openvino_config.json

{
  "compression": {
    "algorithm": "quantization",
    "export_to_onnx_standard_ops": false,
    "ignored_scopes": [
      "{re}.*Embedding*",
      "{re}.*__add___*",
      "{re}.*layer_norm_*"
    ],
    "initializer": {
      "batchnorm_adaptation": {
        "num_bn_adaptation_samples": 0
      },
      "range": {
        "num_init_samples": 8,
        "type": "mean_min_max"
      }
    },
    "overflow_fix": "disable",
    "scope_overrides": {
      "activations": {
        "activations": {
          "mode": "symmetric"
        },
        "weights": {
          "mode": "symmetric"
        }
      }
    }
  },
  "input_info": [
    {
      "keyword": "input_ids",
      "sample_size": [
        8,
        1024
      ],
      "type": "long"
    },
    {
      "keyword": "attention_mask",
      "sample_size": [
        8,
        1024
      ],
      "type": "long"
    }
  ],
  "log_dir": "/data1/vchua/temp/ov-opt-350m-8bit-kv-cache",
  "optimum_version": "1.8.8",
  "save_onnx_model": false,
  "transformers_version": "4.30.2"
}