dieg0code commited on
Commit
04005d5
·
verified ·
1 Parent(s): 2988cee

Checkpoint iteration 108

Browse files
runs/policy_spatial_v9/metadata_iter_108.json ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "iteration": 108,
3
+ "timestamp": "2026-05-11T08:36:01.458637+00:00",
4
+ "buffer_size": 40365,
5
+ "config": {
6
+ "iterations": 180,
7
+ "episodes_per_iter": 20,
8
+ "mcts_sims": 160,
9
+ "c_puct": 1.5,
10
+ "temp_threshold": 28,
11
+ "add_noise": true,
12
+ "seed": 42,
13
+ "verbose_logs": false,
14
+ "episode_log_every": 25,
15
+ "epochs": 1,
16
+ "batch_size": 224,
17
+ "learning_rate": 0.0003,
18
+ "weight_decay": 0.0001,
19
+ "value_loss_coeff": 0.5,
20
+ "reward_shaping_enabled": true,
21
+ "reward_shaping_scale": 0.1,
22
+ "reward_shaping_gamma": 0.99,
23
+ "reward_shaping_material_weight": 0.6,
24
+ "reward_shaping_mobility_weight": 0.4,
25
+ "reward_shaping_draw_penalty": 0.1,
26
+ "league_enabled": true,
27
+ "league_path": "checkpoints/league_ratings.json",
28
+ "league_checkpoint_pool_top_k": 3,
29
+ "league_checkpoint_pool_min_games": 4,
30
+ "league_selfplay_checkpoint_prob": 0.55,
31
+ "league_champion_games": 6,
32
+ "buffer_size": 50000,
33
+ "val_split": 0.1,
34
+ "shuffle_train_val_split": true,
35
+ "train_recent_fraction": 0.7,
36
+ "train_recent_window_fraction": 0.4,
37
+ "d_model": 128,
38
+ "nhead": 8,
39
+ "num_layers": 6,
40
+ "dim_feedforward": 512,
41
+ "dropout": 0.1,
42
+ "log_dir": "/kaggle/working/logs",
43
+ "checkpoint_dir": "/kaggle/working/checkpoints",
44
+ "save_every": 6,
45
+ "keep_last_n_local_checkpoints": 3,
46
+ "keep_last_n_log_versions": 2,
47
+ "keep_last_n_hf_checkpoints": 3,
48
+ "onnx_path": "ataxx_model.onnx",
49
+ "export_onnx": true,
50
+ "hf_enabled": true,
51
+ "hf_repo_id": "dieg0code/ataxx-zero",
52
+ "hf_run_id": "policy_spatial_v9",
53
+ "hf_bootstrap_run_id": "policy_spatial_v8",
54
+ "hf_reset_iteration": true,
55
+ "hf_token_env": "HF_TOKEN",
56
+ "hf_local_dir": "/kaggle/working/hf_checkpoints",
57
+ "max_pending_hf_uploads": 2,
58
+ "hf_upload_future_timeout_s": 120.0,
59
+ "show_progress_bar": false,
60
+ "trainer_log_every_n_steps": 99999,
61
+ "monitor_log_every": 5,
62
+ "epoch_pulse_every": 2,
63
+ "num_workers": 4,
64
+ "persistent_workers": true,
65
+ "strict_probs": false,
66
+ "trainer_devices": 2,
67
+ "trainer_strategy": "ddp_spawn",
68
+ "trainer_precision": "16-mixed",
69
+ "trainer_benchmark": true,
70
+ "ddp_timeout_seconds": 180,
71
+ "mcts_use_amp": true,
72
+ "mcts_cache_size": 100000,
73
+ "mcts_leaf_batch_size": 32,
74
+ "fail_on_selfplay_parallel_error": true,
75
+ "fail_on_hf_upload_error": true,
76
+ "opponent_self_prob": 0.6,
77
+ "opponent_heuristic_prob": 0.25,
78
+ "opponent_random_prob": 0.15,
79
+ "opponent_heuristic_level": "normal",
80
+ "opponent_heuristic_epsilon": 0.1,
81
+ "opponent_heuristic_easy_prob": 0.2,
82
+ "opponent_heuristic_normal_prob": 0.5,
83
+ "opponent_heuristic_hard_prob": 0.3,
84
+ "model_side_swap_prob": 0.5,
85
+ "eval_enabled": true,
86
+ "eval_every": 6,
87
+ "eval_games": 64,
88
+ "eval_sims": 160,
89
+ "eval_heuristic_level": "hard",
90
+ "eval_heuristic_levels": "easy,normal,hard,apex,gambit,sentinel",
91
+ "restore_best_on_regression": true,
92
+ "eval_regression_delta": 0.06,
93
+ "eval_regression_patience": 2,
94
+ "selfplay_workers": 2,
95
+ "selfplay_progress_every_s": 120.0,
96
+ "selfplay_episode_timeout_s": 1800.0,
97
+ "compile_model": true,
98
+ "quiet_mode": false,
99
+ "warmup_games": 320,
100
+ "warmup_epochs": 4,
101
+ "warmup_heuristic_level": "sentinel",
102
+ "warmup_heuristic_levels": "hard,apex,sentinel"
103
+ },
104
+ "stats": {
105
+ "replay_size": 40365,
106
+ "best_eval_score": 0.6458333333333334,
107
+ "train_loss_total": 2.119072914123535,
108
+ "train_loss_value": 0.569379985332489,
109
+ "train_loss_policy": 1.8343825340270996,
110
+ "train_value_mae": 0.5733582973480225,
111
+ "train_policy_accuracy": 0.37041011452674866,
112
+ "train_lr": 0.00029997737146914005,
113
+ "score": 0.6458333333333334,
114
+ "eval_total_wins": 248,
115
+ "eval_total_losses": 136,
116
+ "eval_total_draws": 0,
117
+ "eval_levels": "easy,normal,hard,apex,gambit,sentinel",
118
+ "eval_score_easy": 0.484375,
119
+ "eval_score_normal": 0.390625,
120
+ "eval_score_hard": 0.953125,
121
+ "eval_score_apex": 0.875,
122
+ "eval_score_gambit": 0.265625,
123
+ "eval_score_sentinel": 0.90625
124
+ }
125
+ }