config.json

{
  "architectures": [
    "KairosModel"
  ],
  "classifier_dropout": 0,
  "context_length": 2048,
  "cross_attention_pe_flip": false,
  "d_ff": 1024,
  "d_kv": 64,
  "d_model": 256,
  "decoder_start_token_id": 0,
  "dense_act_fn": "relu",
  "dropout_rate": 0.1,
  "dtype": "float32",
  "eos_token_id": 1,
  "feed_forward_proj": "relu",
  "initializer_factor": 0.05,
  "input_patch_size": 128,
  "input_patch_stride": 128,
  "instance_rope_input_feature_dim": 128,
  "is_cross_attention_pe": true,
  "is_encoder_decoder": true,
  "is_gated_act": false,
  "layer_norm_epsilon": 1e-06,
  "levels": 3,
  "loss_weight_scheme": "log_decay",
  "max_period": "original_rope_init",
  "min_period": "original_rope_init",
  "model_type": "kairos",
  "moe_inter_dim": 1408,
  "n_activated_experts": 3,
  "n_expert_groups": 1,
  "n_limited_groups": 1,
  "n_null_experts": 2,
  "n_positions": 512,
  "num_decoder_layers": 4,
  "num_decoder_segments": 2,
  "num_heads": 4,
  "num_layers": 4,
  "pad_token_id": 0,
  "position_embedding_type": "instance_wise_rope",
  "prediction_length": 64,
  "pretrained_model_path": "",
  "quantiles": [
    0.1,
    0.2,
    0.3,
    0.4,
    0.5,
    0.6,
    0.7,
    0.8,
    0.9
  ],
  "reg_token_id": 1,
  "relative_attention_max_distance": 128,
  "relative_attention_num_buckets": 32,
  "rope_init": "exp",
  "scale_method": "log",
  "score_func": "softmax",
  "seq_balance_factor": 0.0001,
  "target_dist": [
    0.05,
    0.1,
    0.55,
    0.15,
    0.15
  ],
  "transformers_version": "4.56.1",
  "update_bias_rate": 0.01,
  "use_cache": true,
  "use_reg_token": true,
  "vocab_size": 2
}