config.json

{
  "add_cross_attention": false,
  "add_router_probs": false,
  "architectures": [
    "SwitchTransformersForConditionalGeneration"
  ],
  "bos_token_id": 0,
  "d_ff": 37,
  "d_kv": 8,
  "d_model": 32,
  "decoder_sparse_step": 0,
  "decoder_start_token_id": 0,
  "dense_act_fn": "relu",
  "dropout_rate": 0.1,
  "dtype": "float32",
  "encoder_sparse_step": 0,
  "eos_token_id": 1,
  "expert_capacity": 100,
  "initializer_factor": 0.002,
  "is_decoder": false,
  "is_encoder_decoder": true,
  "layer_norm_epsilon": 1e-06,
  "model_type": "switch_transformers",
  "num_decoder_layers": 2,
  "num_experts": 8,
  "num_heads": 4,
  "num_layers": 2,
  "num_sparse_decoder_layers": 3,
  "num_sparse_encoder_layers": 3,
  "pad_token_id": 0,
  "relative_attention_max_distance": 128,
  "relative_attention_num_buckets": 8,
  "router_aux_loss_coef": 0.001,
  "router_bias": false,
  "router_dtype": "float32",
  "router_ignore_padding_tokens": false,
  "router_jitter_noise": 0.0,
  "router_z_loss_coef": 0.001,
  "tie_word_embeddings": true,
  "transformers_version": "5.16.0.dev0",
  "use_cache": true,
  "vocab_size": 32100
}