2023-08-12 17:49:31 +02:00
""" Does quite a few checks and prepares the incoming request for the target endpoint, so it can be streamed """
2023-06-30 02:49:56 +02:00
2023-07-19 23:51:28 +02:00
import json
2023-08-04 03:30:56 +02:00
import yaml
2023-08-06 21:42:07 +02:00
import fastapi
2023-06-30 02:49:56 +02:00
2023-06-28 15:21:14 +02:00
from dotenv import load_dotenv
2023-08-03 01:46:49 +02:00
2023-08-04 03:30:56 +02:00
import streaming
2023-08-06 21:42:07 +02:00
import moderation
2023-08-03 01:46:49 +02:00
2023-08-07 23:28:24 +02:00
from db import users
from helpers import tokens , errors
2023-06-28 15:21:14 +02:00
load_dotenv ( )
2023-08-09 11:43:05 +02:00
models_list = json . load ( open ( ' models.json ' ) )
2023-08-04 03:30:56 +02:00
with open ( ' config/credits.yml ' , encoding = ' utf8 ' ) as f :
credits_config = yaml . safe_load ( f )
2023-08-03 01:46:49 +02:00
async def handle ( incoming_request ) :
2023-06-30 02:49:56 +02:00
""" Transfer a streaming response from the incoming request to the target endpoint """
2023-07-25 02:42:53 +02:00
2023-08-07 23:28:24 +02:00
path = incoming_request . url . path . replace ( ' v1/v1/ ' , ' v1/ ' )
2023-06-30 02:49:56 +02:00
2023-08-04 03:30:56 +02:00
# METHOD
2023-08-01 20:19:00 +02:00
if incoming_request . method not in [ ' GET ' , ' POST ' , ' PUT ' , ' DELETE ' , ' PATCH ' ] :
2023-08-12 17:56:21 +02:00
return await errors . error ( 405 , f ' Method " { incoming_request . method } " is not allowed. ' , ' Change the request method to the correct one. ' )
2023-07-19 23:51:28 +02:00
2023-08-04 03:30:56 +02:00
# PAYLOAD
2023-07-19 23:51:28 +02:00
try :
2023-07-25 02:42:53 +02:00
payload = await incoming_request . json ( )
2023-07-19 23:51:28 +02:00
except json . decoder . JSONDecodeError :
2023-07-25 02:42:53 +02:00
payload = { }
2023-08-12 17:56:21 +02:00
# Tokenise w/ tiktoken
2023-08-01 20:19:00 +02:00
try :
2023-08-06 21:42:07 +02:00
input_tokens = await tokens . count_for_messages ( payload [ ' messages ' ] )
2023-08-03 01:46:49 +02:00
except ( KeyError , TypeError ) :
2023-08-01 20:19:00 +02:00
input_tokens = 0
2023-08-12 17:56:21 +02:00
# Check user auth
2023-08-04 03:30:56 +02:00
received_key = incoming_request . headers . get ( ' Authorization ' )
2023-08-01 20:19:00 +02:00
2023-08-04 03:30:56 +02:00
if not received_key :
2023-08-12 17:56:21 +02:00
return await errors . error ( 401 , ' No NovaAI API key given! ' , ' Add " Authorization: Bearer nv-... " to your request headers. ' )
2023-08-01 20:19:00 +02:00
2023-08-04 03:30:56 +02:00
if received_key . startswith ( ' Bearer ' ) :
received_key = received_key . split ( ' Bearer ' ) [ 1 ]
2023-08-01 20:19:00 +02:00
2023-08-04 03:30:56 +02:00
user = await users . by_api_key ( received_key . strip ( ) )
2023-08-01 20:19:00 +02:00
if not user :
2023-08-12 17:56:21 +02:00
return await errors . error ( 401 , ' Invalid NovaAI API key! ' , ' Create a new NovaOSS API key. ' )
2023-08-03 01:46:49 +02:00
ban_reason = user [ ' status ' ] [ ' ban_reason ' ]
if ban_reason :
2023-08-12 17:56:21 +02:00
return await errors . error ( 403 , f ' Your NovaAI account has been banned. Reason: " { ban_reason } " . ' , ' Contact the staff for an appeal. ' )
2023-08-03 01:46:49 +02:00
if not user [ ' status ' ] [ ' active ' ] :
2023-08-12 17:56:21 +02:00
return await errors . error ( 418 , ' Your NovaAI account is not active (paused). ' , ' Simply re-activate your account using a Discord command or the web panel. ' )
2023-08-03 01:46:49 +02:00
2023-08-09 11:43:05 +02:00
if ' /models ' in path :
return fastapi . responses . JSONResponse ( content = models_list )
2023-08-12 17:56:21 +02:00
# Calculate cost of tokens & check for nsfw prompts
2023-08-04 03:30:56 +02:00
costs = credits_config [ ' costs ' ]
cost = costs [ ' other ' ]
2023-08-03 01:46:49 +02:00
2023-08-08 01:04:35 +02:00
policy_violation = False
2023-08-06 21:42:07 +02:00
2023-08-04 03:30:56 +02:00
if ' chat/completions ' in path :
for model_name , model_cost in costs [ ' chat-models ' ] . items ( ) :
if model_name in payload [ ' model ' ] :
cost = model_cost
2023-08-03 01:46:49 +02:00
2023-08-08 01:04:35 +02:00
policy_violation = await moderation . is_policy_violated ( payload [ ' messages ' ] )
elif ' /moderations ' in path :
pass
2023-08-06 21:42:07 +02:00
else :
inp = payload . get ( ' input ' , payload . get ( ' prompt ' ) )
2023-08-07 23:28:24 +02:00
if inp :
if len ( inp ) > 2 and not inp . isnumeric ( ) :
2023-08-08 01:04:35 +02:00
policy_violation = await moderation . is_policy_violated ( inp )
2023-08-07 23:28:24 +02:00
2023-08-08 01:04:35 +02:00
if policy_violation :
2023-08-12 17:56:21 +02:00
return await errors . error ( 400 , f ' The request contains content which violates this model \' s policies for " { policy_violation } " . ' , ' We currently don \' t support any NSFW models. ' )
2023-08-06 21:42:07 +02:00
2023-08-04 03:30:56 +02:00
role_cost_multiplier = credits_config [ ' bonuses ' ] . get ( user [ ' role ' ] , 1 )
cost = round ( cost * role_cost_multiplier )
2023-08-03 01:46:49 +02:00
if user [ ' credits ' ] < cost :
2023-08-12 17:56:21 +02:00
return await errors . error ( 429 , ' Not enough credits. ' , ' Wait or earn more credits. Learn more on our website or Discord server. ' )
2023-08-01 20:19:00 +02:00
2023-08-12 17:56:21 +02:00
# Send the completion request
2023-06-28 15:21:14 +02:00
2023-08-06 00:43:36 +02:00
if ' chat/completions ' in path and not payload . get ( ' stream ' ) is True :
2023-08-04 03:30:56 +02:00
payload [ ' stream ' ] = False
2023-08-03 03:50:04 +02:00
2023-08-06 21:42:07 +02:00
media_type = ' text/event-stream ' if payload . get ( ' stream ' , False ) else ' application/json '
return fastapi . responses . StreamingResponse (
2023-08-04 03:30:56 +02:00
content = streaming . stream (
user = user ,
path = path ,
payload = payload ,
credits_cost = cost ,
input_tokens = input_tokens ,
incoming_request = incoming_request ,
) ,
2023-08-06 21:42:07 +02:00
media_type = media_type
2023-08-04 03:30:56 +02:00
)