README.md

metadata

license: mit
tags:
  - int8
  - Intel® Neural Compressor
  - neural-compressor
  - PostTrainingDynamic
datasets:
  - cnn_dailymail
metrics:
  - rougeLsum

INT8 DistilBart finetuned on CNN DailyMail

Post-training dynamic quantization

This is an INT8 PyTorch model quantized with huggingface/optimum-intel through the usage of Intel® Neural Compressor.

The original fp32 model comes from the fine-tuned model sysresearch101/t5-large-finetuned-xsum-cnn.

Below linear modules are fallbacked to fp32 for less than 1% relative accuracy loss:

Evaluation result

	INT8	FP32
Accuracy (eval-rougeLsum)	41.4707	41.8117
Model size	722M	1249M

Load with optimum:

from optimum.intel.neural_compressor.quantization import IncQuantizedModelForSeq2SeqLM
int8_model = IncQuantizedModelForSeq2SeqLM.from_pretrained(
    'Intel/bart-large-cnn-int8-dynamic',
)