Spaces:
Running
Running
| from starlette.responses import JSONResponse, FileResponse | |
| from gradio.data_classes import FileData, GradioModel | |
| from sse_starlette.sse import EventSourceResponse | |
| from typing import (List, Tuple, Optional) | |
| from fastapi import FastAPI, Request | |
| 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 sys | |
| import os | |
| # --- === CONFIG === --- | |
| ENV_HANDLE = "env"#or "url on env" | |
| IMAGE_HANDLE = "url"# or "base64" | |
| API_BASE = "env"# or "openai" | |
| api_key = os.environ['OPENAI_API_KEY'] | |
| base_url = os.environ.get('OPENAI_BASE_URL', "https://api.openai.com/v1") | |
| def_models = '["gpt-3.5-turbo", "gpt-3.5-turbo-0125", "gpt-3.5-turbo-1106", "gpt-3.5-turbo-16k", "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", "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-mini", "gpt-4o-mini-2024-07-18"]' | |
| # --- === 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 | |
| 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 | |
| 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 | |
| if API_BASE == "env": | |
| try: | |
| response = requests.get(f"{base_url}/models", headers={"Authorization": f"Bearer {api_key}"}) | |
| response.raise_for_status() | |
| if not ('data' in response.json()): | |
| base_url = "https://api.openai.com/v1" | |
| except Exception as e: | |
| print(f"Error testing API endpoint: {e}") | |
| else: | |
| base_url = "https://api.openai.com/v1" | |
| def loadModels(): | |
| global models, modelList | |
| 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] | |
| } | |
| 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( | |
| f"{base_url}/moderations", | |
| headers={ | |
| "Content-Type": "application/json", | |
| "Authorization": f"Bearer {api_key}" | |
| }, | |
| json={"input": encodeChat(messages)} | |
| ) | |
| response.raise_for_status() | |
| moderation_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/moderations", | |
| headers={ | |
| "Content-Type": "application/json", | |
| "Authorization": f"Bearer {api_key}" | |
| }, | |
| json={"input": encodeChat(messages)} | |
| ) | |
| response.raise_for_status() | |
| moderation_result = response.json() | |
| except requests.exceptions.RequestException as e: | |
| print(f"Error during moderation request to fallback URL: {e}") | |
| return False | |
| try: | |
| if any(result["flagged"] for result in moderation_result["results"]): | |
| return moderation_result | |
| except KeyError: | |
| if moderation_result["flagged"]: | |
| return moderation_result | |
| return False | |
| async def streamChat(params): | |
| async with aiohttp.ClientSession() as session: | |
| try: | |
| async with session.post(f"{base_url}/chat/completions", headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}, json=params) as r: | |
| r.raise_for_status() | |
| async for line in r.content: | |
| if line: | |
| line_str = line.decode('utf-8') | |
| if line_str.startswith("data: "): | |
| line_str = line_str[6:].strip() | |
| if line_str == "[DONE]": | |
| continue | |
| try: | |
| message = json.loads(line_str) | |
| yield message | |
| except json.JSONDecodeError: | |
| continue | |
| except aiohttp.ClientError: | |
| try: | |
| async with session.post("https://api.openai.com/v1/chat/completions", headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}, json=params) as r: | |
| r.raise_for_status() | |
| async for line in r.content: | |
| if line: | |
| line_str = line.decode('utf-8') | |
| if line_str.startswith("data: "): | |
| line_str = line_str[6:].strip() | |
| if line_str == "[DONE]": | |
| continue | |
| try: | |
| message = json.loads(line_str) | |
| yield message | |
| except json.JSONDecodeError: | |
| continue | |
| except aiohttp.ClientError: | |
| return | |
| 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", "gpt-4o", "gpt-4o-2024-05-13", "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", "gpt-4o", "gpt-4o-2024-05-13", "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[ | |
| Optional[MultimodalMessage], | |
| Optional[MultimodalMessage], | |
| ]], | |
| system_message, | |
| model_name, | |
| max_tokens, | |
| temperature, | |
| top_p, | |
| seed, | |
| random_seed, | |
| consent | |
| ): | |
| if not consent: | |
| yield """[CONSENT] You must agree to the terms to use this application. | |
| ``` | |
| By using our application, which integrates with OpenAI's API, you acknowledge and agree to the following terms regarding the data you provide: | |
| 1. Data Collection: This application may collect data shared through the Gradio endpoint or the API endpoint. | |
| 2. Privacy: Please avoid sharing any personal information. | |
| 3. Data Retention and Removal: Data files are deleted every 30 days, and the API is restarted to ensure data clearance. | |
| 4. Scope of Data Collected: The data collected includes model settings, chat history, and responses from the model. This applies only to 'chat' endpoints (Gradio and API) and excludes moderation checks. | |
| 5. Data Usage: The collected data is periodically reviewed, and if any concerning activity is detected, the code is updated accordingly. | |
| By 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. | |
| ``` | |
| To agree to user consent, please do the followings: | |
| 1. Scroll down to find the section labeled 'Additional Inputs' below this page. | |
| 2. Find and click the check box that says 'User Consent [I agree to the terms and conditions. (can't make a button for it)]'. | |
| 3. After agreeing, click either the `🗑️ Clear` button, the `↩️ Undo` button, or the `🔄 Retry` button located above the message input area.""" | |
| return | |
| messages = [{"role": "system", "content": system_message}] | |
| for val in history: | |
| 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) | |
| user_message = handleMultimodalData(model_name, "user", message) | |
| if user_message: | |
| messages.append(user_message) | |
| mode = moderate(messages) | |
| 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 | |
| response = "" | |
| completion = 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 | |
| }) | |
| async for token in completion: | |
| response += token['choices'][0]['delta'].get("content", "") | |
| yield response | |
| loadModels();checkModels();loadENV(); | |
| lastUpdateMessage = "Strict moderation added." | |
| demo = gr.ChatInterface( | |
| respond, | |
| title="GPT-4O-mini", | |
| description=f"A simple proxy to OpenAI!<br/>You can use this space as a proxy! click [here](/api/v1/docs) to view the documents. <strong>[last update: {lastUpdateMessage}]</strong><br/>Also you can only submit images to vision/4o models but can submit txt/code/etc. files to all models.", | |
| multimodal=True, | |
| additional_inputs=[ | |
| gr.Textbox(value="You are a helpful assistant.", label="System message"), | |
| gr.Dropdown(choices=models, value="gpt-4o-mini-2024-07-18", 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="User Consent [I agree to the terms and conditions. (can't make a button for it)]", value=False) | |
| ], | |
| ) | |
| app = FastAPI() | |
| def html(): | |
| return FileResponse("index.html") | |
| async def test_endpoint(): | |
| return JSONResponse(content=modelList) | |
| async def chat_completion(request: Request): | |
| try: | |
| body = await request.json() | |
| if not body.get("messages") or not body.get("model"): | |
| return JSONResponse(content={"error": { "code": "MISSING_VALUE", "message": "Both 'messages' and 'model' are required fields."}}, status_code=400) | |
| if not body.get("model") in models: | |
| return JSONResponse(content={"error": { "code": "INVALID_MODEL", "message": "The model name provided in the request does not exists in predefined list of models."}}, status_code=400) | |
| params = { | |
| key: value for key, value in { | |
| "model": body.get("model"), | |
| "messages": body.get("messages"), | |
| "max_tokens": body.get("max_tokens"), | |
| "temperature": body.get("temperature"), | |
| "top_p": body.get("top_p"), | |
| "frequency_penalty": body.get("frequency_penalty"), | |
| "logit_bias": body.get("logit_bias"), | |
| "logprobs": body.get("logprobs"), | |
| "top_logprobs": body.get("top_logprobs"), | |
| "n": body.get("n"), | |
| "presence_penalty": body.get("presence_penalty"), | |
| "response_format": body.get("response_format"), | |
| "seed": body.get("seed"), | |
| "service_tier": body.get("service_tier"), | |
| "stop": body.get("stop"), | |
| "stream": body.get("stream"), | |
| "stream_options": body.get("stream_options"), | |
| "tools": body.get("tools"), | |
| "tool_choice": body.get("tool_choice"), | |
| "parallel_tool_calls": body.get("parallel_tool_calls"), | |
| "user": rnd(), | |
| }.items() if value is not None | |
| } | |
| if body.get("stream"): | |
| async def event_generator(): | |
| async for event in streamChat(params): | |
| yield json.dumps(event) | |
| return EventSourceResponse(event_generator()) | |
| else: | |
| try: | |
| response = requests.post(f"{base_url}/chat/completions", headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}, json=params) | |
| response.raise_for_status() | |
| except requests.exceptions.RequestException: | |
| try: | |
| response = requests.post("https://api.openai.com/v1/chat/completions", headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}, json=params) | |
| response.raise_for_status() | |
| except requests.exceptions.RequestException as e: | |
| return JSONResponse(content={"error": { "code": "SERVER_ERROR", "message": str(e)}}, status_code=400) | |
| completion = response.json() | |
| return JSONResponse(content=completion) | |
| except Exception as e: | |
| return JSONResponse(content={"error": { "code": "SERVER_ERROR", "message": str(e)}}, status_code=400) | |
| app = gr.mount_gradio_app(app, demo, path="/") | |
| class ArgParser(argparse.ArgumentParser): | |
| def __init__(self, *args, **kwargs): | |
| super(ArgParser, self).__init__(*args, **kwargs) | |
| self.add_argument("-s", "--server", type=str, default="0.0.0.0") | |
| self.add_argument("-p", "--port", type=int, default=7860) | |
| self.add_argument("-d", "--dev", default=False, action="store_true") | |
| self.args = self.parse_args(sys.argv[1:]) | |
| if __name__ == "__main__": | |
| args = ArgParser().args | |
| 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) | |