config.json

{
  "_matformer_config_dict": {
    "attention_type": [],
    "bias": false,
    "block_size_for_attention": 128,
    "bos_token_id": 1,
    "compile_flexattn": false,
    "custom_layers": {},
    "decoder": null,
    "default_layer": {
      "attn_impl": "flash",
      "ffn_activation": "swiglu",
      "hooks": {},
      "normalization": "rmsnorm",
      "normalization_position": "post",
      "positional_encoding": "alibi",
      "sliding_window_size": null
    },
    "encoder": null,
    "entropy": null,
    "eos_token_id": 2,
    "ffn_factor": 3.0,
    "has_entropy_model": null,
    "has_text_autoencoder": null,
    "hidden_size": 768,
    "is_causal": true,
    "mask_token_id": null,
    "masked_substitution_rate": null,
    "max_position_embeddings": 1024,
    "model_class": null,
    "name": "BabyLM",
    "num_attention_heads": 12,
    "num_hidden_layers": 12,
    "pad_token_id": 0,
    "rms_norm_eps": 1e-06,
    "rope_theta": 10000.0,
    "sliding_type": null,
    "tie_word_embeddings": false,
    "training_objective": "autoregressive",
    "vocab_size": 32777
  },
  "architectures": [
    "MatformerCausalLM"
  ],
  "attention_type": [],
  "bias": false,
  "block_size_for_attention": 128,
  "bos_token_id": 1,
  "compile_flexattn": false,
  "custom_layers": {},
  "decoder": null,
  "default_layer": {
    "attn_impl": "flash",
    "ffn_activation": "swiglu",
    "hooks": {},
    "normalization": "rmsnorm",
    "normalization_position": "post",
    "positional_encoding": "alibi",
    "sliding_window_size": null
  },
  "encoder": null,
  "entropy": null,
  "eos_token_id": 2,
  "ffn_factor": 3.0,
  "has_entropy_model": null,
  "has_text_autoencoder": null,
  "hidden_size": 768,
  "is_causal": true,
  "mask_token_id": null,
  "masked_substitution_rate": null,
  "max_position_embeddings": 1024,
  "model_class": null,
  "model_type": "matformer",
  "name": "BabyLM",
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "pad_token_id": 0,
  "rms_norm_eps": 1e-06,
  "rope_theta": 10000.0,
  "sliding_type": null,
  "torch_dtype": "bfloat16",
  "training_objective": "autoregressive",
  "transformers_version": "4.53.0.dev0",
  "use_cache": true,
  "vocab_size": 32777
}