config.json

{
  "add_linear_bias": true,
  "architectures": [
    "FM4BioForMaskedLM"
  ],
  "attention_probs_dropout_prob": 0.0,
  "experts_per_token": 2,
  "hidden_act": "swiglu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 2304,
  "initializer_range": 0.02,
  "intermediate_size": 7680,
  "layer_norm_eps": 1e-05,
  "max_position_embeddings": 1024,
  "model_type": "fm4bio",
  "moe": true,
  "normalization_type": "RMSNorm",
  "num_attention_heads": 36,
  "num_experts": 8,
  "num_hidden_layers": 36,
  "output_vocab_size": 512,
  "pad_token_id": 0,
  "position_embedding_type": "rope",
  "rotary_percent": 1.0,
  "seq_len_interpolation_factor": null,
  "tie_word_embeddings": false,
  "torch_dtype": "float32",
  "transformers_version": "4.38.0.dev0",
  "type_vocab_size": 2,
  "use_cache": true,
  "use_lm_head": false,
  "vocab_size": 128
}