config.json

{
  "architectures": [
    "NemotronForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "unk_token_id": 1,
  "bos_token_id": 2,
  "eos_token_id": 3,
  "pad_token_id": 0,
  "hidden_act": "relu2",
  "hidden_size": 18432,
  "initializer_range": 0.0063,
  "intermediate_size": 73728,
  "max_position_embeddings": 4096,
  "model_type": "nemotron",
  "num_attention_heads": 96,
  "num_hidden_layers": 96,
  "num_key_value_heads": 8,
  "pretraining_tp": 1,
  "layer_norm_eps": 1e-05,
  "rope_scaling": null,
  "partial_rotary_factor": 0.5,
  "rope_theta": 10000.0,
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.40.0.dev0",
  "use_cache": true,
  "vocab_size": 256000
}