Skip to content

README.md

Latest commit

 

History

History
145 lines (107 loc) · 3.78 KB

README.md

File metadata and controls

145 lines (107 loc) · 3.78 KB
Raw

Setup Scripts

This directory contains setup scripts for different LocoTrainer deployment scenarios.

Available Scripts

1. setup_locotrainer.sh - CLI Only (Recommended for most users)

Installs LocoTrainer CLI with uv for use with external API providers.

Use case: You want to use LocoTrainer with OpenAI, DashScope, OpenRouter, or other cloud APIs.

Requirements: Python 3.10+

Installation:

curl -O https://raw.githubusercontent.com/LocoreMind/LocoTrainer/main/scripts/setup_locotrainer.sh
chmod +x setup_locotrainer.sh
./setup_locotrainer.sh

What it does:

  • Checks/installs uv
  • Creates venv at ~/.venv/locotrainer
  • Installs locotrainer via uv pip install
  • Creates config template at ~/.locotrainer.env

After installation:

# Edit config with your API key
nano ~/.locotrainer.env

# Run
source ~/.venv/locotrainer/bin/activate
export $(cat ~/.locotrainer.env | xargs)
locotrainer run -q "What are the default LoRA settings in ms-swift?"

2. setup_locotrainer_vllm.sh - vLLM + LocoTrainer-4B (For GPU users)

Installs vLLM + LocoTrainer for local deployment of LocoTrainer-4B model.

Use case: You have an NVIDIA GPU (40GB+ VRAM) and want to run LocoTrainer-4B locally at zero API cost.

Requirements:

  • Python 3.10+
  • NVIDIA GPU with 40GB+ VRAM (A100 40GB recommended for 128K context)
  • CUDA 12.1+

Installation:

curl -O https://raw.githubusercontent.com/LocoreMind/LocoTrainer/main/scripts/setup_locotrainer_vllm.sh
chmod +x setup_locotrainer_vllm.sh
./setup_locotrainer_vllm.sh

What it does:

  • Checks for NVIDIA GPU
  • Checks/installs uv
  • Creates venv at ~/.venv/locotrainer
  • Installs vLLM + locotrainer
  • Creates ~/start_vllm.sh startup script
  • Creates config at ~/.locotrainer.env (points to localhost:8080)

After installation:

# Start vLLM server (background)
nohup ~/start_vllm.sh > ~/vllm.log 2>&1 &

# Or use screen (recommended)
screen -S vllm
~/start_vllm.sh
# Press Ctrl+A D to detach

# Wait for model to load (check logs)
tail -f ~/vllm.log

# Run LocoTrainer
source ~/.venv/locotrainer/bin/activate
export $(cat ~/.locotrainer.env | xargs)
locotrainer run -q "How does ms-swift implement GRPO training?"

3. setup_locotrainer_dev.sh - Development Environment

Sets up development environment from source for contributors.

Use case: You want to contribute to LocoTrainer or modify the source code.

Requirements: Python 3.10+

Installation:

git clone https://github.com/LocoreMind/LocoTrainer.git
cd LocoTrainer
./scripts/setup_locotrainer_dev.sh

What it does:

  • Checks/installs uv
  • Runs uv sync to install dependencies
  • Creates .env from .env.example

After installation:

# Edit config
nano .env

# Run from source
uv run locotrainer run -q "Your question"

Quick Comparison

Script Install Location Use Case GPU Required
setup_locotrainer.sh ~/.venv/locotrainer Cloud API users No
setup_locotrainer_vllm.sh ~/.venv/locotrainer Local model deployment Yes (40GB+)
setup_locotrainer_dev.sh ./LocoTrainer/.venv Development/contribution No

Troubleshooting

Script fails with "uv: command not found"

  • The script should auto-install uv, but if it fails, manually install: 
    curl -LsSf https://astral.sh/uv/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"

vLLM script fails with CUDA errors

  • Verify CUDA version: nvidia-smi
  • vLLM requires CUDA 12.1+
  • Check GPU memory: nvidia-smi --query-gpu=memory.total --format=csv

"locotrainer: command not found" after installation

  • Activate the venv first: source ~/.venv/locotrainer/bin/activate
  • Or use full path: ~/.venv/locotrainer/bin/locotrainer