LocalAI Self-hosted Citizen Developer

LocalAI Self-hosted

LocalAI Self-hosted Citizen Developer LLM Open Source OpenAI API Compatible Privacy GGUF Hugging Face CPU GPU Docker Model Selection Chatbot Automation

LocalAI Setup

# === LocalAI Installation ===

# Docker — Quick Start
# docker run -p 8080:8080 localai/localai:latest

# Docker Compose — Production
# version: '3.8'
# services:
#   localai:
#     image: localai/localai:latest-aio-cpu
#     ports:
#       - "8080:8080"
#     volumes:
#       - ./models:/build/models
#       - ./config:/build/config
#     environment:
#       - THREADS=8
#       - CONTEXT_SIZE=4096
#       - DEBUG=false
#     restart: unless-stopped

# Download Models
# curl -L "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf" \
#   -o models/mistral-7b.gguf

# Model Config — models/mistral.yaml
# name: mistral
# backend: llama-cpp
# parameters:
#   model: mistral-7b.gguf
#   temperature: 0.7
#   top_p: 0.9
#   top_k: 40
#   context_size: 4096
#   threads: 8

# Test API — OpenAI Compatible
# curl http://localhost:8080/v1/chat/completions \
#   -H "Content-Type: application/json" \
#   -d '{
#     "model": "mistral",
#     "messages": [{"role": "user", "content": "Hello!"}],
#     "temperature": 0.7
#   }'

# Python — Use OpenAI SDK
# pip install openai
# from openai import OpenAI
#
# client = OpenAI(base_url="http://localhost:8080/v1", api_key="not-needed")
#
# response = client.chat.completions.create(
#     model="mistral",
#     messages=[
#         {"role": "system", "content": "You are a helpful assistant."},
#         {"role": "user", "content": "อธิบาย Python list comprehension"},
#     ],
#     temperature=0.7,
#     max_tokens=1000,
# )
# print(response.choices[0].message.content)

from dataclasses import dataclass

@dataclass
class ModelChoice:
    model: str
    size: str
    ram_required: str
    speed: str
    quality: str
    use_case: str

models = [
    ModelChoice("Phi-3 Mini", "2.3GB", "4GB", "เร็วมาก", "ดี", "Chat Code"),
    ModelChoice("Mistral 7B", "4.1GB", "8GB", "เร็ว", "ดีมาก", "General Purpose"),
    ModelChoice("Llama 3 8B", "4.7GB", "8GB", "เร็ว", "ดีมาก", "General Purpose"),
    ModelChoice("CodeLlama 7B", "4.1GB", "8GB", "เร็ว", "ดีมาก (Code)", "Code Generation"),
    ModelChoice("Mixtral 8x7B", "26GB", "32GB", "ปานกลาง", "ดีเยี่ยม", "Complex Tasks"),
]

print("=== Model Choices ===")
for m in models:
    print(f"  [{m.model}] Size: {m.size} | RAM: {m.ram_required}")
    print(f"    Speed: {m.speed} | Quality: {m.quality}")
    print(f"    Use Case: {m.use_case}")

Citizen Developer Workflows

# === AI Workflows for Citizen Developers ===

# Workflow 1: Document Summarizer
# from openai import OpenAI
# client = OpenAI(base_url="http://localhost:8080/v1", api_key="x")
#
# def summarize_document(text, max_words=200):
#     response = client.chat.completions.create(
#         model="mistral",
#         messages=[
#             {"role": "system", "content": f"Summarize in {max_words} words. Thai language."},
#             {"role": "user", "content": text},
#         ],
#     )
#     return response.choices[0].message.content

# Workflow 2: Email Draft Generator
# def draft_email(topic, tone="professional"):
#     response = client.chat.completions.create(
#         model="mistral",
#         messages=[
#             {"role": "system", "content": f"Draft a {tone} email in Thai."},
#             {"role": "user", "content": f"Topic: {topic}"},
#         ],
#     )
#     return response.choices[0].message.content

# Workflow 3: Data Analysis Helper
# def analyze_data(data_description, question):
#     response = client.chat.completions.create(
#         model="mistral",
#         messages=[
#             {"role": "system", "content": "You are a data analyst. Give Python code."},
#             {"role": "user", "content": f"Data: {data_description}\nQuestion: {question}"},
#         ],
#     )
#     return response.choices[0].message.content

@dataclass
class Workflow:
    name: str
    input_type: str
    output: str
    model: str
    time: str
    difficulty: str

workflows = [
    Workflow("Document Summary", "Text/PDF", "สรุปย่อ", "Mistral 7B", "5-15 sec", "ง่าย"),
    Workflow("Email Draft", "Topic + Tone", "Draft Email", "Mistral 7B", "5-10 sec", "ง่าย"),
    Workflow("Code Generator", "Description", "Python Code", "CodeLlama 7B", "10-30 sec", "ง่าย"),
    Workflow("Data Analysis", "Data + Question", "Analysis + Code", "Mistral 7B", "10-20 sec", "ปานกลาง"),
    Workflow("Chatbot", "User Message", "AI Response", "Llama 3 8B", "3-10 sec", "ปานกลาง"),
    Workflow("Translation", "Text + Language", "Translated Text", "Mistral 7B", "5-15 sec", "ง่าย"),
]

print("\n=== Citizen Developer Workflows ===")
for w in workflows:
    print(f"  [{w.name}] Difficulty: {w.difficulty}")
    print(f"    Input: {w.input_type} → Output: {w.output}")
    print(f"    Model: {w.model} | Time: {w.time}")

Production Considerations

# === Production LocalAI Setup ===

# Hardware Requirements
@dataclass
class HardwareReq:
    model_size: str
    ram: str
    cpu: str
    gpu: str
    users: str
    cost_estimate: str

requirements = [
    HardwareReq("3B (Phi-3)", "8GB", "4 cores", "ไม่จำเป็น", "1-5", "$500 PC"),
    HardwareReq("7B (Mistral)", "16GB", "8 cores", "แนะนำ", "5-20", "$1,000 PC"),
    HardwareReq("13B (Llama)", "32GB", "8+ cores", "แนะนำ", "10-50", "$2,000 PC"),
    HardwareReq("8x7B (Mixtral)", "64GB", "16 cores", "จำเป็น", "20-100", "GPU Server"),
]

print("Hardware Requirements:")
for h in requirements:
    print(f"  [{h.model_size}] RAM: {h.ram} | CPU: {h.cpu}")
    print(f"    GPU: {h.gpu} | Users: {h.users} | Cost: {h.cost_estimate}")

# Cost Comparison
comparison = {
    "LocalAI (Mistral 7B)": {"setup": "$1,000 one-time", "monthly": "$50 electricity", "per_1k_tokens": "$0"},
    "OpenAI GPT-4o": {"setup": "$0", "monthly": "Variable", "per_1k_tokens": "$0.005-$0.015"},
    "OpenAI GPT-3.5": {"setup": "$0", "monthly": "Variable", "per_1k_tokens": "$0.0005-$0.0015"},
}

print(f"\n\nCost Comparison (10K requests/month):")
for name, costs in comparison.items():
    print(f"  [{name}]")
    for k, v in costs.items():
        print(f"    {k}: {v}")

เคล็ดลับ

• Start Small: เริ่มจาก Phi-3 Mini ทดลองก่อน แล้วค่อยขยาย
• Q4_K_M: ใช้ Quantization Q4_K_M ดีทั้งคุณภาพและขนาด
• OpenAI SDK: ใช้ OpenAI SDK เปลี่ยนแค่ base_url ง่าย
• Privacy: เหมาะกับข้อมูลลับ ไม่ส่งออกนอกเครื่อง
• Docker: ใช้ Docker Compose สำหรับ Production Setup

LocalAI คืออะไร

Open Source AI Server รันเครื่องตัวเอง OpenAI API Compatible LLM Image Audio Embeddings GGUF Hugging Face CPU GPU Docker Privacy Free

สรุป

LocalAI Self-hosted Citizen Developer LLM OpenAI API Compatible Privacy GGUF Model Docker CPU GPU Chatbot Automation Low-code Production Workflow