config.json

{
  "_name_or_path": "melhoushi/slimpj_h896_d17_gbs66_tpp20.0_lp0.4_linear_linear_reverse",
  "architectures": [
    "NestedLlamaForCausalLM"
  ],
  "attention_bias": true,
  "attention_dropout": 0.0,
  "auto_map": {
    "AutoConfig": "melhoushi/slimpj_h896_d17_gbs66_tpp20.0_lp0.4_linear_linear_reverse--configuration_celerity.CelerityConfig",
    "AutoModel": "melhoushi/slimpj_h896_d17_gbs66_tpp20.0_lp0.4_linear_linear_reverse--modeling_celerity.CelerityModel",
    "AutoModelForCausalLM": "melhoushi/slimpj_h896_d17_gbs66_tpp20.0_lp0.4_linear_linear_reverse--modeling_celerity.CelerityForCausalLM",
    "AutoModelForQuestionAnswering": "melhoushi/slimpj_h896_d17_gbs66_tpp20.0_lp0.4_linear_linear_reverse--modeling_celerity.CelerityForQuestionAnswering",
    "AutoModelForSequenceClassification": "melhoushi/slimpj_h896_d17_gbs66_tpp20.0_lp0.4_linear_linear_reverse--modeling_celerity.CelerityForSequenceClassification",
    "AutoModelForTokenClassification": "melhoushi/slimpj_h896_d17_gbs66_tpp20.0_lp0.4_linear_linear_reverse--modeling_celerity.CelerityForTokenClassification"
  },
  "bos_token_id": 1,
  "embeddings_scale": 1.0,
  "eos_token_id": 2,
  "exit_attention": false,
  "exit_decoder_layer": true,
  "exit_layer_indices": [
    2,
    4,
    9,
    13
  ],
  "exit_mlp": false,
  "head_dim": 64,
  "hidden_act": "relu2",
  "hidden_size": 896,
  "initializer_range": 0.02,
  "intermediate_size": 7168,
  "max_position_embeddings": 8192,
  "mhsa_residual_scale": 0.050850000000000006,
  "mlp_bias": true,
  "mlp_residual_scale": 0.029735294117647054,
  "model_type": "nested_llama",
  "norm_type": "layernorm",
  "num_attention_heads": 14,
  "num_hidden_layers": 17,
  "num_key_value_heads": 14,
  "output_exit_layers": [
    2,
    4,
    9,
    13
  ],
  "output_full_model": true,
  "output_logits_scale": 0.0011160714285714285,
  "position_embedding_type": "alibi",
  "pretraining_tp": 1,
  "rms_norm_eps": 1e-05,
  "rope_scaling": null,
  "rope_theta": 10000.0,
  "scale_qk_dot_by_d": true,
  "tie_exit_lm_head": true,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.48.0",
  "tuple_outputs": true,
  "use_cache": false,
  "use_mlp_gating": false,
  "vocab_size": 128256
}