{ "architectures": [ "CLPNCMForNextConceptModeling" ], "model_type": "cl_pncm", "auto_map": { "AutoConfig": "modeling_cl_pncm.CLPNCMHuggingFaceConfig", "AutoModel": "modeling_cl_pncm.CLPNCMForNextConceptModeling" }, "vocab_size": 8192, "pad_token_id": 0, "bos_token_id": 1, "eos_token_id": 2, "mask_token_id": 3, "abstain_token_id": 4, "d_enc": 512, "d_core": 768, "d_dec": 512, "mlp_ratio": 2.6667, "enc_layers": 4, "enc_heads": 8, "enc_kv_heads": 2, "local_window_size": 128, "entropy_split_threshold": 2.4, "min_patch_len": 1, "max_patch_len": 8, "n_coarse_codes": 4096, "vq_ema_decay": 0.99, "vq_commitment_weight": 0.25, "fsq_levels": 3, "fsq_bins": [ 8, 8, 8, 6, 5 ], "core_layers": 8, "core_heads": 12, "core_kv_heads": 4, "policy_temp": 1.0, "tau_high": 0.85, "tau_low": 0.4, "deliberative_beams": 4, "loop_layers": 3, "loop_heads": 8, "max_loops": 6, "halt_tol": 0.001, "share_loop_weights": true, "use_xsa": true, "use_attn_gate": true, "deep_supervision_weighting": "exponential", "dec_layers": 4, "dec_heads": 8, "pointer_dropout": 0.5, "boundary_conf_threshold": 0.75, "stage": 1, "gamma_copy": 0.5, "alpha_loop": 1.0, "eta_contract": 0.2, "contraction_rho": 0.7, "grpo_group_size": 4, "lambda_brier": 1.0, "mu_ponder": 0.02, "beta_kl": 0.01, "c_abstain": 0.15, "learning_rate": 0.0003, "warmup_steps": 200, "num_steps": 5000, "batch_size": 32, "micro_batch_size": 4, "max_grad_norm": 1.0, "adam_beta2": 0.95, "weight_decay": 0.01, "seed": 42, "amp_dtype": "bf16" }