01 — Installation¶

This guide walks you through installing LLM Fine-Tuner on your computer. There are four options depending on your setup.

Option A — Automatic Install (Recommended for Linux / macOS)¶

This is the easiest method. One command does everything: creates a virtual environment, detects your GPU, installs the right version of PyTorch, and sets up a launcher shortcut.

# 1. Download the project
git clone https://github.com/Yog-Sotho/LLM-fine-tuner.git
cd LLM-fine-tuner

# 2. Run the installer
chmod +x install.sh
./install.sh

The installer will ask a few yes/no questions about optional components. If you want to skip all prompts and accept everything automatically:

./install.sh --yes
# or
AUTO_INSTALL=true ./install.sh

What it installs: - Python virtual environment (isolated, won't affect your system) - PyTorch (GPU version if NVIDIA detected, CPU version otherwise) - All core dependencies (transformers, gradio, peft, datasets, trl, accelerate…) - Flash Attention 2 (if your GPU supports it — big speed boost) - Optional: Unsloth, vLLM, evaluation tools, data tools

After installation, launch the app with:

source llm_finetuner_env/bin/activate
llm-finetune

Option B — Manual Install (Windows or any OS)¶

Use this if the automatic installer doesn't work on your system.

Step 1 — Install Python¶

Download Python 3.11 from python.org. During installation on Windows, check "Add Python to PATH".

Verify it worked:

python --version
# Should print: Python 3.11.x

Step 2 — Download the project¶

git clone https://github.com/Yog-Sotho/LLM-fine-tuner.git
cd LLM-fine-tuner

If you don't have git, download the ZIP from GitHub (green "Code" button → "Download ZIP") and unzip it.

Step 3 — Create a virtual environment¶

A virtual environment keeps the project's packages separate from the rest of your computer.

# Create the environment
python -m venv llm_finetuner_env

# Activate it (Linux / macOS)
source llm_finetuner_env/bin/activate

# Activate it (Windows)
llm_finetuner_env\Scripts\activate

You'll see (llm_finetuner_env) at the start of your terminal prompt. That means it's active.

Step 4 — Install PyTorch¶

If you have an NVIDIA GPU (check with nvidia-smi in terminal):

# CUDA 12.x (most modern NVIDIA cards)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

If nvidia-smi reports CUDA 11.x, update the NVIDIA driver first: PyTorch 2.13 and newer (which the app needs) ship no CUDA 11 build. Any current driver supports CUDA 12, RTX 2000 series included.

If you have no GPU or an AMD/Apple GPU:

pip install torch torchvision torchaudio

torchvision is needed for vision-language (image + text) fine-tuning. It must be the build that matches your torch, so install both in one command as above. Adding it later: pip install -e ".[vision]" from the repo folder works for CPU installs; with CUDA wheels, install it from the same PyTorch index as torch.

Step 5 — Install all other dependencies¶

pip install -r requirements.txt

This includes the optional extras (evaluation metrics, augmentation, embedding-model training). Installing the package instead? Pick extras by name, e.g. pip install -e ".[eval,embedding]" or ".[all]".

Step 6 — Install Unsloth (highly recommended)¶

Unsloth makes training 2–5× faster and uses 60–80% less VRAM. It's optional but strongly recommended if you have an NVIDIA GPU.

pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git" --no-deps

Step 7 — Launch¶

python main.py

Open http://localhost:7860 in your browser.

Option C — Google Colab (No GPU required on your machine)¶

Google Colab gives you free GPU access in your browser. Perfect if your computer is too slow.

Go to colab.research.google.com
Create a new notebook
In the first cell, paste and run:

# Clone the repo
!git clone https://github.com/Yog-Sotho/LLM-fine-tuner.git
%cd LLM-fine-tuner

# Install dependencies
!pip install -r requirements.txt -q
!pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git" --no-deps -q

# Launch (SHARE=true gives you a public URL)
import os, subprocess
proc = subprocess.Popen(["python", "main.py"], env={**os.environ, "SHARE": "true"})

A public URL will appear (e.g. https://xxxxx.gradio.live) — open it in your browser.

Tip: In Colab, go to Runtime → Change runtime type → T4 GPU before running for much faster training.

Option D — Docker¶

# Build the image
docker build -t llm-fine-tuner .

# Run it
docker run -p 7860:7860 --gpus all llm-fine-tuner

Then open http://localhost:7860 in your browser.

Note: --gpus all requires the NVIDIA Container Toolkit. Remove it for CPU-only mode.

Verifying the Installation¶

After launching, the top of the UI shows a hardware summary like:

🖥️ GPU: NVIDIA RTX 4090 — 24.0 GB VRAM
💾 RAM: 64.0 GB
🔥 PyTorch: 2.14.1+cu126
✅ Unsloth: available
✅ Flash Attention 2: available

If you see your GPU listed, everything is working correctly.

Updating¶

To get the latest version:

cd LLM-fine-tuner
git pull
pip install -r requirements.txt --upgrade

Next Step¶

→ 02 — Quick Start: Train your first model in 5 minutes.