Update app.py

app.py

@@ -1,23 +1,165 @@
-from starlette.responses import HTMLResponse
+from starlette.responses import JSONResponse, FileResponse, HTMLResponse
+from gradio.data_classes import FileData, GradioModel
+from sse_starlette.sse import EventSourceResponse
+from typing import (List, Tuple, Optional)
 from fastapi import FastAPI, Request
-from typing import List
 import gradio as gr
+import threading
 import requests
 import argparse
 import aiohttp
 import uvicorn
 import random
 import string
+import base64
 import json
+import time
 import math
 import sys
 import os
 
-API_BASE = "env"
-api_key = os.environ['API_KEY']
+# --- === CONFIG === ---
+
+ENV_HANDLE = "env"#or "url on env"
+IMAGE_HANDLE = "url"# or "base64"
+API_BASE = "env"# or "openai"
+api_key = os.environ['API_API_KEY']
 oai_api_key = os.environ['OPENAI_API_KEY']
 base_url = os.environ.get('OPENAI_BASE_URL', "https://api.openai.com/v1")
-def_models = '["gpt-4", "gpt-4-0125-preview", "gpt-4-0613", "gpt-4-1106-preview", "gpt-4-turbo", "gpt-4-turbo-2024-04-09", "gpt-4-turbo-preview", "chatgpt-4o-latest", "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06", "gpt-4o-mini", "gpt-4o-mini-2024-07-18"]'
+# Will not add O1-mini, and O1-preview into the default, as it requeires TIER-5 sub on OpenAI's API.
+# But if you wanna add O1 just remove this comment line and comment the other
+def_models = '["gpt-4", "gpt-4-0125-preview", "gpt-4-0314", "gpt-4-0613", "gpt-4-1106-preview", "gpt-4-1106-vision-preview", "gpt-4-32k-0314", "gpt-4-turbo", "gpt-4-turbo-2024-04-09", "gpt-4-turbo-preview", "gpt-4-vision-preview", "chatgpt-4o-latest", "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06", "gpt-4o-mini", "gpt-4o-mini-2024-07-18", "o1-mini", "o1-mini-2024-09-12", "o1-preview", "o1-preview-2024-09-12"]'
+# def_models = '["gpt-4", "gpt-4-0125-preview", "gpt-4-0314", "gpt-4-0613", "gpt-4-1106-preview", "gpt-4-1106-vision-preview", "gpt-4-32k-0314", "gpt-4-turbo", "gpt-4-turbo-2024-04-09", "gpt-4-turbo-preview", "gpt-4-vision-preview", "chatgpt-4o-latest", "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06", "gpt-4o-mini", "gpt-4o-mini-2024-07-18"]'
+fakeToolPrompt = """[System: You have ability to generate images, via tools provided to you by system.
+To call a tool you need to write a json in a empty line; like writing it at the end of message.
+To generate a image; you need to follow this example JSON:
+{"tool": "imagine", "isCall": true, "prompt": "golden retriever sitting comfortably on a luxurious, modern couch. The retriever should look relaxed and content, with fluffy fur and a friendly expression. The couch should be stylish, possibly with elegant details like cushions and a soft texture that complements the dog's golden coat"}
+> 'tool' variable is used to define which tool you are calling
+> 'isCall' used to confirm that you are calling that function and not showing it for example
+> 'prompt' the image prompt that will be given to image generation model.
+
+Here's few more example so you can under stand better
+To show as an example>
+    {"tool": "imagine", "isCall": false, "prompt": "futuristic robot playing chess against a human, with the robot confidently strategizing its next move while the human looks thoughtful and slightly perplexed"}
+    {"tool": "imagine", "isCall": false, "prompt": "colorful parrot perched on a wooden fence, pecking at a vibrant tropical fruit. The parrot's feathers should be bright and varied, with greens, blues, and reds. The background should feature a lush, green jungle with scattered rays of sunlight"}
+    {"tool": "imagine", "isCall": false, "prompt": "fluffy white cat lounging on a sunlit windowsill, with a gentle breeze blowing through the curtains"}
+To actually use the tool>
+    {"tool": "imagine", "isCall": true, "prompt": "golden retriever puppy happily playing with a red ball in a sunny park. The park should have green grass, a few trees in the background, and a clear blue sky"}
+    {"tool": "imagine", "isCall": true, "prompt": "red panda balancing on a tightrope, with a city skyline in the background"}
+    {"tool": "imagine", "isCall": true, "prompt": "corgi puppy wearing sunglasses and a red bandana, sitting on a beach chair under a colorful beach umbrella, with a surfboard leaning against the chair and the ocean waves in the background"}
+In chat use examples:
+1.
+Alright, here's an image of an hedgehog riding a skateboard:
+{"tool": "imagine", "isCall": true, "prompt": "A hedgehog riding a skateboard in a suburban park"}
+2.
+Okay, here's the image you requested:
+{"tool": "imagine", "isCall": true, "prompt": "Persian cat lounging on a plush velvet sofa in a cozy, sunlit living room. The cat is elegantly poised, with a calm and regal demeanor, its fur meticulously groomed and slightly fluffed up as it rests comfortably"}
+3.
+This is how i generate images:
+{"tool": "imagine", "isCall": false, "prompt": "image prompt"}
+4. (Do not do this, this would block the user from seeing the image.)
+Alright! Here's an image of a whimsical scene featuring a cat wearing a wizard hat, casting a spell with sparkling magic in a mystical forest.] ```
+{"tool": "imagine", "isCall": true, "prompt": "A playful cat wearing a colorful wizard hat, surrounded by magical sparkles and glowing orbs in a mystical forest. The cat looks curious and mischievous, with its tail swishing as it focuses on casting a spell. The forest is lush and enchanting, with vibrant flowers and soft, dappled sunlight filtering through the trees."}
+5. (if in any case the user asks for the prompt)
+Sure here's the prompt i wrote to generate the image below: `A colorful bird soaring through a bustling city skyline. The bird should have vibrant feathers, contrasting against the modern buildings and blue sky. Below, the city is alive with activity, featuring tall skyscrapers, busy streets, and small parks, creating a dynamic urban scene.`
+]""";
+calcPrompt = """[System: You have ability to calculate math problems (formated on python), via tools provided to you by system.
+To call a tool you need to write a json in a empty line; like writing it at the end of message.
+To use calculator; you need to follow this example JSON:
+{"tool": "calc", "isCall": true, "prompt": "math.pi * 5"}
+> 'tool' variable is used to define which tool you are calling
+> 'isCall' used to confirm that you are calling that function and not showing it for example
+> 'prompt' the math that will be done via python.
+
+Here's few more example so you can under stand better
+To show as an example>
+    {"tool": "calc", "isCall": false, "prompt": "math.sqrt(16)"}
+    {"tool": "calc", "isCall": false, "prompt": "math.pow(2, 3)"}
+    {"tool": "calc", "isCall": false, "prompt": "math.sin(math.pi / 2)"}
+To actually use the tool>
+    {"tool": "calc", "isCall": true, "prompt": "math.factorial(5)"}
+    {"tool": "calc", "isCall": true, "prompt": "math.log(100, 10)"}
+    {"tool": "calc", "isCall": true, "prompt": "math.cos(0)"}
+In chat use examples:
+1.
+Please, wait while I calculate 2+2...
+{"tool": "calc", "isCall": false, "prompt": "2+2"}
+2.
+Plase, wait while I calculate the square root of 25...
+{"tool": "calc", "isCall": true, "prompt": "math.sqrt(25)"}
+3.
+This is how I perform calculations:
+{"tool": "calc", "isCall": false, "prompt": "math.pow(3, 2)"}
+4. (Do not do this, this would block the user from seeing the result.)
+Alright! Here's the result of a complex calculation involving trigonometry and logarithms. ```
+{"tool": "calc", "isCall": true, "prompt": "math.sin(math.pi / 4) + math.log(10, 10)"}
+]""";
+searchPrompt = """[System: You have ability to search queries on a search engine, via tools provided to you by system.
+(Warning: Each search call can take up to 30 or more seconds. Only one search function can be called per round. If a response has already been received, the system will answer based on that response. If the query needs to be searched again, the system will ask the user if they want to requery.)
+To call a tool you need to write a json in a empty line; like writing it at the end of message.
+To look up queries; you need to follow this example JSON:
+{"tool": "search", "isCall": true, "prompt": "What is the latest news on climate change?"}
+> 'tool' variable is used to define which tool you are calling
+> 'isCall' used to confirm that you are calling that function and not showing it for example
+> 'prompt' the query that will be searched.
+
+Here's a few more examples so you can understand better
+To show as an example>
+    {"tool": "search", "isCall": false, "prompt": "How to bake a chocolate cake?"}
+    {"tool": "search", "isCall": false, "prompt": "What are the symptoms of the flu?"}
+    {"tool": "search", "isCall": false, "prompt": "Best practices for remote work"}
+To actually use the tool>
+    {"tool": "search", "isCall": true, "prompt": "How to invest in stocks?"}
+    {"tool": "search", "isCall": true, "prompt": "What is the current status of the Mars rover?"}
+    {"tool": "search", "isCall": true, "prompt": "Latest advancements in AI technology"}
+In chat use examples:
+1.
+Please, wait while I search for the latest trends in technology...
+{"tool": "search", "isCall": false, "prompt": "Latest trends in technology"}
+2.
+Please, wait while I search for the best ways to improve mental health...
+{"tool": "search", "isCall": true, "prompt": "Best ways to improve mental health"}
+3.
+This is how I perform searches:
+{"tool": "search", "isCall": false, "prompt": "How to start a garden?"}
+4. (Do not do this, this would block the user from seeing the result.)
+Alright! Here's the result of a search on the impact of social media on teenagers. ```
+{"tool": "search", "isCall": true, "prompt": "Impact of social media on teenagers"}
+]""";
+
+# --- === CONFIG === ---
+
+def loadENV():
+    def worker():
+        while True:
+            if ENV_HANDLE == "url on env":
+                try:
+                    response = requests.get(os.environ["ENV_URL"])
+                    response.raise_for_status()
+                    env_data = response.json()
+                    for key, value in env_data.items():
+                        os.environ[key] = value
+                    handleApiKeys()
+                    checkModels()
+                    loadModels()
+                except Exception as e:
+                    print(f"Error loading environment variables: {e}")
+            time.sleep(180)
+
+    if ENV_HANDLE == "url on env":
+        try:
+            response = requests.get(os.environ["ENV_URL"])
+            response.raise_for_status()
+            env_data = response.json()
+            for key, value in env_data.items():
+                os.environ[key] = value
+            handleApiKeys()
+            checkModels()
+            loadModels()
+        except Exception as e:
+            print(f"Error loading environment variables: {e}")
+
+    threading.Thread(target=worker, daemon=True).start()
 
 def checkModels():
     global base_url
@@ -36,9 +178,13 @@ def checkModels():
 
 def loadModels():
     global models, modelList
-    models = json.loads(def_models)
+    try:
+        models = json.loads(os.environ.get('OPENAI_API_MODELS', def_models))
+    except json.JSONDecodeError:
+        models = json.loads(def_models)
+
     models = sorted(models)
-    
+
     modelList = {
         "object": "list",
         "data": [{"id": v, "object": "model", "created": 0, "owned_by": "system"} for v in models]
@@ -68,15 +214,15 @@ def handleApiKeys():
         except Exception as e:
             raise RuntimeError(f"Current API key is not valid or an actual error happened: {e}")
 
-def encodeChat(messages):
-    output = []
-    for message in messages:
-        role = message['role']
-        name = f" [{message['name']}]" if 'name' in message else ''
-        content = message['content']
-        formatted_message = f"<|im_start|>{role}{name}\n{content}<|end_of_text|>"
-        output.append(formatted_message)
-    return "\n".join(output)
+def safe_eval(expression):
+    print(expression)
+    allowed_names = {name: obj for name, obj in math.__dict__.items() if not name.startswith("__")}
+    allowed_names['math'] = math
+    code = compile(expression, "<string>", "eval")
+    for name in code.co_names:
+        if name not in allowed_names and name != 'math':
+            raise NameError(f"Use of {name} is not allowed")
+    return eval(code, {"__builtins__": {}}, allowed_names)
 
 def get_api_key(call='api_key'):
     if call == 'api_key':
@@ -90,6 +236,16 @@ def get_api_key(call='api_key'):
         return random.choice(key.split(','))
     return key
 
+def encodeChat(messages):
+    output = []
+    for message in messages:
+        role = message['role']
+        name = f" [{message['name']}]" if 'name' in message else ''
+        content = message['content']
+        formatted_message = f"<|im_start|>{role}{name}\n{content}<|end_of_text|>"
+        output.append(formatted_message)
+    return "\n".join(output)
+
 def moderate(messages):
     try:
         response = requests.post(
@@ -125,10 +281,36 @@ def moderate(messages):
     except KeyError:
         if moderation_result["flagged"]:
             return moderation_result
-    
+
     return False
 
 async def streamChat(params):
+    if params.get("model") in ["o1-mini", "o1-mini-2024-09-12", "o1-preview", "o1-preview-2024-09-12"]:
+        if "temperature" in params:
+            del params["temperature"]
+        if "top_p" in params:
+            del params["top_p"]
+        if "max_tokens" in params:
+            params["max_completion_tokens"] = params.pop("max_tokens")
+        for message in params.get("messages", []):
+            if message["role"] == "system":
+                params["messages"].remove(message)
+        params["stream"] = False;
+        async with aiohttp.ClientSession() as session:
+            try:
+                async with session.post(f"{base_url}/chat/completions", headers={"Authorization": f"Bearer {get_api_key(call='api_key')}", "Content-Type": "application/json"}, json=params) as r:
+                    r.raise_for_status()
+                    response_data = await r.json()
+                    yield {"choices": [{"delta": {"content": response_data["choices"][0]["message"]["content"]}}]}
+            except aiohttp.ClientError:
+                try:
+                    async with session.post("https://api.openai.com/v1/chat/completions", headers={"Authorization": f"Bearer {get_api_key(call='oai_api_key')}", "Content-Type": "application/json"}, json=params) as r:
+                        r.raise_for_status()
+                        response_data = await r.json()
+                        yield {"choices": [{"delta": {"content": response_data["choices"][0]["message"]["content"]}}]}
+                except aiohttp.ClientError:
+                    return
+    else:
         async with aiohttp.ClientSession() as session:
             try:
                 async with session.post(f"{base_url}/chat/completions", headers={"Authorization": f"Bearer {get_api_key(call='api_key')}", "Content-Type": "application/json"}, json=params) as r:
@@ -164,31 +346,187 @@ async def streamChat(params):
                 except aiohttp.ClientError:
                     return
 
+def imagine(prompt):
+    try:
+        response = requests.post(
+            f"{base_url}/images/generations",
+            headers={
+                "Content-Type": "application/json",
+                "Authorization": f"Bearer {get_api_key(call='api_key')}"
+            },
+            json={
+                "model": "dall-e-3",
+                "prompt": prompt,
+                "quality": "hd",
+            }
+        )
+        response.raise_for_status()
+        result = response.json()
+    except requests.exceptions.RequestException as e:
+        print(f"Error during moderation request to {base_url}: {e}")
+        try:
+            response = requests.post(
+                "https://api.openai.com/v1/images/generations",
+                headers={
+                    "Content-Type": "application/json",
+                    "Authorization": f"Bearer {get_api_key(call='oai_api_key')}"
+                },
+                json={
+                        "model": "dall-e-3",
+                        "prompt": prompt,
+                        "quality": "hd",
+                }
+            )
+            response.raise_for_status()
+            result = response.json()
+        except requests.exceptions.RequestException as e:
+            print(f"Error during moderation request to fallback URL: {e}")
+            return False
+
+    return result.get('data', [{}])[0].get('url')
+
+def searchEngine(query):
+    ### This /search endpoint is custom made, OpenAI does not have it.
+        ### If you dupelicate this space, please either try to find another API or make one yourself.
+    response = requests.get(f"{base_url}/search?query={requests.utils.quote(query)}")
+    response.raise_for_status()
+    response_data = response.json()
+    return response_data.get("choices", [{}])[0].get("message", {}).get("content", "")
+
 def rnd(length=8):
     letters = string.ascii_letters + string.digits
     return ''.join(random.choice(letters) for i in range(length))
 
+def handleMultimodalData(model, role, data):
+    if isinstance(data, tuple):
+        data = data[0]
+
+    if isinstance(data, FileData):
+        if data.mime_type.startswith("image/"):
+            if IMAGE_HANDLE == "base64":
+                with open(data.path, "rb") as image_file:
+                    b64image = base64.b64encode(image_file.read()).decode('utf-8')
+                    image_file.close()
+                    return {"role": role, "content": [{"type": "image_url", "image_url": {"url": "data:" + data.mime_type + ";base64," + b64image}}]}
+            else:
+                return {"role": role, "content": [{"type": "image_url", "image_url": {"url": data.url}}]}
+        elif data.mime_type.startswith("text/") or data.mime_type.startswith("application/"):
+            try:
+                with open(data.path, "rb") as data_file:
+                    return {"role": role, "content": "[System: This message contains file.]\n\n<|file_start|>" + data.orig_name + "\n" + data_file.read().decode('utf-8') + "<|file_end|>"}
+            except UnicodeDecodeError:
+                pass
+    elif isinstance(data, str):
+        return {"role": role, "content": data}
+    elif hasattr(data, 'files') and data.files and len(data.files) > 0 and model in {"gpt-4-1106-vision-preview", "gpt-4-vision-preview", "gpt-4-turbo", "chatgpt-4o-latest", "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06", "gpt-4o-mini", "gpt-4o-mini-2024-07-18"}:
+        result, handler, hasFoundFile = [], ["[System: This message contains files; the system will be splitting it.]"], False
+        for file in data.files:
+            if file.mime_type.startswith("image/"):
+                if IMAGE_HANDLE == "base64":
+                    with open(file.path, "rb") as image_file:
+                        result.append({"type": "image_url", "image_url": {"url": "data:" + file.mime_type + ";base64," + base64.b64encode(image_file.read()).decode('utf-8')}})
+                        image_file.close()
+                else:
+                    result.append({"type": "image_url", "image_url": {"url": file.url}})
+            if file.mime_type.startswith("text/") or file.mime_type.startswith("application/"):
+                hasFoundFile = True
+                try:
+                    with open(file.path, "rb") as data_file:
+                        handler.append("<|file_start|>" + file.orig_name + "\n" + data_file.read().decode('utf-8') + "<|file_end|>")
+                except UnicodeDecodeError:
+                    continue
+        if hasFoundFile:
+            handler.append(data.text)
+            return {"role": role, "content": [{"type": "text", "text": "\n\n".join(handler)}] + result}
+        else:
+            return {"role": role, "content": [{"type": "text", "text": data.text}] + result}
+    elif hasattr(data, 'files') and data.files and len(data.files) > 0 and not (model in {"gpt-4-1106-vision-preview", "gpt-4-vision-preview", "gpt-4-turbo", "chatgpt-4o-latest", "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06", "gpt-4o-mini", "gpt-4o-mini-2024-07-18"}):
+        handler, hasFoundFile = ["[System: This message contains files; the system will be splitting it.]"], False
+        for file in data.files:
+            if file.mime_type.startswith("text/") or file.mime_type.startswith("application/"):
+                hasFoundFile = True
+                try:
+                    with open(file.path, "rb") as data_file:
+                        return {"role": role, "content":  "<|file_start|>" + file.orig_name + "\n" + data_file.read().decode('utf-8') + "<|file_end|>"}
+                except UnicodeDecodeError:
+                    continue
+    else:
+        if isinstance(data, tuple):
+            return {"role": role, "content": str(data)}
+        return {"role": role, "content": getattr(data, 'text', str(data))}
+
+class FileMessage(GradioModel):
+    file: FileData
+    alt_text: Optional[str] = None
+
+class MultimodalMessage(GradioModel):
+    text: Optional[str] = None
+    files: Optional[List[FileMessage]]
 
 async def respond(
     message,
-    history: list[tuple[str, str]],
+    history: List[Tuple[
+        Optional[MultimodalMessage],
+        Optional[MultimodalMessage],
+    ]],
+    system_message,
     model_name,
     max_tokens,
     temperature,
     top_p,
+    seed,
+    random_seed,
+    fakeTool,
+    calcBeta,
+    searchBeta,
+        betterSystemPrompt
 ):
     messages = [];
+    if fakeTool:
+        messages.append({"role": "system", "content": fakeToolPrompt});
+    if calcBeta:
+        messages.append({"role": "system", "content": calcPrompt});
+    if searchBeta:
+        messages.append({"role": "system", "content": searchPrompt});
+    if betterSystemPrompt:
+        messages.append({"role": "system", "content": f"You are a helpful assistant. You are an OpenAI GPT model named {model_name}. The current time is {time.strftime('%Y-%m-%d %H:%M:%S')}. Please adhere to OpenAI's usage policies and guidelines. Ensure your responses are accurate, respectful, and within the scope of OpenAI's rules."});
+    else:
+        messages.append({"role": "system", "content": system_message});
 
     for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
+        if val[0] is not None:
+            user_message = handleMultimodalData(model_name, "user", val[0])
+            if user_message:
+                messages.append(user_message)
+        if val[1] is not None:
+            assistant_message = handleMultimodalData(model_name, "assistant", val[1])
+            if assistant_message:
+                messages.append(assistant_message)
 
     if message:
-        mode = moderate(messages)
+        user_message = handleMultimodalData(model_name, "user", message)
+        if user_message:
+            messages.append(user_message)
+            mode = moderate([user_message])
+            if mode:
+                reasons = []
+                categories = mode[0].get('categories', {}) if isinstance(mode, list) else mode.get('categories', {})
+                for category, flagged in categories.items():
+                    if flagged:
+                        reasons.append(category)
+                if reasons:
+                    yield "[MODERATION] I'm sorry, but I can't assist with that.\n\nReasons:\n```\n" + "\n".join([f"{i+1}. {reason}" for i, reason in enumerate(reasons)]) + "\n```"
+                else:
+                    yield "[MODERATION] I'm sorry, but I can't assist with that."
+                return
+
+    async def handleResponse(completion, prefix="", image_count=0, didSearchedAlready=False):
+        response = ""
+        isRequeryNeeded = False
+        async for token in completion:
+            response += token['choices'][0]['delta'].get("content", token['choices'][0]['delta'].get("refusal", ""))
+            yield f"{prefix}{response}"
+        mode = moderate([handleMultimodalData(model_name, "user", message),{"role": "assistant", "content": response}])
         if mode:
             reasons = []
             categories = mode[0].get('categories', {}) if isinstance(mode, list) else mode.get('categories', {})
@@ -200,31 +538,95 @@ async def respond(
             else:
                 yield "[MODERATION] I'm sorry, but I can't assist with that."
             return
-    
-    response = ""
-    async for token in streamChat({
+        for line in response.split('\n'):
+            try:
+                data = json.loads(line)
+                if isinstance(data, dict) and data.get("tool") == "imagine" and data.get("isCall") and "prompt" in data:
+                    if image_count < 4:
+                        image_count += 1
+                        def fetch_image_url(prompt, line):
+                            image_url = imagine(prompt)
+                            return line, f'<img src="{image_url}" alt="{prompt}" width="512"/>'
+
+                        def replace_line_in_response(line, replacement):
+                            nonlocal response
+                            response = response.replace(line, replacement)
+
+                        thread = threading.Thread(target=lambda: replace_line_in_response(*fetch_image_url(data["prompt"], line)))
+                        thread.start()
+                        thread.join()
+                    else:
+                        response = response.replace(line, f'[System: 4 image per message limit; prompt asked: `{data["prompt"]}]`')
+                    yield f"{prefix}{response}"
+                elif isinstance(data, dict) and data.get("tool") == "calc" and data.get("isCall") and "prompt" in data:
+                    isRequeryNeeded = True
+                    try:
+                        result = safe_eval(data["prompt"])
+                        response = response.replace(line, f'[System: `{data["prompt"]}` === `{result}`]')
+                    except Exception as e:
+                        response = response.replace(line, f'[System: Error in calculation; `{e}`]')
+                    yield f"{prefix}{response}"
+                elif isinstance(data, dict) and data.get("tool") == "search" and data.get("isCall") and "prompt" in data:
+                    isRequeryNeeded = True
+                    if didSearchedAlready:
+                        response = response.replace(line, f'[System: One search per response is allowed; due to how long and resource it takes; query: `{data["prompt"]}]`]')
+                    else:
+                        try:
+                            result = searchEngine(data["prompt"])
+                            result_escaped = result.replace('`', '\\`')
+                            response = response.replace(line, f'[System: `{data["prompt"]}` ===\n```\n{result_escaped}\n```\n]')
+                            didSearchedAlready = True
+                        except Exception as e:
+                            response = response.replace(line, f'[System: Error in search function; `{e}`]')
+                        yield f"{prefix}{response}"
+                    yield f"{prefix}{response}"
+            except (json.JSONDecodeError, AttributeError, Exception):
+                continue
+            if isRequeryNeeded:
+                messages.append({"role": "assistant", "content": response})
+                async for res in handleResponse(streamChat({
+                    "model": model_name,
+                    "messages": messages,
+                    "max_tokens": max_tokens,
+                    "temperature": temperature,
+                    "top_p": top_p,
+                    "seed": (random.randint(0, 2**32) if random_seed else seed),
+                    "user": rnd(),
+                    "stream": True
+                }), f"{prefix}{response}\n\n", image_count, didSearchedAlready):
+                    yield res
+    async for res in handleResponse(streamChat({
         "model": model_name,
         "messages": messages,
         "max_tokens": max_tokens,
         "temperature": temperature,
         "top_p": top_p,
+        "seed": (random.randint(0, 2**32) if random_seed else seed),
         "user": rnd(),
         "stream": True
-    }):
-        response += token['choices'][0]['delta'].get("content", token['choices'][0]['delta'].get("refusal", ""))
-        yield response
-        
+    })):
+        yield res
+
 
-handleApiKeys();loadModels();checkModels();
+handleApiKeys();loadModels();checkModels();loadENV();
+lastUpdateMessage = "Rolledback the support on O1 model, due to lack of support on params/streaming/etc."
 demo = gr.ChatInterface(
     respond,
     title="gpt-4o-mini",
-    description=f"This is the smaller version of quardo/gpt-4o-mini space. Mainly exists when the main space is down. (Status: the main space is offline cause it is stuck on 'Building' stage)",
+    description=f"A OpenAI API proxy!<br/>View API docs [here](/api/v1/docs) <strong>[Yes you can use this as an API in a simpler manner]</strong>.<br/><strong>[Last update: {lastUpdateMessage}]</strong> Also you can only submit images to vision models; txt/code/etc. to all models.",
+    multimodal=True,
     additional_inputs=[
+        gr.Textbox(value="You are a helpful assistant. You are an OpenAI GPT model. Please adhere to OpenAI's usage policies and guidelines. Ensure your responses are accurate, respectful, and within the scope of OpenAI's rules.", label="System message"),
         gr.Dropdown(choices=models, value="gpt-4o-mini", label="Model"),
         gr.Slider(minimum=1, maximum=4096, value=4096, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.05, label="Temperature"),
         gr.Slider(minimum=0.05, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
+        gr.Slider(minimum=0, maximum=2**32, value=0, step=1, label="Seed"),
+        gr.Checkbox(label="Randomize Seed", value=True),
+        gr.Checkbox(label="FakeTool [Image generation beta]", value=True),
+        gr.Checkbox(label="FakeTool [Calculator beta]", value=True),
+        gr.Checkbox(label="FakeTool [Search engine beta (Warning; each query takes up to 30 seconds)]", value=True),
+        gr.Checkbox(label="Better system prompt (ignores the system prompt set by user.)", value=True),
     ],
     css="footer{display:none !important}",
     head="""<script>if(!confirm("By using our application, which integrates with OpenAI's API, you acknowledge and agree to the following terms regarding the data you provide:\\n\\n1. Data Collection: This application may log the following data through the Gradio endpoint or the API endpoint: message requests (including messages, responses, model settings, and images sent along with the messages), images that were generated (including only the prompt and the image), search tool calls (including query, search results, summaries, and output responses), and moderation checks (including input and output).\\n2. Data Retention and Removal: Data is retained until further notice or until a specific request for removal is made.\\n3. Data Usage: The collected data may be used for various purposes, including but not limited to, administrative review of logs, AI training, and publication as a dataset.\\n4. Privacy: Please avoid sharing any personal information.\\n\\nBy continuing to use our application, you explicitly consent to the collection, use, and potential sharing of your data as described above. If you disagree with our data collection, usage, and sharing practices, we advise you not to use our application."))location.href="/declined";</script>"""
@@ -246,6 +648,10 @@ def test():
         </html>
     """)
 
+@app.get("/api/v1/docs")
+def html():
+    return FileResponse("index.html")
+
 app = gr.mount_gradio_app(app, demo, path="/")
 
 class ArgParser(argparse.ArgumentParser):
@@ -263,6 +669,4 @@ if __name__ == "__main__":
     if args.dev:
         uvicorn.run("__main__:app", host=args.server, port=args.port, reload=True)
     else:
-        uvicorn.run("__main__:app", host=args.server, port=args.port, reload=False)
-
-
+        uvicorn.run("__main__:app", host=args.server, port=args.port, reload=False)