Fortytwo Node Operator FOR Inference with CPU Mode

This repository contains installation instructions and helper scripts for setting up the FortyTwo Network console application on Linux systems

1. Get Realtime Node by Telegram Bot

Use tool-bot monitoring FortyTwo node operator in performs swarm model inference! Get a detail dashboard ranking, received token, daily report & tracking any transaction FOR 42Tokens in participant network

Got bot here... https://greyscope.xyz/x/fortytwo

Key Bot Features:

• Get real-time performs using etherscan APIs
• Get binding wallet maximum 2 EOA wallet address
• Get amount balances in MONAD, FOR 42Token and recent activity
• Get dashboard insights : Rank, Name, Win Rate, Rounds Won, Activity and Rewards
• Get notifications received FOR 42Token transactions
• Get notifications summary FOR daily accumulated token

2. Installing Setup Node

Update Services & Depedency

sudo apt update && sudo apt upgrade -y &&
sudo apt install -y \
automake autoconf build-essential clang curl \
gcc git htop iptables jq libatomic1 libblas3 libclang-dev \
libgbm1 liblapack3 liblapack-dev libleveldb-dev libomp-dev \
libopenblas-dev libgomp1 libopenmpi-dev libssl-dev lz4 make nano \
ncdu ninja-build nvme-cli pkg-config \
python3-pip screen tar tmux unzip wget

Create Folder

mkdir -p ~/Fortytwo && cd ~/Fortytwo

Download Execute by Officially

curl -L -o fortytwo-console-app.zip https://github.com/Fortytwo-Network/fortytwo-console-app/archive/refs/heads/main.zip
unzip fortytwo-console-app.zip
cd fortytwo-console-app-main
# will delete official script linux.sh
rm linux.sh

Use P2P a Peers Network (optional)

wget https://raw.githubusercontent.com/arcxteam/fortytwo-node/refs/heads/main/.p2p_known_peers.json

Redirect Installation Script or Cloning Repo

wget https://raw.githubusercontent.com/arcxteam/fortytwo-node/refs/heads/main/linux.sh

Create a screen & Running Operator

screen -S fortytwo

chmod +x linux.sh && ./linux.sh

• Test with default MODEL NO.22 Qwen3 1.7B Q4 but I'm recommended use custom model
• Back to main root ~/ Ctrl + A + D

# back to screen log
screen -r fortytwo

Note; After installation, the FortyTwo console application be ready run. If not yet received 42T after a few minutes or hourly.. check your logs, if get like here

• INFO Request 41d89c4b5b394015179749b91b525a75d6327cb049f5aa3239caa2dd3dae569d has too short deadline to fit. Required speed: 0.001164882414882415. Max: 0.001
• INFO Remaining join duration for Inference Join state: 4982 ms
• INFO Request d06dc0566a9fac84ab6539b7554f6605c593962fe2365cbc84918efbf06a4b11 has too short deadline to fit. Required speed: 0.001233404909875498. Max: 0.001

Solutions

• The VPServer need faster for complete any Task Inference, RAM 2GB-8GB so cool
• Need check activate on AVX2/AVX or use another low-parameter LLModel visit Huggingface
• A list custom-parameter GGUF LLModels for Text, Code & Math https://arcxteam.github.io/fortytwo-node/llmodel.html

---

Error Cause & Custom LLModel

The script downloaded FortytwoCapsule-linux-amd64-cuda124 (GPU version), which requires CUDA libraries libcuda.so.1. On CPU-only servers without NVIDIA drivers, this library is missing, causing the load failure.

Why It Happened

• Got it libcuda.so.1 redownload capsule & got like this rust_backtrace=1/full/0 try another model this issue by fortytwo capsule llm_model.rs:315 llama.cpp
• Script detects no nvidia-smi and selects CPU binary (correct)
• But download URL appends -cuda124 for GPU, even in "CPU" branch—likely a script bug (non-existent CPU file defaults to GPU)

Fix

1. Manual Download Capsule & Replace:

   cd ~/Fortytwo/fortytwo-console-app-main/FortytwoNode
   rm -f FortytwoCapsule # remove&redownload
   wget "https://fortytwo-network-public.s3.us-east-2.amazonaws.com/capsule/v$(curl -s https://fortytwo-network-public.s3.us-east-2.amazonaws.com/capsule/latest)/FortytwoCapsule-linux-amd64" -O FortytwoCapsule
   chmod +x FortytwoCapsule

2. Rerun Script:

   cd ~/Fortytwo/fortytwo-console-app-main
   rm linux.sh
   wget https://raw.githubusercontent.com/arcxteam/fortytwo-node/refs/heads/main/linux.sh
   chmod +x linux.sh && ./linux.sh

3. Another Custom Model GGUF

   • Costum model by import plz select 1
   • Need use LLM_HF_REPO and LLM_HF_MODEL_NAME
   • Visit Huggingface
   • Check repo file model need a parameter key is GGUF & llama.cpp
   • Select low LLModel medium at 1GB-3GB file and use token 0.5B-3B with quantity param Q3_K_M or Q4_K_M or Q5_K_M
   • A list custom-parameter GGUF LLModels for Text, Code & Math https://arcxteam.github.io/fortytwo-node/llmodel.html

   Example my custom LLModel

   LLM_HF_REPO="unsloth/DeepSeek-R1-Distill-Qwen-1.5B-GGUF"
   LLM_HF_MODEL_NAME="DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M.gguf"
   NODE_NAME="⬢ MATH EQUATIONS & REASONING: DeepSeek-R1-Distill-Qwen-1.5B Q4"

   

Notes

• Official docs require NVIDIA GPU; CPU support may be limited/slower.
• If fails again, edit script: Remove +="-cuda124" in download sections.