config.json

{
  "_name_or_path": "nvidia/prompt-task-and-complexity-classifier",
  "architectures": [
    "DebertaV2Model"
  ],
  "attention_probs_dropout_prob": 0.1,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.1,
  "hidden_size": 768,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_norm_eps": 1e-07,
  "max_position_embeddings": 512,
  "model_type": "deberta-v2",
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "pad_token_id": 0,
  "pooler_dropout": 0.0,
  "pooler_hidden_act": "gelu",
  "pooler_hidden_size": 768,
  "position_biased_input": false,
  "position_buckets": 256,
  "relative_attention": true,
  "torch_dtype": "float32",
  "transformers_version": "4.21.0",
  "type_vocab_size": 0,
  "vocab_size": 128100,
  "target_sizes": {
    "task_type": 8,
    "creativity_scope": 5,
    "reasoning": 5,
    "contextual_knowledge": 5,
    "number_of_few_shots": 5,
    "domain_knowledge": 5,
    "no_label_reason": 5,
    "constraint_ct": 5
  },
  "task_type_map": {
    "0": "Open QA",
    "1": "Closed QA",
    "2": "Summarization",
    "3": "Text Generation",
    "4": "Code Generation",
    "5": "Chatbot",
    "6": "Classification",
    "7": "Rewrite",
    "8": "Brainstorming",
    "9": "Extraction",
    "10": "Other"
  },
  "weights_map": {
    "creativity_scope": [0.0, 0.25, 0.5, 0.75, 1.0],
    "reasoning": [0.0, 0.25, 0.5, 0.75, 1.0],
    "contextual_knowledge": [0.0, 0.25, 0.5, 0.75, 1.0],
    "number_of_few_shots": [0.0, 1.0, 2.0, 3.0, 4.0],
    "domain_knowledge": [0.0, 0.25, 0.5, 0.75, 1.0],
    "no_label_reason": [0.0, 0.25, 0.5, 0.75, 1.0],
    "constraint_ct": [0.0, 0.25, 0.5, 0.75, 1.0]
  },
  "divisor_map": {
    "creativity_scope": 1.0,
    "reasoning": 1.0,
    "contextual_knowledge": 1.0,
    "number_of_few_shots": 1.0,
    "domain_knowledge": 1.0,
    "no_label_reason": 1.0,
    "constraint_ct": 1.0
  },
  "quantized": true,
  "quantization_method": "dynamic",
  "framework": "onnx",
  "optimized_for": "cpu",
  "file_name": "model_quantized.onnx",
  "quantization_config": {
    "format": "QOperator",
    "mode": "IntegerOps",
    "activations_dtype": "QUInt8",
    "weights_dtype": "QInt8",
    "is_static": false
  }
}