# # SPDX-FileCopyrightText: Hadad # SPDX-License-Identifier: Apache-2.0 # import asyncio import docx import gradio as gr import httpx import json import os import pandas as pd import pdfplumber import pytesseract import random import requests import threading import uuid from PIL import Image from pathlib import Path from pptx import Presentation os.system("apt-get update -q -y && apt-get install -q -y tesseract-ocr tesseract-ocr-eng tesseract-ocr-ind libleptonica-dev libtesseract-dev") INTERNAL_AI_GET_SERVER = os.getenv("INTERNAL_AI_GET_SERVER") INTERNAL_TRAINING_DATA = os.getenv("INTERNAL_TRAINING_DATA") LINUX_SERVER_HOSTS = [host for host in json.loads(os.getenv("LINUX_SERVER_HOST", "[]")) if host] LINUX_SERVER_HOSTS_MARKED = set() LINUX_SERVER_HOSTS_ATTEMPTS = {} LINUX_SERVER_PROVIDER_KEYS = [key for key in json.loads(os.getenv("LINUX_SERVER_PROVIDER_KEY", "[]")) if key] LINUX_SERVER_PROVIDER_KEYS_MARKED = set() LINUX_SERVER_PROVIDER_KEYS_ATTEMPTS = {} AI_TYPES = {f"AI_TYPE_{i}": os.getenv(f"AI_TYPE_{i}") for i in range(1, 7)} RESPONSES = {f"RESPONSE_{i}": os.getenv(f"RESPONSE_{i}") for i in range(1, 10)} MODEL_MAPPING = json.loads(os.getenv("MODEL_MAPPING", "{}")) MODEL_CONFIG = json.loads(os.getenv("MODEL_CONFIG", "{}")) MODEL_CHOICES = list(MODEL_MAPPING.values()) if MODEL_MAPPING else [] DEFAULT_CONFIG = json.loads(os.getenv("DEFAULT_CONFIG", "{}")) META_TAGS = os.getenv("META_TAGS") ALLOWED_EXTENSIONS = json.loads(os.getenv("ALLOWED_EXTENSIONS", "[]")) ACTIVE_CANDIDATE = None def get_available_items(items, marked): available = [item for item in items if item not in marked] random.shuffle(available) return available def marked_item(item, marked, attempts): marked.add(item) attempts[item] = attempts.get(item, 0) + 1 if attempts[item] >= 3: def remove_fail(): marked.discard(item) attempts.pop(item, None) threading.Timer(3600, remove_fail).start() class SessionWithID(requests.Session): def __init__(self): super().__init__() self.session_id = str(uuid.uuid4()) def create_session(): return SessionWithID() def get_model_key(display_name): return next((k for k, v in MODEL_MAPPING.items() if v == display_name), list(MODEL_MAPPING.keys())[0] if MODEL_MAPPING else MODEL_CHOICES[0]) def extract_file_content(file_path): ext = Path(file_path).suffix.lower() content = "" try: if ext == ".pdf": with pdfplumber.open(file_path) as pdf: for page in pdf.pages: text = page.extract_text() if text: content += text + "\n" for table in page.extract_tables(): table_str = "\n".join([", ".join(row) for row in table if row]) content += "\n" + table_str + "\n" elif ext in [".doc", ".docx"]: doc = docx.Document(file_path) for para in doc.paragraphs: content += para.text + "\n" elif ext in [".xlsx", ".xls"]: df = pd.read_excel(file_path) content += df.to_csv(index=False) elif ext in [".ppt", ".pptx"]: prs = Presentation(file_path) for slide in prs.slides: for shape in slide.shapes: if hasattr(shape, "text") and shape.text: content += shape.text + "\n" elif ext in [".png", ".jpg", ".jpeg", ".tiff", ".bmp", ".gif", ".webp"]: pytesseract.pytesseract.tesseract_cmd = "/usr/bin/tesseract" image = Image.open(file_path) content += pytesseract.image_to_string(image) + "\n" else: content = Path(file_path).read_text(encoding="utf-8") except Exception as e: content = f"{file_path}: {e}" return content.strip() async def fetch_response_async(host, provider_key, selected_model, messages, model_config, session_id): timeouts = [60, 80, 120, 240] for timeout in timeouts: try: async with httpx.AsyncClient(timeout=timeout) as client: data = {"model": selected_model, "messages": messages, **model_config} resp = await client.post(host, json={**data, "session_id": session_id}, headers={"Authorization": f"Bearer {provider_key}"}) resp.raise_for_status() resp_json = resp.json() if isinstance(resp_json, dict) and resp_json.get("choices"): choice = resp_json["choices"][0] if choice.get("message") and isinstance(choice["message"].get("content"), str): return choice["message"]["content"] return RESPONSES["RESPONSE_2"] except Exception: continue marked_item(provider_key, LINUX_SERVER_PROVIDER_KEYS_MARKED, LINUX_SERVER_PROVIDER_KEYS_ATTEMPTS) return RESPONSES["RESPONSE_2"] async def chat_with_model_async(history, user_input, selected_model_display, sess): if not get_available_items(LINUX_SERVER_PROVIDER_KEYS, LINUX_SERVER_PROVIDER_KEYS_MARKED) or not get_available_items(LINUX_SERVER_HOSTS, LINUX_SERVER_HOSTS_MARKED): return RESPONSES["RESPONSE_3"] if not hasattr(sess, "session_id"): sess.session_id = str(uuid.uuid4()) selected_model = get_model_key(selected_model_display) model_config = MODEL_CONFIG.get(selected_model, DEFAULT_CONFIG) messages = [{"role": "user", "content": user} for user, _ in history] + [{"role": "assistant", "content": assistant} for _, assistant in history if assistant] if INTERNAL_TRAINING_DATA and MODEL_CHOICES and selected_model_display == MODEL_CHOICES[0]: messages.insert(0, {"role": "system", "content": INTERNAL_TRAINING_DATA}) messages.append({"role": "user", "content": user_input}) global ACTIVE_CANDIDATE if ACTIVE_CANDIDATE: result = await fetch_response_async(ACTIVE_CANDIDATE[0], ACTIVE_CANDIDATE[1], selected_model, messages, model_config, sess.session_id) if result != RESPONSES["RESPONSE_2"]: return result ACTIVE_CANDIDATE = None keys = get_available_items(LINUX_SERVER_PROVIDER_KEYS, LINUX_SERVER_PROVIDER_KEYS_MARKED) hosts = get_available_items(LINUX_SERVER_HOSTS, LINUX_SERVER_HOSTS_MARKED) candidates = [(host, key) for host in hosts for key in keys] random.shuffle(candidates) for host, key in candidates: result = await fetch_response_async(host, key, selected_model, messages, model_config, sess.session_id) if result != RESPONSES["RESPONSE_2"]: ACTIVE_CANDIDATE = (host, key) return result return RESPONSES["RESPONSE_2"] async def respond_async(multi_input, history, selected_model_display, sess): message = {"text": multi_input.get("text", "").strip(), "files": multi_input.get("files", [])} if not message["text"] and not message["files"]: yield history, gr.MultimodalTextbox(value=None, interactive=True), sess return combined_input = "" for file_item in message["files"]: file_path = file_item["name"] if isinstance(file_item, dict) and "name" in file_item else file_item combined_input += f"{Path(file_path).name}\n\n{extract_file_content(file_path)}\n\n" if message["text"]: combined_input += message["text"] history.append([combined_input, ""]) ai_response = await chat_with_model_async(history, combined_input, selected_model_display, sess) history[-1][1] = "" def convert_to_string(data): if isinstance(data, (str, int, float)): return str(data) if isinstance(data, bytes): return data.decode("utf-8", errors="ignore") if isinstance(data, (list, tuple)): return "".join(map(convert_to_string, data)) if isinstance(data, dict): return json.dumps(data, ensure_ascii=False) return repr(data) for character in ai_response: history[-1][1] += convert_to_string(character) await asyncio.sleep(0.0001) yield history, gr.MultimodalTextbox(value=None, interactive=True), sess def change_model(new_model_display): return [], create_session(), new_model_display with gr.Blocks(fill_height=True, fill_width=True, title=AI_TYPES["AI_TYPE_4"], head=META_TAGS) as jarvis: user_history = gr.State([]) user_session = gr.State(create_session()) selected_model = gr.State(MODEL_CHOICES[0] if MODEL_CHOICES else "") chatbot = gr.Chatbot(label=AI_TYPES["AI_TYPE_1"], show_copy_button=True, scale=1, elem_id=AI_TYPES["AI_TYPE_2"]) with gr.Row(): msg = gr.MultimodalTextbox(show_label=False, placeholder=RESPONSES["RESPONSE_5"], interactive=True, file_count="single", file_types=ALLOWED_EXTENSIONS) with gr.Accordion(AI_TYPES["AI_TYPE_6"], open=False): model_dropdown = gr.Dropdown(show_label=False, choices=MODEL_CHOICES, value=MODEL_CHOICES[0]) model_dropdown.change(fn=change_model, inputs=[model_dropdown], outputs=[user_history, user_session, selected_model], show_progress="full") msg.submit(fn=respond_async, inputs=[msg, user_history, selected_model, user_session], outputs=[chatbot, msg, user_session], api_name=INTERNAL_AI_GET_SERVER) jarvis.launch(max_file_size="1mb")