This repository contains examples of integrating gpt into application targeting SAS Viya.
The examples use gpt-4 with functions. Many of the functions use @sassoftware/restaf, @sassoftware/restaflib and @sassoftware/restafedit to make REST API calls to SAS Viya.
The functions are small and easy to follow. Feel free to change these to suit own programming style.
The code is in https://github.com/sassoftware/restaf-demos/tree/gpt-samples
---;
- gpt-4 model is the default openai model.
- OPENAI_MODEL - To change the model, set this enviroment variable to the desired model.
- It is the user's reponsibility to get the api key for openai.
- The samples assume that the key is available as environment variable OPENAI_KEY
- Most examples make use of SAS REST API to access SAS Viya. The applications use @sassoftware/restaf, @sassoftware/restaflib and @sassoftware/restafedit to make these calls. See https://sassoftware.github.io/restaf/ for more information.
- The source code in this repository is provided under Apache-2.0 licensing model
Please sas-cli auth login|logCode to setup authentication token.
- OPENAI_KEY - Set this to the api key from openai
- OPENAI_MODEL - (optional) set to the model you want to use(defaults to gpt-4)
- VIYASOURCE - set this to either cas, compute or none. Defaults to cas. Used in prompts that access Viya resources. The 'none' value is useful if you want to use this cli as a standard chat without accessing Viya.
After setting the environment variables listed above, start the application as follows:
npx @sassoftware/gpt-samples@latest
On the prompt, enter chat to start the gpt session
Enter help to get guidance on what is possible
The gpt session will stay active until you enter exit
To exit the cli enter exit again.
The instruction below is for those users who wish to clone the code and extend the application.
- git clone https://github.com/sassoftware/restaf-demos gpt-samples -b gpt-samples
- cd gpt-samples
- npm install
- Define the environment variables described below
This example demonstrates setting up access to gpt and defining a function called basic. This function's sole purpose is to reformat user provided keywords as html, array or JavaScript object. The code is in ./packages/basic/index.js
npm run basic <some text>
Some sample prompt and results:
SAS Viya is a cloud-based, in-memory analytics engine from SAS Institute, an American multinational developer of analytics software. SAS Viya provides quick, accurate results and reveals valuable insights from large amounts of data. It's capable of machine learning, text analytics, forecasting, optimization, and statistics. It can be used through a variety of programming languages including Python, R, Java, and Lua, or through its visual interface.
Sure, can you provide me with the keywords you want to process?
[ 'a', 'b', 'c' ]
<ul><li>a</li><li>b</li><li>c</li></ul>
This is similar to the previous example The basic function now executes SAS code or casl code.
The prompt should look something like this:
run file path to your .sas file or .casl file
The following starter files are included. Use .sas files if env VIYASOURCE is set to compute
- ../../programs/echo.casl
- ../../programs/datastep.casl
- ../../programs/datastep.sas
npm run basicSAS "run file <some path>"
{
"casResults": {
"result": {
"Fetch": {
"_ctb": true,
"attributes": {
"Action": {
"type": "string",
"value": "fetch"
},
"Actionset": {
"type": "string",
"value": "table"
},
"CreateTime": {
"type": "double",
"value": 2022787095.92428
}
},
"label": "Selected Rows from Table A",
"name": "Fetch",
"rows": [
[
1,
1
]
],
"schema": [
{
"attributes": {},
"format": "",
"label": "",
"name": "_Index_",
"type": "int",
"width": 8
},
{
"attributes": {},
"format": "",
"label": "",
"name": "x",
"type": "double",
"width": 8
}
],
"title": "Selected Rows from Table A"
}
}
}
}
Make sure the env variable VIYASOURCE is set to compute.
[
{
"line": "1 options NOSYNTAXCHECK OBS=MAX;%let syscc=0;",
"type": "source",
"version": 1
},
{
"line": "2 data a;",
"type": "source",
"version": 1
},
{
"line": "3 x=1;",
"type": "source",
"version": 1
},
{
"line": "4 run;",
"type": "source",
"version": 1
},
{
"line": "",
"type": "note",
"version": 1
},
{
"line": "NOTE: The data set WORK.A has 1 observations and 1 variables.",
"type": "note",
"version": 1
},
{
"line": "NOTE: DATA statement used (Total process time):",
"type": "note",
"version": 1
},
{
"line": " real time 0.00 seconds",
"type": "note",
"version": 1
},
{
"line": " cpu time 0.00 seconds",
"type": "note",
"version": 1
},
{
"line": " ",
"type": "note",
"version": 1
},
{
"line": "",
"type": "note",
"version": 1
},
{
"line": "5 ",
"type": "source",
"version": 1
},
{
"line": "6 proc print; run;",
"type": "source",
"version": 1
},
{
"line": "",
"type": "note",
"version": 1
},
{
"line": "NOTE: There were 1 observations read from the data set WORK.A.",
"type": "note",
"version": 1
},
{
"line": "NOTE: The PROCEDURE PRINT printed page 1.",
"type": "note",
"version": 1
},
{
"line": "NOTE: PROCEDURE PRINT used (Total process time):",
"type": "note",
"version": 1
},
{
"line": " real time 0.05 seconds",
"type": "note",
"version": 1
},
{
"line": " cpu time 0.06 seconds",
"type": "note",
"version": 1
},
{
"line": " ",
"type": "note",
"version": 1
},
{
"line": "",
"type": "note",
"version": 1
},
{
"line": "7 ",
"type": "source",
"version": 1
},
{
"line": "8 ;",
"type": "source",
"version": 1
},
{
"line": "",
"type": "note",
"version": 1
}
]