333333333e-07, |
| "loss": -0.0102, |
| "step": 45 |
| }, |
| { |
| "epoch": 1.8, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 422.2, |
| "eval_completions/max_terminated_length": 422.2, |
| "eval_completions/mean_length": 279.8625, |
| "eval_completions/mean_terminated_length": 279.8625, |
| "eval_completions/min_length": 168.8, |
| "eval_completions/min_terminated_length": 168.8, |
| "eval_kl": 4.140138626098633e-05, |
| "eval_loss": -0.009393930435180664, |
| "eval_num_tokens": 326948.0, |
| "eval_reward": 0.0375, |
| "eval_reward_std": 0.075, |
| "eval_rewards/equation_reward_func/mean": 0.0125, |
| "eval_rewards/equation_reward_func/std": 0.025, |
| "eval_rewards/format_reward_func/mean": 0.025, |
| "eval_rewards/format_reward_func/std": 0.05, |
| "eval_runtime": 1799.6706, |
| "eval_samples_per_second": 0.011, |
| "eval_steps_per_second": 0.003, |
| "step": 45 |
| }, |
| { |
| "epoch": 2.0, |
| "grad_norm": 0.0, |
| "learning_rate": 3.4666666666666665e-07, |
| "loss": 0.0, |
| "step": 50 |
| }, |
| { |
| "epoch": 2.0, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.025, |
| "eval_completions/max_length": 576.85, |
| "eval_completions/max_terminated_length": 545.95, |
| "eval_completions/mean_length": 349.0125, |
| "eval_completions/mean_terminated_length": 335.475, |
| "eval_completions/min_length": 183.8, |
| "eval_completions/min_terminated_length": 183.8, |
| "eval_kl": 4.1183829307556155e-05, |
| "eval_loss": 1.6466992747155018e-06, |
| "eval_num_tokens": 364182.0, |
| "eval_reward": 0.0, |
| "eval_reward_std": 0.0, |
| "eval_rewards/equation_reward_func/mean": 0.0, |
| "eval_rewards/equation_reward_func/std": 0.0, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2492.2782, |
| "eval_samples_per_second": 0.008, |
| "eval_steps_per_second": 0.002, |
| "step": 50 |
| }, |
| { |
| "epoch": 2.2, |
| "grad_norm": 5.559809324040543e-07, |
| "learning_rate": 2.8e-07, |
| "loss": 0.011, |
| "step": 55 |
| }, |
| { |
| "epoch": 2.2, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0125, |
| "eval_completions/max_length": 527.5, |
| "eval_completions/max_terminated_length": 490.4, |
| "eval_completions/mean_length": 314.025, |
| "eval_completions/mean_terminated_length": 303.941667175293, |
| "eval_completions/min_length": 168.4, |
| "eval_completions/min_terminated_length": 168.4, |
| "eval_kl": 3.9377808570861815e-05, |
| "eval_loss": 0.013314701616764069, |
| "eval_num_tokens": 402008.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0, |
| "eval_rewards/equation_reward_func/std": 0.0, |
| "eval_rewards/format_reward_func/mean": 0.0125, |
| "eval_rewards/format_reward_func/std": 0.025, |
| "eval_runtime": 2279.1265, |
| "eval_samples_per_second": 0.009, |
| "eval_steps_per_second": 0.002, |
| "step": 55 |
| }, |
| { |
| "epoch": 2.4, |
| "grad_norm": 0.0, |
| "learning_rate": 2.1333333333333334e-07, |
| "loss": 0.0224, |
| "step": 60 |
| }, |
| { |
| "epoch": 2.4, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0, |
| "eval_completions/max_length": 485.2, |
| "eval_completions/max_terminated_length": 485.2, |
| "eval_completions/mean_length": 308.925, |
| "eval_completions/mean_terminated_length": 308.925, |
| "eval_completions/min_length": 184.3, |
| "eval_completions/min_terminated_length": 184.3, |
| "eval_kl": 4.113614559173584e-05, |
| "eval_loss": 0.0009593071299605072, |
| "eval_num_tokens": 436085.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0125, |
| "eval_rewards/equation_reward_func/std": 0.025, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2080.7394, |
| "eval_samples_per_second": 0.01, |
| "eval_steps_per_second": 0.002, |
| "step": 60 |
| }, |
| { |
| "epoch": 2.6, |
| "grad_norm": 3.0584305932279676e-07, |
| "learning_rate": 1.4666666666666666e-07, |
| "loss": 0.0, |
| "step": 65 |
| }, |
| { |
| "epoch": 2.6, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0125, |
| "eval_completions/max_length": 485.85, |
| "eval_completions/max_terminated_length": 454.4, |
| "eval_completions/mean_length": 308.85, |
| "eval_completions/mean_terminated_length": 300.3208343505859, |
| "eval_completions/min_length": 176.9, |
| "eval_completions/min_terminated_length": 176.9, |
| "eval_kl": 4.4587254524230954e-05, |
| "eval_loss": 1.782888261914195e-06, |
| "eval_num_tokens": 473289.0, |
| "eval_reward": 0.0, |
| "eval_reward_std": 0.0, |
| "eval_rewards/equation_reward_func/mean": 0.0, |
| "eval_rewards/equation_reward_func/std": 0.0, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2084.9485, |
| "eval_samples_per_second": 0.01, |
| "eval_steps_per_second": 0.002, |
| "step": 65 |
| }, |
| { |
| "epoch": 2.8, |
| "grad_norm": 0.033825669437646866, |
| "learning_rate": 8e-08, |
| "loss": 0.0202, |
| "step": 70 |
| }, |
| { |
| "epoch": 2.8, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0125, |
| "eval_completions/max_length": 440.1, |
| "eval_completions/max_terminated_length": 401.9, |
| "eval_completions/mean_length": 295.2, |
| "eval_completions/mean_terminated_length": 285.4583335876465, |
| "eval_completions/min_length": 185.7, |
| "eval_completions/min_terminated_length": 185.7, |
| "eval_kl": 4.4012069702148436e-05, |
| "eval_loss": 0.003522181883454323, |
| "eval_num_tokens": 508928.0, |
| "eval_reward": 0.0125, |
| "eval_reward_std": 0.025, |
| "eval_rewards/equation_reward_func/mean": 0.0125, |
| "eval_rewards/equation_reward_func/std": 0.025, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 1885.993, |
| "eval_samples_per_second": 0.011, |
| "eval_steps_per_second": 0.003, |
| "step": 70 |
| }, |
| { |
| "epoch": 3.0, |
| "grad_norm": 2.9322006867005257e-07, |
| "learning_rate": 1.3333333333333334e-08, |
| "loss": 0.013, |
| "step": 75 |
| }, |
| { |
| "epoch": 3.0, |
| "eval_clip_ratio/high_max": 0.0, |
| "eval_clip_ratio/high_mean": 0.0, |
| "eval_clip_ratio/low_mean": 0.0, |
| "eval_clip_ratio/low_min": 0.0, |
| "eval_clip_ratio/region_mean": 0.0, |
| "eval_completions/clipped_ratio": 0.0125, |
| "eval_completions/max_length": 540.0, |
| "eval_completions/max_terminated_length": 516.9, |
| "eval_completions/mean_length": 322.5, |
| "eval_completions/mean_terminated_length": 315.29583435058595, |
| "eval_completions/min_length": 171.8, |
| "eval_completions/min_terminated_length": 171.8, |
| "eval_kl": 4.659593105316162e-05, |
| "eval_loss": 1.8635412288858788e-06, |
| "eval_num_tokens": 544500.0, |
| "eval_reward": 0.0, |
| "eval_reward_std": 0.0, |
| "eval_rewards/equation_reward_func/mean": 0.0, |
| "eval_rewards/equation_reward_func/std": 0.0, |
| "eval_rewards/format_reward_func/mean": 0.0, |
| "eval_rewards/format_reward_func/std": 0.0, |
| "eval_runtime": 2330.8276, |
| "eval_samples_per_second": 0.009, |
| "eval_steps_per_second": 0.002, |
| "step": 75 |
| } |
| ], |
| "logging_steps": 5, |
| "max_steps": 75, |
| "num_input_tokens_seen": 544500, |
| "num_train_epochs": 3, |
| "save_steps": 5, |
| "stateful_callbacks": { |
| "TrainerControl": { |
| "args": { |
| "should_epoch_stop": false, |
| "should_evaluate": false, |
| "should_log": false, |
| "should_save": true, |
| "should_training_stop": true |
| }, |
| "attributes": {} |
| } |
| }, |
| "total_flos": 0.0, |
| "train_batch_size": 4, |
| "trial_name": null, |
| "trial_params": null |
| } |
|
|
