config.json

{
  "adapter_reduction": 16,
  "architectures": [
    "ModernALBERTForMaskedLM"
  ],
  "attention_probs_dropout_prob": 0.1,
  "classifier_dropout": null,
  "embedding_size": 128,
  "group_depth": 4,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.1,
  "hidden_size": 1024,
  "initializer_range": 0.02,
  "intermediate_size": 2624,
  "layer_norm_eps": 1e-12,
  "load_balancing_loss_coef": 0.01,
  "max_position_embeddings": 8192,
  "moa_reduction_factor": 4,
  "model_type": "ModernALBERT",
  "num_attention_heads": 16,
  "num_experts": 4,
  "num_hidden_layers": 16,
  "pad_token_id": 0,
  "router_jitter_noise": 0.01,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.2",
  "use_adapter": true,
  "use_cache": true,
  "use_moa": true,
  "vocab_size": 50368
}