{ "architecture": "ChessTransformer200M", "params": 204006404, "encoder_dim": 256, "hidden_dim": 1024, "num_layers": 16, "num_heads": 16, "ffn_ratio": 4, "policy_head_dim": 512, "value_hidden": 512, "vocab_size": 5504, "training": { "steps": 10000, "positions_seen": 10240256, "batch_size": 256, "accum_steps": 4, "effective_batch": 1024, "lr": 0.0002, "best_accuracy": 0.184, "dataset": "avewright/chess-positions-lichess-sf", "gpu": "NVIDIA A40 46GB" } }