获取GPT API KEY
在环境声明OPENAI_API_KEY
export OPENAI_API_KEY='KEY'
source ./.zshrc
下载对应package
pip install wandb openai
下面是demo_gpt.py
from openai import OpenAI
import textwrap
import os
import wandb
client = OpenAI()
# set the API key
gpt_assistant_prompt = "You are a " + input("Who should I be, as I answer your prompt?")
gpt_user_prompt = input("What prompt do you want me to do?")
gpt_prompt = gpt_assistant_prompt, gpt_user_prompt
print(gpt_prompt)
message=[{"role": "assistant", "content": gpt_assistant_prompt}, {"role": "user", "content": gpt_user_prompt}]
temperature=0.2
max_tokens=256
frequency_penalty=0.0
response = client.chat.completions.create(
model="gpt-4",
messages = message,
temperature=temperature,
max_tokens=max_tokens,
frequency_penalty=frequency_penalty
)
print(response.choices[0].message)
wandb.init()
response_text = response.choices[0].message.content
tokens_used = response.usage.total_tokens
prediction_table = wandb.Table(columns=["Prompt", "Response", "Tokens", "Max Tokens", "Frequency Penalty", "Temperature"])
prediction_table.add_data(gpt_prompt, response_text, tokens_used, max_tokens, frequency_penalty, temperature)
wandb.log({'predictions': prediction_table})
wandb.finish()
