{
  "activation": "swiglu",
  "architecture": "MicroLoopForDiffusionLM",
  "architectures": [
    "MicroLoopForDiffusionLM"
  ],
  "attention_implementation": "sdpa",
  "attention_output_gate": false,
  "attn_res_block_size": null,
  "auto_map": {
    "AutoConfig": "configuration_microloop.MicroLoopConfig",
    "AutoModelForCausalLM": "modeling_microloop.MicroLoopForDiffusionLM"
  },
  "bos_token_id": 1,
  "diffusion": {
    "fallback_block_size": 16,
    "objective": "absorbing_mask_mdlm",
    "primary_block_size": 32,
    "training_block_sizes": {
      "16": 0.35,
      "32": 0.6,
      "64": 0.05
    }
  },
  "dropout": 0.0,
  "dtype": "float32",
  "eos_token_id": 2,
  "head_dimension": 40,
  "hidden_size": 240,
  "intermediate_size": 640,
  "is_decoder": true,
  "looping": {
    "layers": [
      4,
      5,
      6
    ],
    "maximum_serving_loops": 3,
    "training_loop_counts": [
      1
    ]
  },
  "max_position_embeddings": 2048,
  "model_type": "microloop_diffusion",
  "mtp_enabled": false,
  "normalization": "rmsnorm",
  "num_attention_heads": 6,
  "num_hidden_layers": 14,
  "num_key_value_heads": 2,
  "pad_token_id": 0,
  "positional_encoding": "rope",
  "qk_norm": "per_head",
  "rms_norm_eps": 1e-05,
  "rope_theta": 10000.0,
  "swiglu_clamp": {
    "enabled": true,
    "gate_max": 10.0,
    "linear_max": 10.0,
    "linear_min": -10.0
  },
  "target_parameters": 10000000,
  "tie_word_embeddings": true,
  "tokenizer": {
    "required_special_tokens": [
      "<pad>",
      "<bos>",
      "<eos>",
      "<mask>",
      "<system>",
      "<user>",
      "<assistant>",
      "<turn_end>",
      "<doc_end>"
    ],
    "type": "byte_level_bpe",
    "vocabulary_size": 4096
  },
  "transformers_version": "5.14.1",
  "use_cache": false,
  "vocab_size": 4096
}