| { | |
| "method": "dense_full", | |
| "width": 1536, | |
| "heads": 12, | |
| "vocab_size": 32768, | |
| "sequence_len": 2048, | |
| "mlp_width": 6144, | |
| "layers": 23, | |
| "pre_layers": 3, | |
| "core_layers": 9, | |
| "post_layers": 3, | |
| "loops": 2, | |
| "window_pattern": "L", | |
| "max_loops": 3, | |
| "act_epsilon": 0.01, | |
| "ponder_weight": 0.01, | |
| "huginn_mean": 3.0, | |
| "huginn_sigma": 0.5, | |
| "huginn_backprop_loops": 2, | |
| "routing_seed": 1729, | |
| "eval_loops": 3, | |
| "ouro_entropy_weight": 0.1, | |
| "ouro_gate_steps": 1024, | |
| "ouro_exit_quantile": 0.5, | |
| "sparse_topk": 256, | |
| "block_size": 128, | |
| "compression_size": 4, | |
| "local_window": 512, | |
| "index_heads": 4, | |
| "index_dim": 128, | |
| "index_rope_dim": 64, | |
| "index_loss_weight": 1.0, | |
| "burnin_steps": 512, | |
| "slots": 64, | |
| "gate_rank": 16, | |
| "conv_size": 4, | |
| "reference_chunk": 32, | |
| "backend": "reference", | |
| "scale": "formal", | |
| "residual_streams": 4, | |
| "residual_gate_init": 0.01, | |
| "sinkhorn_iterations": 20, | |
| "attnres_groups": 8 | |
| } | |