nova-api/api/handler.py

"""Does quite a few checks and prepares the incoming request for the target endpoint, so it can be streamed"""

import os
import json
import yaml
import time
import orjson
import fastapi

from dotenv import load_dotenv

import responder
import moderation

from rich import print
from db.users import UserManager
from helpers import tokens, errors, network

load_dotenv()

users = UserManager()
models_list = json.load(open('cache/models.json', encoding='utf8'))
models = [model['id'] for model in models_list['data']]

with open('config/config.yml', encoding='utf8') as f:
    config = yaml.safe_load(f)

moderation_debug_key_key = os.getenv('MODERATION_DEBUG_KEY')

async def handle(incoming_request: fastapi.Request):
    """Transfer a streaming response 
    Takes the request from the incoming request to the target endpoint.
    Checks method, token amount, auth and cost along with if request is NSFW.
    """

    path = incoming_request.url.path
    path = path.replace('/v1/v1', '/v1')

    ip_address = await network.get_ip(incoming_request)

    if '/models' in path:
        return fastapi.responses.JSONResponse(content=models_list)

    try:
        payload = await incoming_request.json()
    except json.decoder.JSONDecodeError:
        payload = {}
    except UnicodeDecodeError:
        payload = {}

    received_key = incoming_request.headers.get('Authorization')

    if not received_key or not received_key.startswith('Bearer '):
        return await errors.error(401, 'No NovaAI API key given!', 'Add \'Authorization: Bearer nv-...\' to your request headers.')

    key_tags = ''

    if '#' in received_key:
        key_tags = received_key.split('#')[1]
        received_key = received_key.split('#')[0]

    user = await users.user_by_api_key(received_key.split('Bearer ')[1].strip())

    if not user or not user['status']['active']:
        return await errors.error(418, 'Invalid or inactive NovaAI API key!', 'Create a new NovaOSS API key or reactivate your account.')

    ban_reason = user['status']['ban_reason']
    if ban_reason:
        return await errors.error(403, f'Your NovaAI account has been banned. Reason: \'{ban_reason}\'.', 'Contact the staff for an appeal.')

    # Checking for enterprise status
    enterprise_keys = os.environ.get('ENTERPRISE_KEYS')
    if path.startswith('/enterprise/v1') and user.get('api_key') not in enterprise_keys.split():
        return await errors.error(403, 'Enterprise API is not available.', 'Contact the staff for an upgrade.')

    if 'account/credits' in path:
        return fastapi.responses.JSONResponse({'credits': user['credits']})

    costs = config['costs']
    cost = costs['other']

    if 'chat/completions' in path:
        cost = costs['chat-models'].get(payload.get('model'), cost)

    role = user.get('role', 'default')

    try:
        role_cost_multiplier = config['roles'][role]['bonus']
    except KeyError:
        role_cost_multiplier = 1

    cost = round(cost * role_cost_multiplier)

    if user['credits'] < cost:
        return await errors.error(429, 'Not enough credits.', 'Wait or earn more credits. Learn more on our website or Discord server.')


    if 'DISABLE_VARS' not in key_tags:
        payload_with_vars = json.dumps(payload)

        replace_dict = {
            'timestamp': str(int(time.time())),
            'date': time.strftime('%Y-%m-%d'),
            'time': time.strftime('%H:%M:%S'),
            'datetime': time.strftime('%Y-%m-%d %H:%M:%S'),
            'model': payload.get('model', 'unknown'),
        }

        if 'ALLOW_INSECURE_VARS' in key_tags:
            replace_dict.update({
                'my.ip': ip_address,
                'my.id': str(user['_id']),
                'my.role': user.get('role', 'default'),
                'my.credits': str(user['credits']),
                'my.discord': user.get('auth', {}).get('discord', ''),
            })

        for key, value in replace_dict.items():
            payload_with_vars = payload_with_vars.replace(f'[[{key}]]', value)

        payload = json.loads(payload_with_vars)

    policy_violation = False

    if not (moderation_debug_key_key and moderation_debug_key_key in key_tags and 'gpt-3' in payload.get('model', '')):
        if '/moderations' not in path:
            inp = ''

            if 'input' in payload or 'prompt' in payload:
                inp = payload.get('input', payload.get('prompt', ''))

            if isinstance(payload.get('messages'), list):
                inp = ''

                for message in payload.get('messages', []):
                    if message.get('role') == 'user':
                        inp += message.get('content', '') + '\n'

            if 'functions' in payload:
                inp += '\n'.join([function.get('description', '') for function in payload.get('functions', [])])

            if inp and len(inp) > 2 and not inp.isnumeric():
                policy_violation = await moderation.is_policy_violated(inp)

    if policy_violation:
        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.'
        )

    if 'chat/completions' in path and not payload.get('stream', False):
        payload['stream'] = False
    if 'chat/completions' in path and not payload.get('model'):
        payload['model'] = 'gpt-3.5-turbo'

    media_type = 'text/event-stream' if payload.get('stream', False) else 'application/json'

    if (model := payload.get('model')) not in models and model is not None:
        return await errors.error(404, 'Model not found.', 'Check the model name and try again.')

    return fastapi.responses.StreamingResponse(
        content=responder.respond(
            user=user,
            path=path,
            payload=payload,
            credits_cost=cost,
            input_tokens=0,
            incoming_request=incoming_request,
        ),
        media_type=media_type
    )