config.json

{
  "activation_function": "silu",
  "architectures": [
    "VoxtralRealtimeEncoder"
  ],
  "attention_dropout": 0.0,
  "dtype": "float32",
  "head_dim": 8,
  "hidden_act": "silu",
  "hidden_size": 32,
  "initializer_range": 0.02,
  "intermediate_size": 32,
  "max_position_embeddings": 256,
  "model_type": "voxtral_realtime_encoder",
  "num_attention_heads": 2,
  "num_hidden_layers": 2,
  "num_mel_bins": 80,
  "rms_norm_eps": 1e-06,
  "rope_parameters": {
    "rope_theta": 10000.0,
    "rope_type": "default"
  },
  "sliding_window": 750,
  "transformers_version": "5.16.0.dev0",
  "vocab_size": 99
}