README.md

metadata

language:
  - en
license: apache-2.0
tags:
  - token-classfication
  - int8
  - PostTrainingStatic
datasets:
  - conll2003
metrics:
  - accuracy
model-index:
  - name: distilbert-base-uncased-finetuned-conll03-english-int8-static
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: Conll2003
          type: conll2003
        metrics:
          - name: Accuracy
            type: accuracy
            value: 0.9858650364082395

INT8 distilbert-base-uncased-finetuned-conll03-english

Post-training static quantization

This is an INT8 PyTorch model quantized with Intel® Neural Compressor.

The original fp32 model comes from the fine-tuned model elastic/distilbert-base-uncased-finetuned-conll03-english.

The calibration dataloader is the train dataloader. The default calibration sampling size 100 isn't divisible exactly by batch size 8, so the real sampling size is 104.

Test result

• Batch size = 8
• Amazon Web Services c6i.xlarge (Intel ICE Lake: 4 vCPUs, 8g Memory) instance.

	INT8	FP32
Throughput (samples/sec)	148.144	77.108
Accuracy (eval-accuracy)	0.9859	0.9882
Model size (MB)	64.5	253

Load with Intel® Neural Compressor (build from source):

from neural_compressor.utils.load_huggingface import OptimizedModel
int8_model = OptimizedModel.from_pretrained(
    'Intel/distilbert-base-uncased-finetuned-conll03-english-int8-static',
)

Notes:

• The INT8 model has better performance than the FP32 model when the CPU is fully occupied. Otherwise, there will be the illusion that INT8 is inferior to FP32.