config.json

{
  "architectures": [
    "BlastModelForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "auto_map": {
    "AutoConfig": "modeling_blast.BlastLlamaConfig",
    "AutoModelForCausalLM": "modeling_blast.BlastModelForCausalLM"
  },
  "blast_num_blocks": [
    16
  ],
  "blast_rank": {
    "down_proj": 1488,
    "gate_proj": 1488,
    "k_proj": 1024,
    "o_proj": 1024,
    "q_proj": 1024,
    "up_proj": 1488,
    "v_proj": 1024
  },
  "bos_token_id": 1,
  "eos_token_id": 2,
  "hidden_act": "silu",
  "hidden_size": 4096,
  "indices": [
    0,
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    9,
    10,
    11,
    12,
    13,
    14,
    15,
    16,
    17,
    18,
    19,
    20,
    21,
    22,
    23,
    24,
    25,
    26,
    27,
    28,
    29,
    30,
    31
  ],
  "initializer_range": 0.02,
  "intermediate_size": 11008,
  "max_position_embeddings": 2048,
  "max_sequence_length": 2048,
  "mlp_bias": false,
  "model_type": "blast_llama",
  "num_attention_heads": 32,
  "num_hidden_layers": 32,
  "num_key_value_heads": 32,
  "pad_token_id": 0,
  "precompute_matrix": true,
  "pretraining_tp": 1,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 10000.0,
  "target_modules": [
    "q_proj",
    "k_proj",
    "v_proj",
    "o_proj",
    "gate_proj",
    "up_proj",
    "down_proj"
  ],
  "tie_word_embeddings": false,
  "torch_dtype": "float16",
  "transformers_version": "4.44.2",
  "use_cache": true,
  "vocab_size": 32000
}