config.json

{
  "activation_function": "silu",
  "architectures": [
    "MobilintExaoneForCausalLM"
  ],
  "attention_dropout": 0.0,
  "auto_map": {
    "AutoConfig": "proxy_exaone.MobilintExaoneConfig",
    "AutoModelForCausalLM": "proxy_exaone.MobilintExaoneForCausalLM"
  },
  "bos_token_id": 1,
  "embed_dropout": 0.0,
  "eos_token_id": 361,
  "head_dim": 128,
  "hidden_size": 4096,
  "initializer_range": 0.02,
  "intermediate_size": 14336,
  "layer_norm_epsilon": 1e-05,
  "max_position_embeddings": 16384,
  "model_type": "mobilint-exaone",
  "mxq_path": "EXAONE-Deep-7.8B-W8.mxq",
  "num_attention_heads": 32,
  "num_key_value_heads": 8,
  "num_layers": 32,
  "pad_token_id": 0,
  "rope_scaling": {
    "factor": 8.0,
    "high_freq_factor": 4.0,
    "low_freq_factor": 1.0,
    "original_max_position_embeddings": 4096,
    "rope_type": "llama3"
  },
  "rope_theta": 1000000.0,
  "target_cores": [
    "0:0"
  ],
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.43.1",
  "use_cache": true,
  "vocab_size": 102400,
  "npu_prefill_chunk_size": {
    "global4": 512,
    "global8": 512,
    "single": 512
  }
}