bert4torch_config.json

{
  "model": "llama",
  "hidden_size": 2048,
  "intermediate_size": 5632,
  "num_attention_heads": 32,
  "num_key_value_heads": 4,
  "num_hidden_layers": 22,
  "hidden_act": "silu",
  "vocab_size": 64793,
  "segment_vocab_size": 0,
  "layer_norm_eps": 1e-06,
  "rope_rank": "updown",
  "flash_attention": true,
  "tie_emb_prj_weight": true,
  "is_causal": true
}