lact-muon-nope-postnorm-nheads2-chunk2048-760m/config.json

{
  "_name_or_path": "configs/lact/760M_lact_swiglu_nh4_fwlow_rank_momentum_muon.json",
  "architectures": [
    "LaCTForCausalLM"
  ],
  "attention_bias": false,
  "attn_qk_norm": false,
  "bos_token_id": 1,
  "elementwise_affine": true,
  "eos_token_id": 2,
  "fuse_cross_entropy": true,
  "fuse_norm": true,
  "fuse_swiglu": true,
  "fw_init_gain": 0.5,
  "hidden_act": "swish",
  "hidden_ratio": 4,
  "hidden_size": 1536,
  "initializer_range": 0.02,
  "inter_multi": 1,
  "intermediate_size": null,
  "lact_chunk_size": 2048,
  "last_layer_fuse_norm": true,
  "learnable_ttt_scale": true,
  "lr_dim": 1,
  "lr_parameterization": "mamba",
  "max_position_embeddings": 32768,
  "model_type": "lact_swiglu",
  "norm_eps": 1e-06,
  "num_attn_heads": 24,
  "num_hidden_layers": 24,
  "num_lact_heads": 4,
  "qkv_bias": false,
  "qkv_silu": true,
  "rope_theta": 1000000,
  "tie_word_embeddings": false,
  "torch_dtype": "float32",
  "transformers_version": "4.45.2",
  "ttt_loss_type": "dot_product",
  "ttt_nope": true,
  "ttt_prenorm": false,
  "use_cache": false,
  "use_momentum": true,
  "use_muon": true,
  "vocab_size": 32000,
  "w0_w2_low_rank": 32,
  "window_size": 2048
}