README.md

@@ -20,6 +20,8 @@ We aim to fix that! NovaAI provides several AI models for you to use for free.
 
 
 ###### *I founded FoxGPT (called *NovaGPT* back then)
+Old, slow, deprecated* FoxGPT vs new NovaAI repository star count:
+
 <a href="https://star-history.com/#NovaOSS/nova-api&FoxGPT/gpt&Date">
   <picture>
     <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=NovaOSS/nova-api,FoxGPT/gpt&type=Date&theme=dark" />
@@ -28,6 +30,8 @@ We aim to fix that! NovaAI provides several AI models for you to use for free.
   </picture>
 </a>
 
+<img alt="'Emotional damage' meme, with a man with a worried face and the yellow caption 'emotional damage'" src="https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Findianmemetemplates.com%2Fwp-content%2Fuploads%2Femotional-damage-1024x575.jpg&f=1&nofb=1&ipt=b325721ee0a7b9e11603a9bd484c8042b82e1704e639887107c6ce3e0d9b389e&ipo=images" height=100>
+
 ## NovaOSS APIs
 Our infrastructure might seem a bit confusing, but it's actually quite simple. Just the first one really matters for you, if you want to access our AI API. The other ones are just for the team.
 

api/chunks.py

@@ -0,0 +1,30 @@
+import json
+
+from helpers import chat
+
+async def process_chunks(
+    chunks,
+    is_chat: bool,
+    chat_id: int,
+    target_request: dict,
+    model: str=None,
+):
+    """This function processes the response chunks from the providers and yields them.
+    """
+    async for chunk in chunks:
+        chunk = chunk.decode("utf8").strip()
+        send = False
+
+        if is_chat and '{' in chunk:
+            data = json.loads(chunk.split('data: ')[1])
+            chunk = chunk.replace(data['id'], chat_id)
+            send = True
+
+            if target_request['module'] == 'twa' and data.get('text'):
+                chunk = await chat.create_chat_chunk(chat_id=chat_id, model=model, content=['text'])
+
+            if (not data['choices'][0]['delta']) or data['choices'][0]['delta'] == {'role': 'assistant'}:
+                send = False
+
+        if send and chunk:
+            yield chunk + '\n\n'

api/core.py

@@ -50,9 +50,6 @@ async def get_users(discord_id: int, incoming_request: fastapi.Request):
     if not user:
         return await errors.error(404, 'Discord user not found in the API database.', 'Check the `discord_id` parameter.')
 
-    # turn the ObjectId into a string
-    user['_id'] = str(user['_id'])
-
     return user
 
 async def new_user_webhook(user: dict) -> None:
@@ -93,8 +90,6 @@ async def create_user(incoming_request: fastapi.Request):
     user = await manager.create(discord_id)
     await new_user_webhook(user)
 
-    user['_id'] = str(user['_id'])
-
     return user
 
 @router.put('/users')
@@ -131,23 +126,24 @@ async def run_checks(incoming_request: fastapi.Request):
     if auth_error:
         return auth_error
 
-    results = {}
+    try:
+        chat = await checks.client.test_chat()
+    except Exception as exc:
+        print(exc)
+        chat = None
 
-    funcs = [
-        checks.client.test_chat_non_stream_gpt4,
-        checks.client.test_chat_stream_gpt3,
-        checks.client.test_function_calling,
-        checks.client.test_image_generation,
-        # checks.client.test_speech_to_text,
-        checks.client.test_models
-    ]
+    try:
+        moderation = await checks.client.test_api_moderation()
+    except Exception:
+        moderation = None
 
-    for func in funcs:
-        try:
-            result = await func()
-        except Exception as exc:
-            results[func.__name__] = str(exc)
-        else:
-            results[func.__name__] = result
+    try:
+        models = await checks.client.test_models()
+    except Exception:
+        models = None
 
-    return results
+    return {
+        'chat/completions': chat,
+        'models': models,
+        'moderations': moderation,
+    }

api/handler.py

@@ -124,14 +124,7 @@ async def handle(incoming_request: fastapi.Request):
                 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', [])])
+                inp = '\n'.join([message['content'] for message in payload['messages']])
 
             if inp and len(inp) > 2 and not inp.isnumeric():
                 policy_violation = await moderation.is_policy_violated(inp)
@@ -155,7 +148,7 @@ async def handle(incoming_request: fastapi.Request):
             path=path,
             payload=payload,
             credits_cost=cost,
-            input_tokens=0,
+            input_tokens=-1,
             incoming_request=incoming_request,
         ),
         media_type=media_type

api/main.py

@@ -2,12 +2,12 @@
 
 import fastapi
 import pydantic
+import functools
 
 from rich import print
 from dotenv import load_dotenv
-
+from json import JSONDecodeError
 from bson.objectid import ObjectId
-
 from slowapi.errors import RateLimitExceeded
 from slowapi.middleware import SlowAPIMiddleware
 from fastapi.middleware.cors import CORSMiddleware
@@ -17,6 +17,7 @@ from helpers import network
 
 import core
 import handler
+import moderation
 
 load_dotenv()
 
@@ -65,5 +66,17 @@ async def root():
 
 @app.route('/v1/{path:path}', methods=['GET', 'POST', 'PUT', 'DELETE', 'PATCH'])
 async def v1_handler(request: fastapi.Request):
-    res = await handler.handle(incoming_request=request)
+    res = await handler.handle(request)
     return res
+
+@functools.lru_cache()
+@app.post('/moderate')
+async def moderate(request: fastapi.Request):
+    try:
+        prompt = await request.json()
+        prompt = prompt['text']
+    except (KeyError, JSONDecodeError):
+        return fastapi.Response(status_code=400)
+
+    result = await moderation.is_policy_violated__own_model(prompt)
+    return result or ''

api/responder.py

@@ -11,6 +11,7 @@ import starlette
 from rich import print
 from dotenv import load_dotenv
 
+import chunks
 import proxies
 import provider_auth
 import after_request
@@ -20,6 +21,24 @@ from helpers import network, chat, errors
 
 load_dotenv()
 
+## Loads config which contains rate limits
+with open('config/config.yml', encoding='utf8') as f:
+    config = yaml.safe_load(f)
+
+## Where all rate limit requested data will be stored.
+# Rate limit data is **not persistent** (It will be deleted on server stop/restart).
+user_last_request_time = {}
+
+DEMO_PAYLOAD = {
+    'model': 'gpt-3.5-turbo',
+    'messages': [
+        {
+            'role': 'user',
+            'content': '1+1='
+        }
+    ]
+}
+
 async def respond(
     path: str='/v1/chat/completions',
     user: dict=None,
@@ -33,22 +52,27 @@ async def respond(
     """
 
     is_chat = False
+    is_stream = payload.get('stream', False)
 
     model = None
-    is_stream = False
 
     if 'chat/completions' in path:
         is_chat = True
         model = payload['model']
 
+    if is_chat and is_stream:
+        chat_id = await chat.create_chat_id()
+        yield await chat.create_chat_chunk(chat_id=chat_id, model=model, content=chat.CompletionStart)
+        yield await chat.create_chat_chunk(chat_id=chat_id, model=model, content=None)
+
     json_response = {}
 
     headers = {
         'Content-Type': 'application/json',
-        'User-Agent': 'axios/0.21.1',
+        'User-Agent': 'null'
     }
 
-    for _ in range(10):
+    for _ in range(5):
         # Load balancing: randomly selecting a suitable provider
         # If the request is a chat completion, then we need to load balance between chat providers
         # If the request is an organic request, then we need to load balance between organic providers
@@ -91,11 +115,10 @@ async def respond(
                     cookies=target_request.get('cookies'),
                     ssl=False,
                     timeout=aiohttp.ClientTimeout(
-                        connect=0.3,
+                        connect=0.5,
                         total=float(os.getenv('TRANSFER_TIMEOUT', '500'))
                     ),
                 ) as response:
-                    is_stream = response.content_type == 'text/event-stream'
 
                     if response.status == 429:
                         continue
@@ -121,27 +144,35 @@ async def respond(
                             if 'Too Many Requests' in str(exc):
                                 continue
 
-                        async for chunk in response.content.iter_any():
-                            chunk = chunk.decode('utf8').strip()
-                            yield chunk + '\n\n'
+                        async for chunk in chunks.process_chunks(
+                            chunks=response.content.iter_any(),
+                            is_chat=is_chat,
+                            chat_id=chat_id,
+                            model=model,
+                            target_request=target_request
+                        ):
+                            yield chunk
 
                     break
 
             except Exception as exc:
+                # print(f'[!] {type(exc)} - {exc}')
                 continue
 
             if (not json_response) and is_chat:
                 print('[!] chat response is empty')
                 continue
     else:
-        yield await errors.yield_error(500, 'Sorry, the provider is not responding. We\'re possibly getting rate-limited.', 'Please try again later.')
+        yield await errors.yield_error(500, 'Sorry, the API is not responding.', 'Please try again later.')
         return
 
+    if is_chat and is_stream:
+        yield await chat.create_chat_chunk(chat_id=chat_id, model=model, content=chat.CompletionStop)
+        yield 'data: [DONE]\n\n'
+
     if (not is_stream) and json_response:
         yield json.dumps(json_response)
 
-    print(f'[+] {path} -> {model or ""}')
-
     await after_request.after_request(
         incoming_request=incoming_request,
         target_request=target_request,
@@ -152,3 +183,5 @@ async def respond(
         is_chat=is_chat,
         model=model,
     )
+
+    print(f'[+] {path} -> {model or ""}')

checks/client.py

@@ -2,7 +2,6 @@
 
 import os
 import time
-import json
 import httpx
 import openai
 import asyncio
@@ -11,7 +10,6 @@ import traceback
 from rich import print
 from typing import List
 from dotenv import load_dotenv
-from pydantic import BaseModel
 
 load_dotenv()
 
@@ -20,7 +18,7 @@ MODEL = 'gpt-3.5-turbo'
 MESSAGES = [
     {
         'role': 'user',
-        'content': 'Just respond with the number "1337", nothing else.'
+        'content': '1+1=',
     }
 ]
 
@@ -45,12 +43,12 @@ async def test_server():
     else:
         return time.perf_counter() - request_start
 
-async def test_chat_non_stream_gpt4() -> float:
-    """Tests non-streamed chat completions with the GPT-4 model."""
+async def test_chat_non_stream(model: str=MODEL, messages: List[dict]=None) -> dict:
+    """Tests an API api_endpoint."""
 
     json_data = {
-        'model': 'gpt-4',
-        'messages': MESSAGES,
+        'model': model,
+        'messages': messages or MESSAGES,
         'stream': False
     }
 
@@ -65,52 +63,10 @@ async def test_chat_non_stream_gpt4() -> float:
         )
         response.raise_for_status()
 
-    assert '1337' in response.json()['choices'][0]['message']['content'], 'The API did not return a correct response.'
+    assert '2' in response.json()['choices'][0]['message']['content'], 'The API did not return a correct response.'
     return time.perf_counter() - request_start
 
-async def test_chat_stream_gpt3() -> float:
-    """Tests the text stream endpoint with the GPT-3.5-Turbo model."""
-
-    json_data = {
-        'model': 'gpt-3.5-turbo',
-        'messages': MESSAGES,
-        'stream': True,
-    }
-
-    request_start = time.perf_counter()
-
-    async with httpx.AsyncClient() as client:
-        response = await client.post(
-            url=f'{api_endpoint}/chat/completions',
-            headers=HEADERS,
-            json=json_data,
-            timeout=10,
-        )
-        response.raise_for_status()
-
-    chunks = []
-    resulting_text = ''
-
-    async for chunk in response.aiter_text():
-        for subchunk in chunk.split('\n\n'):
-            chunk = subchunk.replace('data: ', '').strip()
-
-            if chunk == '[DONE]':
-                break
-
-            if chunk:
-                chunks.append(json.loads(chunk))
-
-                try:
-                    resulting_text += json.loads(chunk)['choices'][0]['delta']['content']
-                except KeyError:
-                    pass
-
-    assert '1337' in resulting_text, 'The API did not return a correct response.'
-
-    return time.perf_counter() - request_start
-
-async def test_image_generation() -> float:
+async def test_sdxl():
     """Tests the image generation endpoint with the SDXL model."""
 
     json_data = {
@@ -133,48 +89,6 @@ async def test_image_generation() -> float:
     assert '://' in response.json()['data'][0]['url']
     return time.perf_counter() - request_start
 
-class StepByStepAIResponse(BaseModel):
-    """Demo response structure for the function calling test."""
-    title: str
-    steps: List[str]
-
-async def test_function_calling():
-    """Tests function calling functionality with newer GPT models."""
-
-    json_data = {
-        'stream': False,
-        'model': 'gpt-3.5-turbo-0613',
-        'messages': [
-            {"role": "user", "content": "Explain how to assemble a PC"}
-        ],
-        'functions': [
-            {
-                'name': 'get_answer_for_user_query',
-                'description': 'Get user answer in series of steps',
-                'parameters': StepByStepAIResponse.schema()
-            }
-        ],
-        'function_call': {'name': 'get_answer_for_user_query'}
-    }
-
-    request_start = time.perf_counter()
-
-    async with httpx.AsyncClient() as client:
-        response = await client.post(
-            url=f'{api_endpoint}/chat/completions',
-            headers=HEADERS,
-            json=json_data,
-            timeout=10,
-        )
-        response.raise_for_status()
-
-    res = response.json()
-    output = json.loads(res['choices'][0]['message']['function_call']['arguments'])
-    print(output)
-
-    assert output.get('title') and output.get('steps'), 'The API did not return a correct response.'
-    return time.perf_counter() - request_start
-
 async def test_models():
     """Tests the models endpoint."""
 
@@ -208,20 +122,17 @@ async def demo():
         else:
             raise ConnectionError('API Server is not running.')
 
-        # print('[lightblue]Checking if function calling works...')
-        # print(await test_function_calling())
+        print('Checking non-streamed chat completions...')
+        print(await test_chat_non_stream())
 
-        # print('Checking non-streamed chat completions...')
-        # print(await test_chat_non_stream_gpt4())
+        # print('[lightblue]Checking if SDXL image generation works...')
+        # print(await test_sdxl())
 
-        # print('Checking streamed chat completions...')
-        # print(await test_chat_stream_gpt3())
+        # print('[lightblue]Checking if the moderation endpoint works...')
+        # print(await test_api_moderation())
 
-        # print('[lightblue]Checking if  image generation works...')
-        # print(await test_image_generation())
-
-        # print('Checking the models endpoint...')
-        # print(await test_models())
+        print('Checking the models endpoint...')
+        print(await test_models())
 
     except Exception as exc:
         print('[red]Error: ' + str(exc))

playground/functioncalling.py

@@ -1,82 +0,0 @@
-import os
-import json
-import openai
-
-from dotenv import load_dotenv
-
-load_dotenv()
-
-openai.api_base = 'http://localhost:2332/v1'
-openai.api_key = os.environ['NOVA_KEY']
-
-# Example dummy function hard coded to return the same weather
-# In production, this could be your backend API or an external API
-def get_current_weather(location, unit='fahrenheit'):
-    """Get the current weather in a given location"""
-    weather_info = {
-        'location': location,
-        'temperature': '72',
-        'unit': unit,
-        'forecast': ['sunny', 'windy'],
-    }
-    return json.dumps(weather_info)
-
-def run_conversation():
-    # Step 1: send the conversation and available functions to GPT
-    messages = [{'role': 'user', 'content': 'What\'s the weather like in Boston?'}]
-    functions = [
-        {
-            'name': 'get_current_weather',
-            'description': 'Get the current weather in a given location',
-            'parameters': {
-                'type': 'object',
-                'properties': {
-                    'location': {
-                        'type': 'string',
-                        'description': 'The city and state, e.g. San Francisco, CA',
-                    },
-                    'unit': {'type': 'string', 'enum': ['celsius', 'fahrenheit']},
-                },
-                'required': ['location'],
-            },
-        }
-    ]
-    response = openai.ChatCompletion.create(
-        model='gpt-3.5-turbo-0613',
-        messages=messages,
-        functions=functions,
-        function_call='auto',  # auto is default, but we'll be explicit
-    )
-    response_message = response['choices'][0]['message']
-
-    # Step 2: check if GPT wanted to call a function
-    if response_message.get('function_call'):
-        # Step 3: call the function
-        # Note: the JSON response may not always be valid; be sure to handle errors
-        available_functions = {
-            'get_current_weather': get_current_weather,
-        }  # only one function in this example, but you can have multiple
-        function_name = response_message['function_call']['name']
-        fuction_to_call = available_functions[function_name]
-        function_args = json.loads(response_message['function_call']['arguments'])
-        function_response = fuction_to_call(
-            location=function_args.get('location'),
-            unit=function_args.get('unit'),
-        )
-
-        # Step 4: send the info on the function call and function response to GPT
-        messages.append(response_message)  # extend conversation with assistant's reply
-        messages.append(
-            {
-                'role': 'function',
-                'name': function_name,
-                'content': function_response,
-            }
-        )  # extend conversation with function response
-        second_response = openai.ChatCompletion.create(
-            model='gpt-3.5-turbo-0613',
-            messages=messages,
-        )  # get a new response from GPT where it can see the function response
-        return second_response
-
-print(run_conversation())

setup.py

@@ -0,0 +1,23 @@
+from fastapi import FastAPI
+from fastapi.responses import PlainTextResponse
+from fastapi.requests import Request
+from fastapi.responses import Response
+from slowapi import Limiter, _rate_limit_exceeded_handler
+from slowapi.util import get_remote_address
+from slowapi.errors import RateLimitExceeded
+
+limiter = Limiter(key_func=lambda: "test", default_limits=["5/minute"])
+app = FastAPI()
+app.state.limiter = limiter
+app.add_exception_handler(RateLimitExceeded, _rate_limit_exceeded_handler)
+
+# Note: the route decorator must be above the limit decorator, not below it
+@app.get("/home")
+@limiter.limit("5/minute")
+async def homepage(request: Request):
+    return PlainTextResponse("test")
+
+@app.get("/mars")
+@limiter.limit("5/minute")
+async def homepage(request: Request, response: Response):
+    return {"key": "value"}