⚡️ 测试微软 Phi-4-Mini-Instruct 模型：轻量级 AI 任务的利器

pip install transformers torch

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

# 检查是否有 GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")

# 加载模型和 tokenizer
model_id = "microsoft/Phi-4-mini-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    trust_remote_code=True
).to(device)

# 构建对话
messages = [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "用简单的语言解释监督学习和非监督学习的区别。"}
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# 编码并转移到设备
inputs = tokenizer(prompt, return_tensors="pt").to(device)

# 生成回复
output_ids = model.generate(
    **inputs,
    max_new_tokens=900,
    do_sample=False,
    temperature=0.0
)

# 解码并打印输出
response = tokenizer.decode(output_ids[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True)
print("生成的回复:")
print(response)

生成的回复：
Supervised learning is like having a teacher who shows you examples of the answers and then asks you to predict the answers for new questions. You learn from the examples provided, and your goal is to get the answers right. Unsupervised learning, on the other hand, is like exploring a new place without a map. You look at the data and try to find patterns or groupings on your own, without any guidance on what you're looking for.