config.json

{
  "architectures": [
    "NVEsmForMaskedLM"
  ],
  "attention_probs_dropout_prob": 0.0,
  "attn_input_format": "bshd",
  "attn_mask_type": "padding",
  "auto_map": {
    "AutoConfig": "esm_nv.NVEsmConfig",
    "AutoModel": "esm_nv.NVEsmModel",
    "AutoModelForMaskedLM": "esm_nv.NVEsmForMaskedLM",
    "AutoModelForTokenClassification": "esm_nv.NVEsmForTokenClassification"
  },
  "classifier_dropout": null,
  "dtype": "float32",
  "emb_layer_norm_before": false,
  "encoder_activation": "gelu",
  "esmfold_config": null,
  "fuse_qkv_params": true,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 2560,
  "initializer_range": 0.02,
  "intermediate_size": 10240,
  "is_folding_model": false,
  "layer_norm_eps": 1e-05,
  "mask_token_id": 32,
  "max_position_embeddings": 1026,
  "max_seq_length": null,
  "micro_batch_size": null,
  "model_type": "nv_esm",
  "num_attention_heads": 40,
  "num_hidden_layers": 36,
  "pad_token_id": 1,
  "padded_vocab_size": 64,
  "position_embedding_type": "rotary",
  "qkv_weight_interleaved": true,
  "token_dropout": true,
  "transformers_version": "4.57.3",
  "use_cache": true,
  "vocab_list": null,
  "vocab_size": 33
}