{
 "name": "mini-beatrix-2.5s",
 "d_model": 1024,
 "n_layers": 20,
 "n_heads": 16,
 "context": 4096,
 "vocab_size": 256,
 "tokenizer": "byte-trigram",
 "hub_layers": [
  0,
  1,
  2,
  3,
  4,
  5,
  6,
  7,
  8,
  9,
  10,
  11,
  12,
  13,
  14,
  15,
  16,
  17,
  18,
  19
 ],
 "hub_K": 64,
 "hub_D": 128,
 "tau": 0.1,
 "bank_experts": 3,
 "bank_ff": 1024,
 "head_K": 256,
 "head_D": 256,
 "gate_init": -3.0,
 "tie_embeddings": false,
 "hub_chunk": 256,
 "hub_const": 4,
 "hub_ckpt": 0,
 "model_type": "mini-beatrix",
 "architectures": [
  "MiniBeatrixForCausalLM"
 ],
 "auto_map": {
  "AutoConfig": "modeling_minibeatrix.MiniBeatrixConfig",
  "AutoModelForCausalLM": "modeling_minibeatrix.MiniBeatrixForCausalLM"
 },
 "torch_dtype": "float32",
 "use_cache": false,
 "trunk": {
  "craft": "mini-beatrix-2s",
  "step": 61422,
  "base_model_id": "alephllm/mini-beatrix-2s@step61422",
  "note": "the core weights are bit-identical to AbstractPhil/mini-beatrix-2s; 2.5s adds the arm library"
 },
 "default_arm": null
}