{
  "_name_or_path": "/tmp/tmpmsy21owk",
  "alibi": true,
  "apply_residual_connection_post_layernorm": false,
  "architectures": [
    "FalconForCausalLM"
  ],
  "attention_dropout": 0.0,
  "auto_map": {
    "AutoConfig": "configuration_falcon.FalconConfig",
    "AutoModel": "tiiuae/falcon-rw-7b--modeling_falcon.FalconModel",
    "AutoModelForCausalLM": "modeling_falcon.FalconForCausalLM",
    "AutoModelForQuestionAnswering": "tiiuae/falcon-rw-7b--modeling_falcon.FalconForQuestionAnswering",
    "AutoModelForSequenceClassification": "tiiuae/falcon-rw-7b--modeling_falcon.FalconForSequenceClassification",
    "AutoModelForTokenClassification": "tiiuae/falcon-rw-7b--modeling_falcon.FalconForTokenClassification"
  },
  "bias": true,
  "bos_token_id": 1,
  "eos_token_id": 2,
  "hidden_dropout": 0.0,
  "hidden_size": 4096,
  "initializer_range": 0.02,
  "layer_norm_epsilon": 1e-05,
  "model_type": "falcon",
  "multi_query": false,
  "new_decoder_architecture": false,
  "num_attention_heads": 64,
  "num_hidden_layers": 36,
  "num_kv_heads": 64,
  "parallel_attn": false,
  "quantization_config": {
    "bnb_4bit_compute_dtype": "bfloat16",
    "bnb_4bit_quant_type": "fp4",
    "bnb_4bit_use_double_quant": true,
    "llm_int8_enable_fp32_cpu_offload": false,
    "llm_int8_has_fp16_weight": false,
    "llm_int8_skip_modules": [
      "lm_head"
    ],
    "llm_int8_threshold": 6.0,
    "load_in_4bit": false,
    "load_in_8bit": true,
    "quant_method": "bitsandbytes"
  },
  "torch_dtype": "float16",
  "transformers_version": "4.37.1",
  "use_cache": true,
  "vocab_size": 65024
}