config.json

{
  "architectures": [
    "ZaryaLMForCausalLM"
  ],
  "attention_dropout": 0.0,
  "attention_scalar_gate": false,
  "bos_token_id": 1,
  "diffusion_loss_proportion": 0.5,
  "dtype": "float32",
  "eos_token_id": 2,
  "gate_multiplier": 2.0,
  "head_dim": 128,
  "hidden_size": 512,
  "intermediate_size": 1440,
  "mask_token_id": 32768,
  "max_position_embeddings": 2048,
  "mlp_scalar_gate": false,
  "model_type": "zarya_lm",
  "noise_eps": 0.001,
  "nope_every_n": null,
  "nope_layers": [],
  "num_attention_heads": 8,
  "num_hidden_layers": 9,
  "num_key_value_heads": 4,
  "pad_token_id": 0,
  "qk_norm": true,
  "rms_norm_eps": 1e-06,
  "rope_theta": 1000000.0,
  "slot_borders": [
    0.06,
    0.2,
    0.4,
    0.6,
    0.8,
    1.0
  ],
  "slot_sizes": [
    2,
    4,
    8,
    16,
    32,
    64
  ],
  "tie_word_embeddings": true,
  "training_objective": "zarya-slotted-pretrain-v1",
  "transformers_version": "5.17.0",
  "vocab_size": 32769,
  "zarya_tokenizer": {
    "base_tokenizer_sha256": "fc71a154536f6c84a5bf729d06324032126603ef5183c0b4979407e4fad68a8d",
    "base_vocab_size": 32768,
    "mask_token": "<|mdm_mask|>",
    "mask_token_id": 32768
  },
  "auto_map": {
    "AutoConfig": "configuration_zarya_lm.ZaryaLMConfig",
    "AutoModelForCausalLM": "modeling_zarya_lm.ZaryaLMForCausalLM"
  },
  "_name_or_path": "d0rj/zarya-51M-base"
}