config.json

{
    "model_type": "custom_seq2seq",
  "architectures": [
    "custom_seq2seq_llm"
  ],
  "auto_map": {
    "AutoConfig": "modeling_custom_seq2seq_llm.Seq2SeqConfig",
    "AutoModelForSeq2SeqLM": "modeling_custom_seq2seq_llm.CustomSeq2SeqLLM"
  },
  "attention_probs_dropout_prob": 0.0,
  "bos_token_id": 1,
  "decoder_start_token_id": 3,
  "eos_token_id": 2,
  "head_dim": 64,
  "hidden_act": "silu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 1024,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_norm_eps": 1e-12,
  "max_position_embeddings": 1024,
  "num_attention_heads": 16,
  "num_decoder_layers": 32,
  "num_encoder_layers": 16,
  "num_key_value_heads": 4,
  "pad_token_id": 3,
  "rotary_emb_base": 10000.0,
  "rotary_emb_dim": 32,
  "rotary_emb_interleaved": false,
  "rotary_emb_scale_base": null,
  "transformers_version": "4.44.2",
  "use_cache": true,
  "vocab_size": 48256
}