我想要一天分享一點「LLM從底層堆疊的技術」,並且每篇文章長度控制在三分鐘以內,讓大家不會壓力太大,但是又能夠每天成長一點。
先列出目前擁有的材料:
我們自己微調了 GPT,那如果想看看原生 GPT4 的性能可以使用以下程式:
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(model = "gpt-4",
messages = [{"role": "user",
"content": ""},
{"role": "assistant",
"content": "Explain why space and time are important from Kant's point of view"},
{"role": "assistant",
"content": ""}],
temperature = 0.03,
max_tokens = 256,
top_p = 1,
frequency_penalty = 0,
presence_penalty = 0)