AI: 本地模型部署

ollama run llama3

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

sudo systemctl start ollama

# Download Llama 3 8B model (about 4.7 GB)
ollama pull llama3

# Download Qwen2 7B
ollama pull qwen2

# Download Mistral 7B
ollama pull mistral

# Start interactive chat with Llama 3
ollama run llama3

# >>> Hello! Can you explain what a neural network is in one paragraph?
# A neural network is a computational model inspired by the way biological
# neurons work in the human brain. It consists of layers of interconnected
# nodes (neurons) that process information...

# Type /bye to exit
# >>> /bye

# List all downloaded models
ollama list

# Output example:
# NAME            ID              SIZE    MODIFIED
# llama3:latest   365c0bd3c000    4.7 GB  2 hours ago
# mistral:latest  2b3e8435e228    4.1 GB  5 days ago
# qwen2:latest    e3d7e5e5e5e5    4.4 GB  1 day ago

pip install ollama

import ollama

# Simple chat with local Llama 3
response = ollama.chat(
    model="llama3",
    messages=[
        {"role": "user", "content": "Explain quantum computing in one sentence."},
    ],
)

print(response["message"]["content"])

import time
import ollama
from openai import OpenAI

# Local model latency
start = time.time()
local_response = ollama.chat(
    model="llama3",
    messages=[{"role": "user", "content": "What is machine learning?"}],
)
local_time = time.time() - start

# Cloud API latency
client = OpenAI()
start = time.time()
cloud_response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "What is machine learning?"}],
)
cloud_time = time.time() - start

print(f"Local model: {local_time:.2f}s")
print(f"Cloud API:   {cloud_time:.2f}s")
print(f"Local output: {local_response['message']['content'][:100]}")
print(f"Cloud output: {cloud_response.choices[0].message.content[:100]}")

import requests
import json

# Call Ollama REST API directly
url = "http://localhost:11434/api/chat"
payload = {
    "model": "llama3",
    "messages": [{"role": "user", "content": "Write a haiku about programming."}],
    "stream": False,
}

response = requests.post(url, json=payload)
result = response.json()
print(result["message"]["content"])

import ollama
from openai import OpenAI

def local_translate(text: str, source: str = "English", target: str = "Chinese") -> str:
    prompt = (
        f"You are a professional {source}-{target} translator. "
        f"Translate the following {source} text to {target}. "
        f"Only output the translation, no explanation.\n\n{text}"
    )
    response = ollama.chat(
        model="llama3",
        messages=[{"role": "user", "content": prompt}],
    )
    return response["message"]["content"]

def cloud_translate(text: str, source: str = "English", target: str = "Chinese") -> str:
    client = OpenAI()
    prompt = (
        f"You are a professional {source}-{target} translator. "
        f"Translate the following {source} text to {target}. "
        f"Only output the translation, no explanation.\n\n{text}"
    )
    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
    )
    return response.choices[0].message.content

test_sentences = [
    "The patient presented with persistent cough and mild fever for three days.",
    "Artificial intelligence is transforming healthcare diagnostics.",
    "The stock market rallied after the central bank announced rate cuts.",
]

print("=" * 60)
print("Local AI Translator vs Cloud API Translator")
print("=" * 60)

for i, sentence in enumerate(test_sentences, 1):
    local_result = local_translate(sentence)
    cloud_result = cloud_translate(sentence)
    print(f"\n[{i}] EN: {sentence}")
    print(f"    Local:  {local_result}")
    print(f"    Cloud:  {cloud_result}")
    print(f"    {'--- Match ---' if local_result.strip() == cloud_result.strip() else '--- Different ---'}")

# Step 1: Install and start Ollama
ollama serve

# Step 2: Download model (in another terminal)
ollama pull llama3

# Step 3: Install Python dependencies
pip install ollama openai

# Step 4: Run the translator
python local_translator.py

============================================================
Local AI Translator vs Cloud API Translator
============================================================

[1] EN: The patient presented with persistent cough and mild fever for three days.
    Local:  Patient presents with persistent cough and mild fever for three days.
    Cloud:  Patient shows persistent cough and mild fever for three days.
    --- Different ---

[2] EN: Artificial intelligence is transforming healthcare diagnostics.
    Local:  AI is transforming medical diagnosis.
    Cloud:  AI is revolutionizing the medical diagnosis field.
    --- Different ---

# Step 1: Install Ollama from https://ollama.com

# Step 2: Download a model
ollama pull llama3

# Step 3: Start a chat
ollama run llama3

# >>> Tell me three interesting facts about the planet Mars.

# Step 4: List installed models
ollama list

import ollama

def interactive_translator():
    print("Local AI Translator (type 'quit' to exit)")
    print("-" * 40)
    while True:
        text = input("EN> ")
        if text.lower() == "quit":
            break
        response = ollama.chat(
            model="llama3",
            messages=[
                {
                    "role": "user",
                    "content": f"Translate to Chinese, only output the translation:\n\n{text}",
                },
            ],
        )
        print(f"CN> {response['message']['content']}\n")

interactive_translator()

import ollama
from openai import OpenAI
import time

PROMPT = """Analyze the following customer review and extract:
1. Sentiment (Positive/Negative/Neutral)
2. Key complaint or praise
3. Suggested action for the company

Review: "The headphones sound amazing but the Bluetooth keeps disconnecting every 10 minutes. I have to restart them constantly. Great audio quality but unusable for calls."
"""

def compare_outputs():
    # Local model
    start = time.time()
    local = ollama.chat(
        model="llama3",
        messages=[{"role": "user", "content": PROMPT}],
    )
    local_time = time.time() - start
    local_text = local["message"]["content"]

    # Cloud API
    client = OpenAI()
    start = time.time()
    cloud = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": PROMPT}],
    )
    cloud_time = time.time() - start
    cloud_text = cloud.choices[0].message.content

    print("PROMPT:", PROMPT[:80], "...")
    print("=" * 60)
    print(f"[Local - {local_time:.1f}s]\n{local_text}")
    print("=" * 60)
    print(f"[Cloud - {cloud_time:.1f}s]\n{cloud_text}")
    print("=" * 60)

    # TODO: Write a brief analysis comparing:
    # - Accuracy of sentiment extraction
    # - Completeness of key points
    # - Actionability of suggestions
    # - Response time difference

compare_outputs()