openvino_config.json

{
  "compression": [
    {
      "algorithm": "movement_sparsity",
      "ignored_scopes": [
        "{re}.*MobileBertEmbeddings.*",
        "{re}.*Bottleneck.*",
        "{re}.*OutputBottleneck.*",
        "{re}.*qa_outputs.*"
      ],
      "params": {
        "enable_structured_masking": true,
        "importance_regularization_factor": 0.065,
        "warmup_end_epoch": 10,
        "warmup_start_epoch": 3
      },
      "sparse_structure_by_scopes": [
        {
          "mode": "block",
          "sparse_factors": [
            16,
            16
          ],
          "target_scopes": "{re}.*MobileBertAttention.*"
        },
        {
          "axis": 0,
          "mode": "per_dim",
          "target_scopes": "{re}.*MobileBertIntermediate.*"
        },
        {
          "axis": 1,
          "mode": "per_dim",
          "target_scopes": "{re}.*MobileBertOutput.*"
        },
        {
          "axis": 1,
          "mode": "per_dim",
          "target_scopes": "{re}.*FFNOutput.*"
        }
      ]
    },
    {
      "algorithm": "quantization",
      "export_to_onnx_standard_ops": false,
      "ignored_scopes": [
        "{re}.*__add___[0-1]",
        "{re}.*__truediv__*"
      ],
      "initializer": {
        "batchnorm_adaptation": {
          "num_bn_adaptation_samples": 16
        },
        "range": {
          "num_init_samples": 128,
          "params": {
            "max_percentile": 99.99,
            "min_percentile": 0.01
          },
          "type": "percentile"
        }
      },
      "overflow_fix": "disable",
      "preset": "mixed",
      "scope_overrides": {
        "activations": {
          "{re}.*matmul_0": {
            "mode": "symmetric"
          }
        }
      }
    }
  ],
  "input_info": [
    {
      "keyword": "input_ids",
      "sample_size": [
        32,
        384
      ],
      "type": "long"
    },
    {
      "keyword": "token_type_ids",
      "sample_size": [
        32,
        384
      ],
      "type": "long"
    },
    {
      "keyword": "attention_mask",
      "sample_size": [
        32,
        384
      ],
      "type": "long"
    }
  ],
  "log_dir": "/nvme2/yujiepan/workspace/jpqd-test/LOGS/optimum-mobilebert-qa/0127_ttev_ftz8ih_epo18lr2e-06teacher0.95cosDecayRestart",
  "optimum_version": "1.6.1",
  "save_onnx_model": false,
  "transformers_version": "4.25.1"
}