通过 Huggingface 类库启动 qwen1.5 模型, 这里用的是 cpu，如果用mps，也就是 M2 的 GPU，推理时很慢而且 GPU 的使用率也不高，按理说0.5b 参数的这个模型应该是很快的。目前使用 cpu 可以启动，也可以进行问答。

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch

model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen1.5-0.5B-Chat",
    device_map = 'auto'
)
model.to("cpu")

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B-Chat")

prompt = "创建 user table, username string, age int"
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to("cpu")

generated_ids = model.generate(
    model_inputs.input_ids,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)