Implementation:Microsoft LoRA NLG Environment Setup Script

Overview

NLG_Environment_Setup_Script provides the concrete commands for installing Python dependencies and downloading pretrained GPT-2 checkpoints required by the LoRA NLG fine-tuning pipeline. This implementation covers two shell-level entry points: pip install -r requirement.txt and bash download_pretrained_checkpoints.sh.

Type

External Tool Doc

Source

examples/NLG/requirement.txt (lines 1-7)
examples/NLG/download_pretrained_checkpoints.sh (lines 1-11)

Signature

Dependency Installation

pip install -r requirement.txt

The requirement.txt file contains the following pinned dependencies:

--find-links https://download.pytorch.org/whl/torch_stable.html
torch==1.7.1+cu101
transformers==3.3.1
spacy
tqdm
tensorboard
progress

Checkpoint Download

bash download_pretrained_checkpoints.sh

The script performs the following operations:

#!/bin/bash
echo "downloading pretrained model checkpoints..."
mkdir pretrained_checkpoints
cd pretrained_checkpoints
wget https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-pytorch_model.bin
wget https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-medium-pytorch_model.bin
wget https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-large-pytorch_model.bin
cd ..
echo "script complete!"

Input / Output

Direction	Description
Input	A clean Python environment (virtualenv or conda) with `pip` and `wget` available.
Output	Installed Python packages: torch, transformers, spacy, tqdm, tensorboard, progress. Pretrained checkpoint files in `pretrained_checkpoints/`: `gpt2-pytorch_model.bin` (GPT-2 Small, ~500MB) `gpt2-medium-pytorch_model.bin` (GPT-2 Medium, ~1.5GB) `gpt2-large-pytorch_model.bin` (GPT-2 Large, ~3GB)

Notes

The --find-links directive in requirement.txt points pip to the PyTorch wheel index for CUDA 10.1 builds. If using a different CUDA version, modify this line accordingly.
The loralib package is not listed in requirement.txt and must be installed separately, typically via pip install loralib or from the repository root.
The regex package is also required by the BPE encoder (encoder.py) but is not listed in the requirements file. Install it via pip install regex.

Metadata

Field	Value
Source	microsoft/LoRA
Type	External Tool Doc
Last Updated	2026-02-10

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment