config.json

{
  "model_type": "minimind",
  "architectures": [
    "MiniMindForEmbedding"
  ],
  "hidden_size": 768,
  "num_hidden_layers": 8,
  "num_attention_heads": 8,
  "num_key_value_heads": 4,
  "head_dim": 96,
  "vocab_size": 6400,
  "intermediate_size": 2432,
  "max_position_embeddings": 32768,
  "rope_theta": 1000000.0,
  "rms_norm_eps": 1e-06,
  "bos_token_id": 1,
  "eos_token_id": 2,
  "tie_word_embeddings": false,
  "use_moe": true,
  "embed_dim": 768,
  "mrl_dims": [
    768,
    512,
    256,
    128,
    64
  ],
  "pooling": "last_token",
  "torch_dtype": "float16",
  "num_experts": 4,
  "num_experts_per_tok": 1,
  "moe_intermediate_size": 2432,
  "norm_topk_prob": true
}