基于这篇文章，我们现在可以将 gemini 与 openai 库一起使用。所以，我决定在这篇文章中尝试一下
目前仅提供聊天完成 api 和嵌入 api。
在本文中，我尝试使用 python 和 javascript。
python
首先，我们来搭建环境。
pip install openai python-dotenv
接下来，让我们运行以下代码。
import os
from dotenv import load_dotenv
from openai import openai
load_dotenv()
google_api_key = os.getenv("google_api_key")
client = openai(
api_key=google_api_key,
base_url="https://generativelanguage.googleapis.com/v1beta/"
)
response = client.chat.completions.create(
model="gemini-1.5-flash",
n=1,
messages=[
{"role": "system", "content": "you are a helpful assistant."},
{
"role": "user",
"content": "explain briefly(less than 30 words) to me how ai works."
}
]
)
print(response.choices[0].message.content)
返回了以下响应。
ai mimics human intelligence by learning patterns from data, using algorithms to solve problems and make decisions.
在内容字段中，您可以指定字符串或“类型”：“文本”。
import os
from dotenv import load_dotenv
from openai import openai
load_dotenv()
google_api_key = os.getenv("google_api_key")
client = openai(
api_key=google_api_key,
base_url="https://generativelanguage.googleapis.com/v1beta/"
)
response = client.chat.completions.create(
model="gemini-1.5-flash",
n=1,
messages=[
{"role": "system", "content": "you are a helpful assistant."},
{
"role": "user",
"content": [
{
"type": "text",
"text": "explain briefly(less than 30 words) to me how ai works.",
},
]
}
]
)
print(response.choices[0].message.content)
但是，图像和音频输入出现错误。
图像输入示例代码
import os
from dotenv import load_dotenv
from openai import openai
load_dotenv()
google_api_key = os.getenv("google_api_key")
client = openai(
api_key=google_api_key,
base_url="https://generativelanguage.googleapis.com/v1beta/"
)
# png to base64 text
import base64
with open("test.png", "rb") as image:
b64str = base64.b64encode(image.read()).decode("utf-8")
response = client.chat.completions.create(
model="gemini-1.5-flash",
# model="gpt-4o",
n=1,
messages=[
{"role": "system", "content": "you are a helpful assistant."},
{
"role": "user",
"content": [
{
"type": "text",
"text": "describe the image in the image below.",
},
{
"type": "image_url",
"image_url": {
"url": f"data:image/png;base64,{b64str}"
}
}
]
}
]
)
print(response.choices[0].message.content)
音频输入示例代码
import os
from dotenv import load_dotenv
from openai import openai
load_dotenv()
google_api_key = os.getenv("google_api_key")
client = openai(
api_key=google_api_key,
base_url="https://generativelanguage.googleapis.com/v1beta/"
)
# png to base64 text
import base64
with open("test.wav", "rb") as audio:
b64str = base64.b64encode(audio.read()).decode("utf-8")
response = client.chat.completions.create(
model="gemini-1.5-flash",
# model="gpt-4o-audio-preview",
n=1,
modalities=["text"],
messages=[
{"role": "system", "content": "you are a helpful assistant."},
{
"role": "user",
"content": [
{
"type": "text",
"text": "what does he say?",
},
{
"type": "input_audio",
"input_audio": {
"data": b64str,
"format": "wav",
}
}
]
}
]
)
print(response.choices[0].message.content)
返回了以下错误响应。
openai.badrequesterror: error code: 400 - [{'error': {'code': 400, 'message': 'request contains an invalid argument.', 'status': 'invalid_argument'}}]
目前仅支持文字输入，不过后续似乎会支持图片和音频输入。
javascript
让我们看一下 javascript 示例代码。
首先，我们来搭建环境。
npm init -y npm install openai npm pkg set type=module
接下来，让我们运行以下代码。
import openai from "openai";
const google_api_key = process.env.google_api_key;
const openai = new openai({
apikey: google_api_key,
baseurl: "https://generativelanguage.googleapis.com/v1beta/"
});
const response = await openai.chat.completions.create({
model: "gemini-1.5-flash",
messages: [
{ role: "system", content: "you are a helpful assistant." },
{
role: "user",
content: "explain briefly(less than 30 words) to me how ai works",
},
],
});
console.log(response.choices[0].message.content);
运行代码时，请确保在 .env 文件中包含 api 密钥。 .env 文件将在运行时加载。
node --env-file=.env run.js
返回了以下响应。
AI systems learn from data, identify patterns, and make predictions or decisions based on those patterns.
我们可以在同一个库中使用其他模型，这真是太棒了。
就我个人而言，我对此感到很高兴，因为 openai 使编辑对话历史记录变得更加容易。
