Checkpoint iteration 10

runs/policy_spatial_v13/metadata_iter_010.json

@@ -0,0 +1,139 @@
+{
+  "iteration": 10,
+  "timestamp": "2026-05-24T05:15:06.812822+00:00",
+  "buffer_size": 3249,
+  "config": {
+    "iterations": 600,
+    "episodes_per_iter": 6,
+    "mcts_sims": 800,
+    "c_puct": 1.25,
+    "temp_threshold": 28,
+    "add_noise": true,
+    "seed": 42,
+    "verbose_logs": true,
+    "episode_log_every": 25,
+    "epochs": 1,
+    "batch_size": 128,
+    "learning_rate": 0.0003,
+    "weight_decay": 0.0001,
+    "value_loss_coeff": 0.5,
+    "reward_shaping_enabled": true,
+    "reward_shaping_scale": 0.1,
+    "reward_shaping_gamma": 0.99,
+    "reward_shaping_material_weight": 0.6,
+    "reward_shaping_mobility_weight": 0.4,
+    "reward_shaping_draw_penalty": 0.1,
+    "league_enabled": true,
+    "league_path": "checkpoints/league_ratings.json",
+    "league_checkpoint_pool_top_k": 3,
+    "league_checkpoint_pool_min_games": 4,
+    "league_selfplay_checkpoint_prob": 0.55,
+    "league_champion_games": 6,
+    "buffer_size": 50000,
+    "val_split": 0.1,
+    "shuffle_train_val_split": true,
+    "train_recent_fraction": 0.7,
+    "train_recent_window_fraction": 0.4,
+    "d_model": 384,
+    "nhead": 12,
+    "num_layers": 12,
+    "dim_feedforward": 1536,
+    "dropout": 0.1,
+    "log_dir": "/kaggle/working/logs",
+    "checkpoint_dir": "/kaggle/working/checkpoints",
+    "save_every": 1,
+    "keep_last_n_local_checkpoints": 3,
+    "keep_last_n_log_versions": 2,
+    "keep_last_n_hf_checkpoints": 3,
+    "onnx_path": "ataxx_model.onnx",
+    "export_onnx": false,
+    "hf_enabled": true,
+    "hf_repo_id": "dieg0code/ataxx-zero",
+    "hf_run_id": "policy_spatial_v13",
+    "hf_bootstrap_run_id": "",
+    "hf_reset_iteration": false,
+    "hf_token_env": "HF_TOKEN",
+    "hf_local_dir": "/kaggle/working/hf_checkpoints",
+    "wandb_enabled": true,
+    "wandb_project": "ataxx-zero",
+    "wandb_entity": "dieg0code-ai",
+    "wandb_run_name": "policy_spatial_v13",
+    "max_pending_hf_uploads": 12,
+    "hf_upload_future_timeout_s": 900.0,
+    "show_progress_bar": false,
+    "trainer_log_every_n_steps": 99999,
+    "monitor_log_every": 5,
+    "epoch_pulse_every": 2,
+    "num_workers": 2,
+    "persistent_workers": true,
+    "strict_probs": false,
+    "trainer_devices": 2,
+    "trainer_strategy": "ddp_spawn",
+    "trainer_precision": "16-mixed",
+    "trainer_benchmark": true,
+    "ddp_timeout_seconds": 180,
+    "mcts_use_amp": true,
+    "mcts_cache_size": 100000,
+    "mcts_leaf_batch_size": 32,
+    "mcts_dirichlet_alpha": 0.1,
+    "mcts_dirichlet_frac": 0.25,
+    "fail_on_selfplay_parallel_error": false,
+    "fail_on_hf_upload_error": false,
+    "opponent_self_prob": 0.45,
+    "opponent_heuristic_prob": 0.0,
+    "opponent_random_prob": 0.05,
+    "opponent_heuristic_level": "normal",
+    "opponent_heuristic_epsilon": 0.0,
+    "opponent_heuristic_easy_prob": 0.2,
+    "opponent_heuristic_normal_prob": 0.5,
+    "opponent_heuristic_hard_prob": 0.3,
+    "model_side_swap_prob": 0.5,
+    "eval_enabled": true,
+    "eval_every": 6,
+    "eval_games": 64,
+    "eval_sims": 256,
+    "eval_heuristic_level": "hard",
+    "eval_heuristic_levels": "easy,normal,hard,apex,gambit,sentinel",
+    "restore_best_on_regression": false,
+    "eval_regression_delta": 0.06,
+    "eval_regression_patience": 2,
+    "baseline_checkpoint": "liga",
+    "baseline_hf_run_id": "policy_spatial_v8",
+    "baseline_composite": 0.85,
+    "baseline_h2h_min_score": 0.25,
+    "eval_absolute_patience": 2,
+    "eval_absolute_delta": 0.03,
+    "eval_absolute_action": "abort",
+    "eval_absolute_abort_mode": "h2h",
+    "eval_absolute_min_iteration": 150,
+    "selfplay_workers": 2,
+    "selfplay_progress_every_s": 120.0,
+    "selfplay_episode_timeout_s": 1800.0,
+    "compile_model": false,
+    "quiet_mode": false,
+    "warmup_games": 0,
+    "warmup_epochs": 0,
+    "warmup_heuristic_level": "sentinel",
+    "warmup_heuristic_levels": "hard,apex,sentinel",
+    "pretrain_dataset_path": "/kaggle/working/data/curated/v13_pretrain.npz",
+    "pretrain_epochs": 3,
+    "human_replay_path": "/kaggle/working/data/curated/v13_human_replay.npz",
+    "human_batch_fraction": 0.2,
+    "human_value_mask": true,
+    "symmetry_augmentation": true,
+    "value_head_depth": 2,
+    "count_head_enabled": true,
+    "count_loss_coeff": 0.1,
+    "eval_composite_uses_h2h_only": true
+  },
+  "stats": {
+    "replay_size": 3249,
+    "best_eval_score": 0.00390625,
+    "train_loss_total": 4.827633857727051,
+    "train_loss_value": 1.9580296277999878,
+    "train_loss_policy": 3.8380746841430664,
+    "train_value_mae": 0.9904497861862183,
+    "train_policy_accuracy": 0.03966083005070686,
+    "train_lr": 0.0002999979769811034
+  }
+}