server.md

aitg server

intro

this provides a lightweight and plug-and-play http server component for utilizing gpt models!

run server

first you want to run the server. if you want to use docker see the docs for docker. otherwise, you want to run the aitg.srv module.

for example:

KEY=secret MODEL=/tmp/PT_GPTNEO125_ATG python -m aitg.srv gpt --host 127.0.0.1 --port 6000

api

a very simple api that provides various methods.

• either GET or POST can be used
• file extension on the request path specifies the return data type
  • .json = JSON
  • .mp = MsgPack
  • .mpz = LZ4(MsgPack)
• request body is always json data

routes

• info - get info about the server and model
• encode - tokenize text to string
• decode - turn a token sequence to a string
• gen_gpt - generate text

try on your shell!

give it a try with curl:

curl --request GET \
  --url http://localhost:6000/gen_gpt.json \
  --header 'Content-Type: application/json' \
  --data '{
        "key": "secret",
        "prompt": "I like elephants because",
        "max_length": 24
}'