Implementation:ContextualAI HALOs Install Sh

Knowledge Sources	ContextualAI HALOs
Domains	Infrastructure, DevOps
Last Updated	2026-02-08 03:00 GMT

Overview

Concrete tool for provisioning the HALOs training environment provided by a shell script.

Description

The install.sh script creates a conda environment named halos with Python 3.10.14 and installs all dependencies required for LLM alignment training, sampling, labeling, and evaluation. It pins exact versions of PyTorch (2.4.0), Flash Attention (2.6.3), Transformers (4.51.3), PEFT (0.12.0), Datasets (2.20.0), Accelerate (0.33.0), and vLLM (0.6.3.post1). It also clones the LM Evaluation Harness and pre-downloads evaluation tasks.

Usage

Run this script on a fresh machine with conda and CUDA 12.1 installed, before executing any training or evaluation workflow.

Code Reference

Source Location

• Repository: ContextualAI/HALOs
• File: install.sh
• Lines: L1-31

Signature

#!/bin/bash

conda create --name halos python=3.10.14
conda activate halos

conda install pip
pip install packaging ninja
conda install pytorch=2.4.0 pytorch-cuda=12.1 -c pytorch -c nvidia
pip install flash-attn==2.6.3 --no-build-isolation
pip install transformers==4.51.3
pip install peft==0.12.0
pip install datasets==2.20.0
pip install accelerate==0.33.0
pip install vllm==0.6.3.post1
pip install alpaca-eval immutabledict langdetect wandb omegaconf openai hydra-core==1.3.2

# lm-eval
git clone --depth 1 https://github.com/EleutherAI/lm-evaluation-harness
cd lm-evaluation-harness
pip install -e .
# download tasks for offline eval
python << EOF
from lm_eval import tasks
task_names = ["winogrande", "mmlu", "gsm8k_cot", "bbh_cot_fewshot", "arc_easy", "arc_challenge", "hellaswag", "ifeval"]
task_dict = tasks.get_task_dict(task_names)

from datasets import load_dataset
load_dataset("tatsu-lab/alpaca_eval", "alpaca_eval")
EOF

Import

bash install.sh

I/O Contract

Inputs

Name	Type	Required	Description
Machine	Hardware	Yes	Fresh machine with conda and CUDA 12.1 installed
Network	Connectivity	Yes	Internet access for downloading packages and datasets

Outputs

Name	Type	Description
halos environment	conda env	Activated Python 3.10.14 environment with all dependencies
lm-evaluation-harness	Git repo	Cloned and installed evaluation framework
Cached tasks	Files	Pre-downloaded benchmark task configs and AlpacaEval dataset

Usage Examples

Standard Installation

# Run from the repository root
bash install.sh

# Verify the environment is active
conda activate halos
python -c "import transformers; print(transformers.__version__)"
# Should print: 4.51.3

Related Pages

Implements Principle

• Principle:ContextualAI_HALOs_Environment_Setup

Requires Environment

• Environment:ContextualAI_HALOs_CUDA_12_1_Training_Environment