config.json

{
  "architectures": [
    "ZFusionAIForCausalLM"
  ],
  "bos_token_id": 0,
  "dtype": "bfloat16",
  "eos_token_id": 1,
  "hidden_size": 768,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "kv_lora_rank": 128,
  "max_position_embeddings": 65536,
  "model_type": "zfusionai_v2",
  "num_attention_heads": 8,
  "num_hidden_layers": 20,
  "num_key_value_heads": 2,
  "pad_token_id": 2,
  "qk_rope_head_dim": 32,
  "rms_norm_eps": 1e-06,
  "rope_scaling": {
    "factor": 2.0,
    "original_max_position_embeddings": 32768,
    "type": "yarn"
  },
  "rope_theta": 100000.0,
  "transformers_version": "4.57.1",
  "use_cache": false,
  "v_head_dim": 64,
  "vocab_size": 32768
}