quantization

Step-by-Step

This document describes the step-by-step instructions to run large language models (LLMs) on 4th Gen Intel® Xeon® Scalable Processor (codenamed Sapphire Rapids) with PyTorch and Intel® Extension for PyTorch.

The scripts run_clm.py, run_mlm.py and run_plm.py provide three quantization approaches respectively (dynamic, static, qat) based on Intel® Neural Compressor and return last token prediction accuracy by trainer.

The large language model quantization is moved to text-generation now.

Prerequisite

1. Create Environment

# Installation
git clone https://github.com/intel/intel-extension-for-transformers.git itrex
cd itrex
pip install -r requirements.txt
pip install -v .
cd examples/huggingface/pytorch/language-modeling/quantization
pip install -r requirements.txt

Run

Here is how to run the scripts:

Causal Language Modeling (CLM)

python run_clm.py \
    --model_name_or_path EleutherAI/gpt-neo-125M \
    --dataset_name wikitext \
    --dataset_config_name wikitext-2-raw-v1 \
    --tune \
    --quantization_approach static \
    --do_train \
    --do_eval \
    --output_dir ./tmp/clm_output \
    --overwrite_output_dir

Masked Language Modeling (MLM)

python run_mlm.py \
    --model_name_or_path bert-base-uncased \
    --dataset_name wikitext \
    --dataset_config_name wikitext-2-raw-v1 \
    --tune \
    --quantization_approach static \
    --do_train \
    --do_eval \
    --output_dir ./tmp/mlm_output \
    --overwrite_output_dir

Permutation Language Modeling (PLM)

    python run_plm.py \
    --model_name_or_path xlnet-base-cased \
    --dataset_name wikitext \
    --dataset_config_name wikitext-2-raw-v1 \
    --tune \
    --quantization_approach static \
    --do_train \
    --do_eval \
    --output_dir ./tmp/plm_output \
    --overwrite_output_dir