Ollama & LLM Cheatsheet

A practical reference for running large language models locally with Ollama.

Installation

Quick Install Script

curl -fsSL https://ollama.com/install.sh | sh

APT / Manual Binary

# APT (script adds repo automatically)
curl -fsSL https://ollama.com/install.sh | sh
 
# Manual binary
curl -L https://ollama.com/download/ollama-linux-amd64.tgz | sudo tar -C /usr/local -xz
ollama serve
 
# Docker
docker run -d --gpus all -v ollama:/root/.ollama -p 11434:11434 ollama/ollama
 
# Verify
ollama --version

Model Management

ollama list                              # List downloaded models
ollama pull llama3.2                     # Pull default model
ollama pull llama3.2:1b                  # Specific tag
ollama pull llama3.2:q4_K_M              # Quantized variant
ollama pull mixtral:8x7b                # Mixtral MoE
ollama push myuser/mymodel:tag           # Push to registry
ollama rm llama3.2                       # Remove model
ollama cp llama3.2 my-copy              # Copy locally
ollama show llama3.2                     # Show model details
ollama show --modelfile llama3.2         # Show underlying Modelfile
ollama show --license llama3.2           # License information
ollama show --parameters llama3.2        # Parameters

Running Models

Interactive vs Single Prompt

ollama run llama3.2                      # Interactive chat session
ollama run llama3.2 "Explain quantum computing"  # Single shot
ollama run llama3.2 --verbose            # Show timing stats
echo "What is 2+2?" | ollama run llama3.2  # Pipe input
ollama run llama3.2 --raw "Q:?"          # Skip template

Custom Parameters

ollama run llama3.2 --temperature 0.1    # Deterministic output
ollama run llama3.2 --top-p 0.9          # Nucleus sampling
ollama run llama3.2 --seed 42            # Reproducible
ollama run llama3.2 --num-predict 100    # Max output tokens
ollama run llama3.2 --num-ctx 8192       # Extended context
ollama run llama3.2 --format json        # Request JSON output
ollama run llama3.2 --keep-alive 5m      # Keep in memory 5 min

Interactive Commands

/help                                    # Available commands
/set temperature 0.7                     # Change parameter
/set num_ctx 4096                        # Set context window
/set seed 42                             # Set random seed
/show                                    # Current settings
/load modelname                          # Switch model
/exit                                    # Quit session

Ollama API

List, Generate, Chat

curl http://localhost:11434/api/tags                    # List models
curl http://localhost:11434/api/generate -d '{           # Generate
  "model": "llama3.2", "prompt": "Hello", "stream": false
}'
curl http://localhost:11434/api/chat -d '{               # Chat
  "model": "llama3.2",
  "messages": [{"role": "user", "content": "Hello"}]
}'
curl http://localhost:11434/api/embeddings -d '{         # Embeddings
  "model": "llama3.2", "prompt": "Hello world"
}'

Advanced API with Options

curl http://localhost:11434/api/generate -d '{
  "model": "llama3.2",
  "prompt": "Write a haiku",
  "options": {
    "temperature": 0.8, "top_k": 40, "top_p": 0.95,
    "num_predict": 100, "seed": 42
  },
  "stream": false
}'

Custom Port & CORS

OLLAMA_HOST=0.0.0.0:11435 ollama serve   # Custom port
OLLAMA_ORIGINS="*" ollama serve           # Allow all origins
OLLAMA_ORIGINS="https://app.example.com" ollama serve

Configuration

Environment Variables

export OLLAMA_HOST=0.0.0.0               # Bind all interfaces
export OLLAMA_MODELS=/mnt/data/models    # Custom model dir
export OLLAMA_KEEP_ALIVE=5m              # Idle unload timeout
export OLLAMA_KEEP_ALIVE=-1              # Keep always loaded
export OLLAMA_NUM_PARALLEL=4             # Max concurrent reqs
export OLLAMA_MAX_LOADED_MODELS=2        # Models in memory
export OLLAMA_FLASH_ATTENTION=1          # Flash attention
export OLLAMA_DEBUG=1                    # Debug logging

Custom Modelfile

FROM llama3.2
PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER num_ctx 4096
PARAMETER stop "<|eot_id|>"
SYSTEM You are a helpful AI assistant.
TEMPLATE """
{{ if .System }}<|start_header_id|>system<|end_header_id|>
{{ .System }}<|eot_id|>{{ end }}
<|start_header_id|>user<|end_header_id|>
{{ .Prompt }}<|eot_id|>
<|start_header_id|>assistant<|end_header_id|>
"""

Build

ollama create my-model -f ./Modelfile
ollama create my-model:cust -f Modelfile.custom

Systemd Service

[Unit]
Description=Ollama Service
After=network-online.target
[Service]
ExecStart=/usr/local/bin/ollama serve
Environment=OLLAMA_HOST=0.0.0.0
Environment=OLLAMA_KEEP_ALIVE=5m
Environment=OLLAMA_NUM_PARALLEL=4
Environment=CUDA_VISIBLE_DEVICES=0
Restart=always
RestartSec=3
[Install]
WantedBy=default.target

sudo systemctl daemon-reload && sudo systemctl enable --now ollama.service
journalctl -u ollama -f                  # Follow logs

GPU Configuration

# NVIDIA
export CUDA_VISIBLE_DEVICES=0            # Single GPU
export CUDA_VISIBLE_DEVICES=0,1          # Multiple GPUs
 
# AMD
export ROCR_VISIBLE_DEVICES=0            # AMD GPU select
docker run -d --device=/dev/kfd --device=/dev/dri -v ollama:/root/.ollama ollama/ollama:rocm
 
# CPU only
OLLAMA_NO_GPU=1 ollama serve

Performance Tuning

# Context length
ollama run llama3.2 --num-ctx 32768      # 32K context
# Param in Modelfile: PARAMETER num_ctx 16384
 
# NUMA (multi-socket)
numactl --cpunodebind=0 --membind=0 ollama serve
# Flash attention
OLLAMA_FLASH_ATTENTION=1 ollama serve

Security

# UFW
ufw allow from 192.168.1.0/24 to any port 11434
ufw deny 11434
 
# iptables
iptables -A INPUT -p tcp --dport 11434 -s 127.0.0.1 -j ACCEPT
iptables -A INPUT -p tcp --dport 11434 -j DROP
 
# Rate limiting
iptables -A INPUT -p tcp --dport 11434 -m limit --limit 10/minute -j ACCEPT

Model Quantization

# Quality/size tradeoffs:
ollama pull llama3.2:q2_K                # 2-bit, smallest
ollama pull llama3.2:q3_K_M              # 3-bit medium
ollama pull llama3.2:q4_K_M              # 4-bit, best tradeoff
ollama pull llama3.2:q5_K_M              # 5-bit medium
ollama pull llama3.2:q8_0                # 8-bit, highest quality
ollama pull llama3.2:iq4_xs              # IQ format, improved 4-bit

Troubleshooting

# OOM: use smaller model, shorter context, fewer parallel requests
ollama pull llama3.2:1b                  # Smaller model
OLLAMA_NUM_PARALLEL=1 ollama serve       # Serial processing
 
# Slow inference: enable flash attention, use quantized model
OLLAMA_FLASH_ATTENTION=1 ollama serve
ollama pull llama3.2:q4_K_M
 
# CUDA errors: check driver, restart serve
nvidia-smi                               # Check driver
OLLAMA_DEBUG=1 ollama serve              # Debug mode
 
# Connection refused: verify server running
curl http://localhost:11434/api/tags     # Test endpoint
ss -tlnp | grep 11434                    # Check binding

Integration

OpenAI-Compatible Endpoint

from openai import OpenAI
client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")
resp = client.chat.completions.create(
    model="llama3.2",
    messages=[{"role": "user", "content": "Hello"}]
)

curl http://localhost:11434/v1/chat/completions -d '{
  "model": "llama3.2",
  "messages": [{"role": "user", "content": "Hi"}]
}'

LangChain

from langchain_ollama import ChatOllama
llm = ChatOllama(model="llama3.2", temperature=0.7)
print(llm.invoke("Tell me a joke").content)

Open WebUI

docker run -d -p 3000:8080 \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  ghcr.io/open-webui/open-webui:main
# Access http://localhost:3000

REST Management

curl http://localhost:11434/api/create -d '{"name":"m","modelfile":"FROM llama3.2\nSYSTEM You are helpful."}'
curl http://localhost:11434/api/copy -d '{"source":"llama3.2","destination":"backup"}'
curl -X DELETE http://localhost:11434/api/delete -d '{"name":"backup"}'