{
  "name": "eqlm-small-140m",
  "hf_config": {
    "name": "eqlm-small-140m"
  },
  "block_size": 2048,
  "n_embd": 1024,
  "intermediate_size": 4096,
  "num_attention_heads": 8,
  "num_key_value_heads": 8,
  "vocab_size": 32768,
  "padding_multiple": 64,
  "padded_vocab_size": 32768,
  "rope_settings": {
    "use_rope": true,
    "rope_condense_ratio": 1,
    "rope_base": 50000
  },
  "use_abacus": false,
  "randomize_positions_from": null,
  "block_class_name": "TransformerPreNormBlock",
  "norm_class_name": "RMSNorm",
  "attn_impl": "flash",
  "norm_eps": 1e-05,
  "mlp_class_name": "BaseMLP",
  "nonlin_name": "ReLU2",
  "bias": false,
  "qk_bias": false,
  "init_strategy": "scaled-zero",
  "init_orthogonal": true,
  "skip_initialization": false,
  "mup_model_scaling_factor": 1,
  "use_fused_head": "pytorch",
  "debias_attention": false,
  "center_attention": false,
  "clip_qkv": null,
  "qk_norm": true,
  "logit_softcap": null,
  "causal": true,
  "activation_checkpoint_impl": "per-block",
  "simple_ops": false,
  "strategy": "ddp",
  "n_backbone_layers": 7,
  "n_fp_blocks": 1,
  "tie_embeddings": true,
  "solver": "anderson",
  "max_iter": 64,
  "min_iter": 6,
  "tol": 0.0003,
  "anderson_m": 5,
  "anderson_beta": 1.0,
  "backward_type": "onestep",
  "backward_max_iter": 64,
  "backward_min_iter": 6,
  "backward_tol": 0.0003,
  "adjoint_grad_clip": null,
  "layer_scale_init": 0.75,
  "gamma_max": 0.75,
  "fp_lr_scale": 0.5,
  "fp_wd": 0.1,
  "recurrent_embedding_dimension": 1024,
  "model_class_name": "EQLM",
  "_class_name": "EQLMConfig"
}