config.json

{
  "_name_or_path": "./12l-10_stacked",
  "architectures": [
    "CustomTransformerForMaskedLM"
  ],
  "attention_probs_dropout_prob": 0.1,
  "attn_out_bias": false,
  "attn_out_dropout_prob": 0.0,
  "attn_qkv_bias": false,
  "auto_map": {
    "AutoConfig": "model.CustomTransformerConfig",
    "AutoModel": "model.CustomTransformerModel",
    "AutoModelForMaskedLM": "model.CustomTransformerForMaskedLM",
    "AutoModelForSequenceClassification": "model.CustomTransformerForSequenceClassification"
  },
  "bos_token_id": 1,
  "classifier_dropout": null,
  "context_len": 128,
  "deterministic_fa2": false,
  "embedding_size": 768,
  "eos_token_id": 2,
  "ffn_dim_multiplier": null,
  "ffn_hidden_dims": 1536,
  "global_attn_every_n_layers": 3,
  "hidden_size": 768,
  "layernorm_eps": 1e-06,
  "local_attn_rotary_emb_base": -1,
  "local_attn_rotary_emb_dim": null,
  "mask_token_id": 3,
  "model_type": "custom_transformer",
  "moe_aux_loss_coef": 0.01,
  "moe_eps": 1e-06,
  "moe_num_experts": 15,
  "moe_routed_experts": 1,
  "moe_shared_experts": 1,
  "num_attention_heads": 12,
  "num_dims": 768,
  "num_heads": 12,
  "num_kv_heads": 12,
  "num_layers": 12,
  "pad_token_id": 0,
  "rope_theta": 100000.0,
  "rotary_emb_base": 10000,
  "rotary_emb_dim": 64,
  "rotary_emb_interleaved": false,
  "rotary_emb_scale_base": null,
  "sliding_window": 128,
  "torch_dtype": "float32",
  "transformers_version": "4.48.0",
  "use_cache": false,
  "use_fa2": true,
  "use_flash": true,
  "use_lossfreebalance": true,
  "use_moe": true,
  "use_sdpa_attn_mask": false,
  "vocab_size": 50368
}