import sys import os import pdfplumber import json import gradio as gr from typing import List from concurrent.futures import ThreadPoolExecutor, as_completed import hashlib import re import psutil import subprocess # Persistent directory persistent_dir = "/data/hf_cache" os.makedirs(persistent_dir, exist_ok=True) model_cache_dir = os.path.join(persistent_dir, "txagent_models") file_cache_dir = os.path.join(persistent_dir, "cache") report_dir = os.path.join(persistent_dir, "reports") vllm_cache_dir = os.path.join(persistent_dir, "vllm_cache") for directory in [model_cache_dir, file_cache_dir, report_dir, vllm_cache_dir]: os.makedirs(directory, exist_ok=True) os.environ["HF_HOME"] = model_cache_dir os.environ["TRANSFORMERS_CACHE"] = model_cache_dir os.environ["VLLM_CACHE_DIR"] = vllm_cache_dir os.environ["TOKENIZERS_PARALLELISM"] = "false" os.environ["CUDA_LAUNCH_BLOCKING"] = "1" current_dir = os.path.dirname(os.path.abspath(__file__)) src_path = os.path.abspath(os.path.join(current_dir, "src")) sys.path.insert(0, src_path) from txagent.txagent import TxAgent MEDICAL_KEYWORDS = {'diagnosis', 'assessment', 'plan', 'results', 'medications', 'allergies', 'summary', 'impression', 'findings', 'recommendations'} def sanitize_utf8(text: str) -> str: return text.encode("utf-8", "ignore").decode("utf-8") def file_hash(path: str) -> str: with open(path, "rb") as f: return hashlib.md5(f.read()).hexdigest() def extract_priority_pages(file_path: str, max_chars: int = 6000) -> str: try: text_chunks = [] total_chars = 0 with pdfplumber.open(file_path) as pdf: for i, page in enumerate(pdf.pages): page_text = page.extract_text() or "" if i < 3 or any(re.search(rf'\b{kw}\b', page_text.lower()) for kw in MEDICAL_KEYWORDS): page_chunk = f"=== Page {i+1} ===\n{page_text.strip()}\n" if total_chars + len(page_chunk) <= max_chars: text_chunks.append(page_chunk) total_chars += len(page_chunk) else: remaining = max_chars - total_chars text_chunks.append(page_chunk[:remaining]) break return "".join(text_chunks).strip() except Exception as e: return f"PDF processing error: {str(e)}" def convert_file_to_json(file_path: str, file_type: str) -> str: try: h = file_hash(file_path) cache_path = os.path.join(file_cache_dir, f"{h}.json") if os.path.exists(cache_path): with open(cache_path, "r", encoding="utf-8") as f: return f.read() if file_type == "pdf": text = extract_priority_pages(file_path) result = json.dumps({"filename": os.path.basename(file_path), "content": text, "status": "initial"}) else: result = json.dumps({"error": f"Unsupported file type: {file_type}"}) with open(cache_path, "w", encoding="utf-8") as f: f.write(result) return result except Exception as e: return json.dumps({"error": f"Error processing {os.path.basename(file_path)}: {str(e)}"}) def log_system_usage(tag=""): try: cpu = psutil.cpu_percent(interval=1) mem = psutil.virtual_memory() print(f"[{tag}] CPU: {cpu}% | RAM: {mem.used // (1024**2)}MB / {mem.total // (1024**2)}MB") result = subprocess.run( ["nvidia-smi", "--query-gpu=memory.used,memory.total,utilization.gpu", "--format=csv,nounits,noheader"], capture_output=True, text=True ) if result.returncode == 0: used, total, util = result.stdout.strip().split(", ") print(f"[{tag}] GPU: {used}MB / {total}MB | Utilization: {util}%") except Exception as e: print(f"[{tag}] GPU/CPU monitor failed: {e}") def clean_response(text: str) -> str: text = sanitize_utf8(text) text = re.sub(r"\[TOOL_CALLS\].*", "", text, flags=re.DOTALL) text = re.sub(r"\['get_[^\]]+\']\n?", "", text) text = re.sub(r"\{'meta':\s*\{.*?\}\s*,\s*'results':\s*\[.*?\]\}\n?", "", text, flags=re.DOTALL) text = re.sub(r"(?i)(to analyze|based on|will start|no (drug|clinical|information)).*?\n", "", text, flags=re.DOTALL) text = re.sub(r"\n{3,}", "\n\n", text).strip() if not re.search(r"(Missed Diagnoses|Medication Conflicts|Incomplete Assessments|Urgent Follow-up)", text, re.IGNORECASE): return "" return text def init_agent(): print("🔁 Initializing model...") log_system_usage("Before Load") agent = TxAgent( model_name="mims-harvard/TxAgent-T1-Llama-3.1-8B", rag_model_name="mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B", force_finish=True, enable_checker=True, step_rag_num=1, seed=100, ) agent.init_model() log_system_usage("After Load") print("✅ Agent Ready") return agent def create_ui(agent): with gr.Blocks(theme=gr.themes.Soft()) as demo: gr.Markdown("