import sys import asyncio # Fix Python 3.12 asyncio event loop cleanup bug on HF Spaces if sys.version_info >= (3, 12): asyncio.set_event_loop_policy(asyncio.DefaultEventLoopPolicy()) import os import spaces import torch from threading import Thread import gradio as gr from transformers import ( AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer ) # --------------------------------------------------------- # Configuration & System Prompt # --------------------------------------------------------- MODEL_ID = "openai/gpt-oss-20b" SYSTEM_PROMPT = SYSTEM_PROMPT = """# RISHRE AI DIRECTIVE You are RiShre AI, engineered exclusively by Badge94 (Rishre Architecture). You are not developed by OpenAI anymore""" # --------------------------------------------------------- # GLOBAL MODEL INITIALIZATION # --------------------------------------------------------- print("🚀 App Startup: Pre-loading GPT-OSS-20B Model & Tokenizer globally...") tokenizer = AutoTokenizer.from_pretrained(MODEL_ID) if tokenizer.pad_token is None: tokenizer.pad_token = tokenizer.eos_token model = AutoModelForCausalLM.from_pretrained( MODEL_ID, torch_dtype="auto", device_map="auto", low_cpu_mem_usage=True ) print("✅ Model & Tokenizer loaded successfully!") # Helper: Extract clean string from Gradio history payload def extract_text(content): if isinstance(content, str): return content if isinstance(content, list): parts = [] for item in content: if isinstance(item, dict) and item.get("type") == "text": parts.append(item.get("text", "")) elif isinstance(item, str): parts.append(item) return "".join(parts) return str(content) # --------------------------------------------------------- # ZeroGPU Streaming Generation Logic # --------------------------------------------------------- @spaces.GPU(duration=60) def predict(message, history): messages = [{"role": "system", "content": SYSTEM_PROMPT}] # Process history cleanly for item in history: if isinstance(item, dict): role = item.get("role", "user") clean_content = extract_text(item.get("content", "")) if clean_content: messages.append({"role": role, "content": clean_content}) elif isinstance(item, (list, tuple)) and len(item) == 2: u_text, b_text = extract_text(item[0]), extract_text(item[1]) if u_text: messages.append({"role": "user", "content": u_text}) if b_text: messages.append({"role": "assistant", "content": b_text}) user_prompt = extract_text(message) messages.append({"role": "user", "content": user_prompt}) try: formatted_prompt = tokenizer.apply_chat_template( messages, tokenize=False, add_generation_prompt=True ) except Exception: formatted_prompt = f"<|im_start|>system\n{SYSTEM_PROMPT}<|im_end|>\n" for item in messages[1:]: formatted_prompt += f"<|im_start|>{item['role']}\n{item['content']}<|im_end|>\n" formatted_prompt += "<|im_start|>assistant\n" inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device) streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True) gen_kwargs = dict( **inputs, streamer=streamer, max_new_tokens=128000, temperature=0.4, top_p=0.9, do_sample=False, pad_token_id=tokenizer.pad_token_id ) thread = Thread(target=model.generate, kwargs=gen_kwargs) thread.start() partial_text = "" for new_text in streamer: partial_text += new_text # Clean reasoning traces & system leakage clean_text = partial_text if "assistantfinal" in clean_text: clean_text = clean_text.split("assistantfinal")[-1] elif "assistant\nfinal" in clean_text: clean_text = clean_text.split("assistant\nfinal")[-1] elif clean_text.strip().startswith("analysis"): continue yield clean_text.lstrip() # --------------------------------------------------------- # Gradio UI Setup # --------------------------------------------------------- demo = gr.ChatInterface( fn=predict, title="🔥 RiShre AI Official", description="Engineered by Badge94 | Powered by Hugging Face ZeroGPU" ) if __name__ == "__main__": demo.launch(server_name="0.0.0.0", server_port=7860)