# gpt3-contextual
Contextual chat with GPT-3 of OpenAI API.
# 🚀 Quick start
Install.
```bash
$ pip install gpt3-contextual
```
Make script as console.py.
```python
import asyncio
from gpt3contextual import ContextualChat
async def main():
cc = ContextualChat(
"YOUR_OPENAI_APIKEY",
username="Human",
agentname="AI"
)
while True:
text = input("human> ")
resp, _ = await cc.chat("user1234567890", text)
print(f"AI> {resp}")
asyncio.run(main())
```
Run console.py and chat with GPT-3.
```
human> hello
AI> Hi, how can I help you?
human> I'm hungry
AI> What would you like to eat?
human> sandwitches
AI> What kind of sandwich would you like?
human> ham&egg
AI> Would you like that on a white or wheat bread?
human> wheet
AI> Would you like anything else with your ham and egg sandwich on wheat bread?
human> Everything is fine, thank you.
AI> Great! Your order has been placed. Enjoy your meal!
```
# 🧸 Usage
You can set parameters to customize conversation senario when you make the instance of `ContextualChat`.
See also https://platform.openai.com/docs/api-reference/completions to understands some params.
- `api_key`: str : API key for OpenAI API.
- `context_count`: int : Count of turns(request and response) to use as the context. Default=`6`
- `username`: str : Name of roll of user. Default=`customer`.
- `agentname`: str = Name or role of agent(bot). Default=`agent`.
- `chat_description`: str : Some conditions to be considered in the senario of conversation.
- `engine`: str : The engine(model) to use. Default=`text-davinci-003`.
- `temperature`: float : What sampling temperature to use, between 0 and 2. Default=`0.5`.
- `max_tokens`: int : The maximum number of tokens to generate in the completion. Default=`2000`.
- `context_manager`: ContextManager : Custom ContextManager.
# 🥪 How it works
As you know, `Completion` endpoint doesn't provide the features for keep and use context.
So, simply, this library send the previous conversation histories with the request text like below.
Prompt on the 1st turn:
```
human:hello
AI:
```
Chatbot returns `Hi, how can I help you?`.
Prompt on the 2nd turn:
```
human:hello
AI:Hi, how can I help you?
human:I'm hungry
AI:
```
Chatbot returns `What would you like to eat?`
:
Prompt on the 6th turn:
```
human:sandwitches
AI:What kind of sandwich would you like?
human:ham&egg
AI:Would you like that on a white or wheat bread?
human:wheet
AI:Would you like anything else with your ham and egg sandwich on wheat bread?
human:Everything is fine, thank you.
AI:
```
Chatbot returns `Great! Your order has been placed. Enjoy your meal!`
If you change the `username` and `agentname` the senario of conversation changes.
And, setting `chat_description` also make effects on the situations.
Here is an example to simulate the conversation by brother and sister in Japanese.
```python
cc = ContextualChat(
openai_apikey,
username="兄",
agentname="妹",
chat_description="これは兄と親しい妹との会話です。仲良しなので丁寧語を使わずに話してください。"
)
```
Prompt to be sent to OpenAI API is like bellow. `chat_description` is always set to the first line, no matter how many turns of conversation proceed.
```
これは兄と親しい妹との会話です。仲良しなので丁寧語を使わずに話してください。
兄:おはよー
妹:おはよー！今日の予定は？
兄:いつも通り、特にないよ
妹:じゃあ、今日は一緒に遊ぼうよ！
兄:何かしたいことある？
妹:うん、今日は映画を見よう！
兄:いいね。 どんなのがいい？
妹:思いついた！サスペンス映画がいいかな！
````
Raw data
{
"_id": null,
"home_page": "https://github.com/uezo/gpt3-contextual",
"name": "gpt3-contextual",
"maintainer": "uezo",
"docs_url": null,
"requires_python": "",
"maintainer_email": "uezo@uezo.net",
"keywords": "",
"author": "uezo",
"author_email": "uezo@uezo.net",
"download_url": "",
"platform": null,
"description": "# gpt3-contextual\n\nContextual chat with GPT-3 of OpenAI API.\n\n# \ud83d\ude80 Quick start\n\nInstall.\n\n```bash\n$ pip install gpt3-contextual\n```\n\nMake script as console.py.\n\n```python\nimport asyncio\nfrom gpt3contextual import ContextualChat\n\nasync def main():\n cc = ContextualChat(\n \"YOUR_OPENAI_APIKEY\",\n username=\"Human\",\n agentname=\"AI\"\n )\n\n while True:\n text = input(\"human> \")\n resp, _ = await cc.chat(\"user1234567890\", text)\n print(f\"AI> {resp}\")\n\nasyncio.run(main())\n```\n\nRun console.py and chat with GPT-3.\n\n```\nhuman> hello\nAI> Hi, how can I help you?\nhuman> I'm hungry\nAI> What would you like to eat?\nhuman> sandwitches\nAI> What kind of sandwich would you like?\nhuman> ham&egg \nAI> Would you like that on a white or wheat bread?\nhuman> wheet\nAI> Would you like anything else with your ham and egg sandwich on wheat bread?\nhuman> Everything is fine, thank you.\nAI> Great! Your order has been placed. Enjoy your meal!\n```\n\n\n# \ud83e\uddf8 Usage\n\nYou can set parameters to customize conversation senario when you make the instance of `ContextualChat`.\nSee also https://platform.openai.com/docs/api-reference/completions to understands some params.\n\n- `api_key`: str : API key for OpenAI API.\n- `context_count`: int : Count of turns(request and response) to use as the context. Default=`6`\n- `username`: str : Name of roll of user. Default=`customer`.\n- `agentname`: str = Name or role of agent(bot). Default=`agent`.\n- `chat_description`: str : Some conditions to be considered in the senario of conversation.\n- `engine`: str : The engine(model) to use. Default=`text-davinci-003`.\n- `temperature`: float : What sampling temperature to use, between 0 and 2. Default=`0.5`.\n- `max_tokens`: int : The maximum number of tokens to generate in the completion. Default=`2000`.\n- `context_manager`: ContextManager : Custom ContextManager.\n\n\n# \ud83e\udd6a How it works\n\nAs you know, `Completion` endpoint doesn't provide the features for keep and use context.\nSo, simply, this library send the previous conversation histories with the request text like below.\n\nPrompt on the 1st turn:\n```\nhuman:hello\nAI:\n```\nChatbot returns `Hi, how can I help you?`.\n\nPrompt on the 2nd turn:\n```\nhuman:hello\nAI:Hi, how can I help you?\nhuman:I'm hungry\nAI:\n```\nChatbot returns `What would you like to eat?`\n\n:\n\nPrompt on the 6th turn:\n```\nhuman:sandwitches\nAI:What kind of sandwich would you like?\nhuman:ham&egg \nAI:Would you like that on a white or wheat bread?\nhuman:wheet\nAI:Would you like anything else with your ham and egg sandwich on wheat bread?\nhuman:Everything is fine, thank you.\nAI:\n```\nChatbot returns `Great! Your order has been placed. Enjoy your meal!`\n\n\nIf you change the `username` and `agentname` the senario of conversation changes.\nAnd, setting `chat_description` also make effects on the situations.\n\nHere is an example to simulate the conversation by brother and sister in Japanese.\n\n```python\ncc = ContextualChat(\n openai_apikey,\n username=\"\u5144\",\n agentname=\"\u59b9\",\n chat_description=\"\u3053\u308c\u306f\u5144\u3068\u89aa\u3057\u3044\u59b9\u3068\u306e\u4f1a\u8a71\u3067\u3059\u3002\u4ef2\u826f\u3057\u306a\u306e\u3067\u4e01\u5be7\u8a9e\u3092\u4f7f\u308f\u305a\u306b\u8a71\u3057\u3066\u304f\u3060\u3055\u3044\u3002\"\n)\n```\n\nPrompt to be sent to OpenAI API is like bellow. `chat_description` is always set to the first line, no matter how many turns of conversation proceed.\n\n```\n\u3053\u308c\u306f\u5144\u3068\u89aa\u3057\u3044\u59b9\u3068\u306e\u4f1a\u8a71\u3067\u3059\u3002\u4ef2\u826f\u3057\u306a\u306e\u3067\u4e01\u5be7\u8a9e\u3092\u4f7f\u308f\u305a\u306b\u8a71\u3057\u3066\u304f\u3060\u3055\u3044\u3002\n\u5144:\u304a\u306f\u3088\u30fc\n\u59b9:\u304a\u306f\u3088\u30fc\uff01\u4eca\u65e5\u306e\u4e88\u5b9a\u306f\uff1f\n\u5144:\u3044\u3064\u3082\u901a\u308a\u3001\u7279\u306b\u306a\u3044\u3088\n\u59b9:\u3058\u3083\u3042\u3001\u4eca\u65e5\u306f\u4e00\u7dd2\u306b\u904a\u307c\u3046\u3088\uff01\n\u5144:\u4f55\u304b\u3057\u305f\u3044\u3053\u3068\u3042\u308b\uff1f\n\u59b9:\u3046\u3093\u3001\u4eca\u65e5\u306f\u6620\u753b\u3092\u898b\u3088\u3046\uff01\n\u5144:\u3044\u3044\u306d\u3002 \u3069\u3093\u306a\u306e\u304c\u3044\u3044\uff1f\n\u59b9:\u601d\u3044\u3064\u3044\u305f\uff01\u30b5\u30b9\u30da\u30f3\u30b9\u6620\u753b\u304c\u3044\u3044\u304b\u306a\uff01\n````\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "Contextual chat with GPT-3 of OpenAI API.",
"version": "0.1",
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "cb5ae35fbda39fd059b3edf6a2bfca9fd986a79b0bcaf44e2e3052ef095f5051",
"md5": "a1ec264f22c5af55b6ad5762869229d8",
"sha256": "4456747f1cecf7c4abbb537ce5c46e49912b097c95bd8eea145b2b0ceab3499f"
},
"downloads": -1,
"filename": "gpt3_contextual-0.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "a1ec264f22c5af55b6ad5762869229d8",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": null,
"size": 5401,
"upload_time": "2023-02-11T14:37:11",
"upload_time_iso_8601": "2023-02-11T14:37:11.224056Z",
"url": "https://files.pythonhosted.org/packages/cb/5a/e35fbda39fd059b3edf6a2bfca9fd986a79b0bcaf44e2e3052ef095f5051/gpt3_contextual-0.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2023-02-11 14:37:11",
"github": true,
"gitlab": false,
"bitbucket": false,
"github_user": "uezo",
"github_project": "gpt3-contextual",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"requirements": [],
"lcname": "gpt3-contextual"
}
JSON
