metadata
language:
- en
license: apache-2.0
tags:
- multiple-choice
- int8
- PostTrainingStatic
datasets:
- swag
metrics:
- accuracy
model-index:
- name: bert-base-uncased-finetuned-swag-int8-static
results:
- task:
name: Multiple-choice
type: multiple-choice
dataset:
name: Swag
type: swag
metrics:
- name: Accuracy
type: accuracy
value: 0.7838148474693298
INT8 bert-base-uncased-finetuned-swag
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 thyagosme/bert-base-uncased-finetuned-swag.
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) | 16.55 | 9.333 |
Accuracy (eval-accuracy) | 0.7838 | 0.7915 |
Model size (MB) | 133 | 418 |
Load with Intel® Neural Compressor (build from source):
from neural_compressor.utils.load_huggingface import OptimizedModel
int8_model = OptimizedModel.from_pretrained(
'Intel/bert-base-uncased-finetuned-swag-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.