pred/config.json

{
  "architectures": [
    "SmolPredictorForConditionalGeneration"
  ],
  "auto_map": {
    "AutoConfig": "configuration_predictor_smol.SmolPredictorConfig",
    "AutoModel": "modeling_predictor_smolv3.SmolPredictorForConditionalGeneration",
    "AutoModelForVision2Seq": "modeling_predictor_smolv3.SmolPredictorForConditionalGeneration"
  },
  "beta_add_one": false,
  "beta_init_mode": "uniform",
  "beta_param_scale": 0.5,
  "continuous_dist": "beta",
  "continuous_eval_quantile": 0.5,
  "contrastive_margin": 0.0,
  "contrastive_temperature": 0.1,
  "contrastive_weight": 0.2,
  "cross_depth": 1,
  "dim_head": 64,
  "dirichlet_budget": 1.0,
  "dirichlet_eps": 1e-06,
  "dirichlet_init_weight_std": null,
  "dirichlet_logit_noise_std": 0.0,
  "dropout": 0.0,
  "dtype": "bfloat16",
  "dual_path_depth": 1,
  "enable_rope_cache": true,
  "ff_mult": 4,
  "force_uniform_ab": false,
  "fps": 2.0,
  "frame_info_ema_alpha": 0.9,
  "frame_temporal_depth": 2,
  "gate_k_ratio": 0.25,
  "gate_mode": "gumbel_softmax",
  "gate_query_scale": 1.0,
  "gate_temperature": 1.0,
  "hidden_size": 768,
  "info_fuse_mode": "pooled_ln",
  "init_concentration": 6.0,
  "init_head": true,
  "init_scale_mean": 1.0,
  "init_weight_std": 0.001,
  "layer_fusion_indices": null,
  "layer_fusion_mode": "none",
  "layer_weighted_num_layers": 0,
  "llm_hidden_size": 768,
  "logistic_normal_init_sigma": 0.7,
  "max_frames": 128,
  "max_scale": 1.8,
  "min_scale": 0.2,
  "mlp_mode": "mlp",
  "model_type": "smol_predictor",
  "num_heads": 16,
  "out_hidden_size": 576,
  "output_dim": 1024,
  "patch_size": 16,
  "pool_gate_mode": "no_ln",
  "regression_head_mode": "mlp",
  "scale_bins": null,
  "self_depth": 2,
  "sim_gamma": 0.05,
  "sim_scale_weight": 0.2,
  "sim_tau": 0.6,
  "sim_temp": 0.15,
  "smol_model_name": "HuggingFaceTB/SmolVLM2-256M-Video-Instruct",
  "spatial_merge_size": 2,
  "temporal_mixer_depth": 1,
  "temporal_use_pos": true,
  "text_encoder_depth": 0,
  "text_encoder_dropout": 0.0,
  "text_encoder_ff_mult": 4,
  "text_encoder_heads": null,
  "transformers_version": "4.57.6",
  "use_cross_attn_gate": true,
  "use_differentiable_importance": false,
  "use_dirichlet_budget": false,
  "use_discrete_action": false,
  "use_frame_info": false,
  "use_learned_pooling": true,
  "use_text": true,
  "use_text_conditioned_spatial": false,
  "use_text_encoder": false,
  "vision_hidden_size": 576,
  "vlp_mode": "mlp",
  "vocab_size": 49280
}