Update app.py

app.py

@@ -1,705 +1,705 @@
-import os
-import json
-import time
-import asyncio
-import aiohttp
-from typing import Dict, List, Set, Optional
-from urllib.parse import quote, urljoin
-from datetime import datetime
-from pathlib import Path
-from datasets import Dataset, DatasetDict
-import huggingface_hub
-
-from fastapi import FastAPI, BackgroundTasks, HTTPException, status
-from fastapi.responses import JSONResponse
-from pydantic import BaseModel, Field
-import uvicorn
-
-# Path for storing caption data
-CAPTIONS_DIR = Path("captions_data")
-CAPTIONS_DIR.mkdir(exist_ok=True)
-
-# Hugging Face configuration
-HF_TOKEN = os.getenv("HF_TOKEN")
-HF_DATASET_ID = os.getenv("HF_DATASET_ID", "fred808/helium")
-
-if not HF_TOKEN:
-    raise ValueError("HF_TOKEN environment variable is required")
-
-def get_caption_file_path(course: str) -> Path:
-    """Get the path to the JSON file for storing course captions"""
-    safe_name = quote(course, safe='')
-    return CAPTIONS_DIR / f"{safe_name}_captions.json"
-
-def save_captions_to_file(course: str, captions: List[Dict]) -> None:
-    """Save captions to a JSON file"""
-    try:
-        file_path = get_caption_file_path(course)
-        with open(file_path, 'w', encoding='utf-8') as f:
-            json.dump(captions, f, indent=2, ensure_ascii=False)
-        print(f"✓ Saved {len(captions)} captions for {course}")
-    except Exception as e:
-        print(f"Error saving captions for {course}: {e}")
-
-def load_captions_from_file(course: str) -> List[Dict]:
-    """Load existing captions from JSON file"""
-    try:
-        file_path = get_caption_file_path(course)
-        if file_path.exists():
-            with open(file_path, 'r', encoding='utf-8') as f:
-                captions = json.load(f)
-                print(f"✓ Loaded {len(captions)} existing captions for {course}")
-                return captions
-    except Exception as e:
-        print(f"Error loading captions for {course}: {e}")
-    return []
-
-# Configuration
-SOURCE_SERVER = "https://fred808-vs2.hf.space"
-CAPTION_SERVERS = [
-    "https://favoredone-tv88mp.hf.space",
-  "https://favoredone-7p1dcf.hf.space",
-  "https://favoredone-k7b4mf.hf.space",
-  "https://favoredone-mzlxc7.hf.space",
-  "https://favoredone-aomfwa.hf.space",
-  "https://favoredone-7g6v04.hf.space",
-  "https://favoredone-dk1skh.hf.space",
-  "https://favoredone-z4yo0y.hf.space",
-  "https://favoredone-f6czeq.hf.space",
-  "https://favoredone-5fo8ga.hf.space",
-  "https://favoredone-zde8x6.hf.space",
-  "https://favoredone-r0biih.hf.space",
-  "https://favoredone-ljdzkf.hf.space",
-  "https://favoredone-irrpe5.hf.space",
-  "https://favoredone-bh9rwz.hf.space",
-  "https://favoredone-u8c4dt.hf.space",
-  "https://favoredone-futwyd.hf.space",
-  "https://favoredone-hg2sot.hf.space",
-  "https://favoredone-pvweug.hf.space",
-  "https://favoredone-z6azk2.hf.space",
-  "https://favoredone-4zid9w.hf.space",
-  "https://favoredone-be7a1r.hf.space",
-  "https://favoredone-ayazxa.hf.space",
-  "https://favoredone-6ckj4m.hf.space",
-  "https://favoredone-whn0xu.hf.space",
-  "https://favoredone-t49exm.hf.space",
-  "https://favoredone-cgrh0a.hf.space",
-  "https://favoredone-r1kb5g.hf.space"
-]
-MODEL_TYPE = "Florence-2-large"  # Explicitly request large model
-
-# FastAPI Models
-class CourseInfo(BaseModel):
-    course_folder: str
-
-class ImageInfo(BaseModel):
-    filename: str
-
-class CaptionRequest(BaseModel):
-    image_url: str
-    model_choice: str = MODEL_TYPE
-
-class CaptionResponse(BaseModel):
-    success: bool
-    caption: Optional[str] = None
-    error: Optional[str] = None
-
-class ServerStatus(BaseModel):
-    url: str
-    model: str
-    busy: bool
-    total_processed: int
-    total_time: float
-    fps: float
-
-class ProcessingStatus(BaseModel):
-    course: str
-    total_images: int
-    processed_images: int
-    progress_percent: float
-    status: str
-
-class StartProcessingRequest(BaseModel):
-    courses: Optional[List[str]] = None  # If None, process all courses
-    continuous: bool = True  # Default to continuous like original
-
-# FastAPI App
-app = FastAPI(
-    title="Caption Coordinator API",
-    description="Distributed caption processing coordinator",
-    version="1.0.0"
-)
-
-# Global state
-processed_images: Dict[str, Set[str]] = {}  # {course: set(image_names)}
-course_captions: Dict[str, List[Dict]] = {}  # {course: [{image, caption, metadata}]}
-failed_images: Dict[str, Set[str]] = {}  # {course: set(image_names)}
-servers = []
-is_processing = False
-current_processing_task = None
-auto_start_processing = True  # Set to False if you don't want auto-start
-
-class CaptionServer:
-    def __init__(self, url):
-        self.url = url
-        self.busy = False
-        self.model = "unknown"
-        self.total_processed = 0
-        self.total_time = 0
-
-    @property
-    def fps(self):
-        return self.total_processed / self.total_time if self.total_time > 0 else 0
-
-# Initialize servers
-def initialize_servers():
-    global servers
-    servers = [CaptionServer(url) for url in CAPTION_SERVERS]
-
-# API Routes
-@app.get("/")
-async def root():
-    return {
-        "message": "Caption Coordinator API", 
-        "status": "running",
-        "auto_processing": auto_start_processing,
-        "is_processing": is_processing
-    }
-
-@app.get("/health")
-async def health():
-    return {
-        "status": "healthy",
-        "servers_available": len([s for s in servers if not s.busy]),
-        "total_servers": len(servers),
-        "is_processing": is_processing,
-        "auto_processing": auto_start_processing
-    }
-
-@app.get("/courses")
-async def get_courses():
-    """Fetch available courses from source server"""
-    try:
-        async with aiohttp.ClientSession() as session:
-            async with session.get(f"{SOURCE_SERVER}/courses") as resp:
-                data = await resp.json()
-                if isinstance(data, dict) and 'courses' in data:
-                    return [c['course_folder'] for c in data['courses'] if isinstance(c, dict)]
-                return []
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Error fetching courses: {e}")
-
-@app.get("/courses/{course}/images")
-async def get_course_images(course: str):
-    """Fetch images list for a course"""
-    try:
-        course_frames = f"{course}_frames" if not course.endswith("_frames") else course
-        url = f"{SOURCE_SERVER}/images/{quote(course_frames)}"
-        async with aiohttp.ClientSession() as session:
-            async with session.get(url) as resp:
-                data = await resp.json()
-                if isinstance(data, dict) and 'images' in data:
-                    return data['images']
-                return []
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Error fetching images: {e}")
-
-@app.get("/servers/status")
-async def get_servers_status():
-    """Get status of all caption servers"""
-    server_statuses = []
-    for server in servers:
-        server_statuses.append(ServerStatus(
-            url=server.url,
-            model=server.model,
-            busy=server.busy,
-            total_processed=server.total_processed,
-            total_time=server.total_time,
-            fps=server.fps
-        ))
-    return server_statuses
-
-@app.get("/processing/status")
-async def get_processing_status():
-    """Get current processing status"""
-    status_info = {}
-    for course in processed_images:
-        total = len(processed_images[course])
-        processed = len(course_captions.get(course, []))
-        failed = len(failed_images.get(course, set()))
-        status_info[course] = {
-            "course": course,
-            "total_images": total,
-            "processed_images": processed,
-            "failed_images": failed,
-            "progress_percent": (processed / total * 100) if total > 0 else 0,
-            "status": "completed" if processed + failed >= total else "processing"
-        }
-    return status_info
-
-@app.post("/processing/start")
-async def start_processing(request: StartProcessingRequest = StartProcessingRequest()):
-    """Start caption processing"""
-    global is_processing, current_processing_task
-    
-    if is_processing:
-        raise HTTPException(status_code=400, detail="Processing is already running")
-    
-    is_processing = True
-    current_processing_task = asyncio.create_task(
-        processing_loop(request.courses, request.continuous)
-    )
-    
-    return {
-        "message": "Processing started", 
-        "continuous": request.continuous,
-        "specific_courses": request.courses
-    }
-
-@app.post("/processing/stop")
-async def stop_processing():
-    """Stop caption processing"""
-    global is_processing, current_processing_task
-    
-    if not is_processing:
-        raise HTTPException(status_code=400, detail="Processing is not running")
-    
-    is_processing = False
-    if current_processing_task:
-        current_processing_task.cancel()
-        try:
-            await current_processing_task
-        except asyncio.CancelledError:
-            pass
-        current_processing_task = None
-    
-    return {"message": "Processing stopped"}
-
-@app.get("/captions/{course}")
-async def get_captions(course: str):
-    """Get captions for a specific course"""
-    captions = load_captions_from_file(course)
-    return {
-        "course": course,
-        "total_captions": len(captions),
-        "captions": captions
-    }
-
-@app.delete("/captions/{course}")
-async def delete_captions(course: str):
-    """Delete captions for a specific course"""
-    try:
-        file_path = get_caption_file_path(course)
-        if file_path.exists():
-            file_path.unlink()
-            if course in processed_images:
-                del processed_images[course]
-            if course in course_captions:
-                del course_captions[course]
-            if course in failed_images:
-                del failed_images[course]
-            return {"message": f"Captions for {course} deleted"}
-        else:
-            raise HTTPException(status_code=404, detail=f"No captions found for {course}")
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Error deleting captions: {e}")
-
-# Core processing functions
-async def fetch_courses() -> List[str]:
-    """Fetch available courses from source server"""
-    async with aiohttp.ClientSession() as session:
-        async with session.get(f"{SOURCE_SERVER}/courses") as resp:
-            data = await resp.json()
-            if isinstance(data, dict) and 'courses' in data:
-                return [c['course_folder'] for c in data['courses'] if isinstance(c, dict)]
-            return []
-
-async def fetch_course_images(course: str) -> List[Dict]:
-    """Fetch images list for a course"""
-    course_frames = f"{course}_frames" if not course.endswith("_frames") else course
-    url = f"{SOURCE_SERVER}/images/{quote(course_frames)}"
-    async with aiohttp.ClientSession() as session:
-        async with session.get(url) as resp:
-            data = await resp.json()
-            if isinstance(data, dict) and 'images' in data:
-                return data['images']
-            return []
-
-async def get_caption(server: str, image_url: str) -> Dict:
-    """Get caption from a specific server"""
-    params = {
-        'image_url': image_url,
-        'model_choice': MODEL_TYPE
-    }
-    try:
-        async with aiohttp.ClientSession() as session:
-            async with session.get(server, params=params, timeout=30) as resp:
-                return await resp.json()
-    except Exception as e:
-        print(f"Error from {server}: {e}")
-        return None
-
-async def get_model_info():
-    """Get model information from caption servers"""
-    model_info = []
-    async with aiohttp.ClientSession() as session:
-        for server in CAPTION_SERVERS:
-            try:
-                health_url = server.rsplit('/analyze', 1)[0] + '/health'
-                async with session.get(health_url) as resp:
-                    info = await resp.json()
-                    model_info.append({
-                        'url': server,
-                        'model': info.get('model_choice', 'unknown')
-                    })
-            except Exception as e:
-                print(f"Couldn't get model info from {server}: {e}")
-    return model_info
-
-async def process_image(server: CaptionServer, course: str, image: Dict) -> Dict:
-    """Process single image through one caption server with better error handling"""
-    if server.busy:
-        return None
-        
-    server.busy = True
-    start_time = time.time()
-    
-    try:
-        # Structure URL correctly: /images/COURSE_NAME_frames/IMAGE.png
-        course_frames = f"{course}_frames" if not course.endswith("_frames") else course
-        image_url = urljoin(SOURCE_SERVER, f"/images/{quote(course_frames)}/{quote(image['filename'])}")
-        result = await get_caption(server.url, image_url)
-        
-        processing_time = time.time() - start_time
-        server.total_time += processing_time
-        
-        if result and result.get('success') and result.get('caption'):
-            server.total_processed += 1
-            metadata = {
-                "image": image['filename'],
-                "caption": result['caption'],
-                "server": server.url,
-                "processing_time": processing_time,
-                "timestamp": datetime.now().isoformat()
-            }
-            print(f"Server {server.url} processed {image['filename']} in {processing_time:.2f}s ({server.fps:.2f} fps)")
-            return metadata
-        else:
-            # Server responded but no caption (might be error or empty response)
-            error_msg = result.get('error', 'Unknown error') if result else 'No response'
-            print(f"Server {server.url} failed for {image['filename']}: {error_msg}")
-            return None
-            
-    except asyncio.TimeoutError:
-        print(f"Server {server.url} timeout for {image['filename']}")
-        return None
-    except Exception as e:
-        print(f"Error processing {image['filename']} on {server.url}: {e}")
-        return None
-        
-    finally:
-        server.busy = False
-
-async def upload_to_huggingface(course: str, metadata_list: List[Dict]):
-    """Upload course captions to Hugging Face dataset"""
-    try:
-        print(f"📤 Uploading {len(metadata_list)} captions for {course} to Hugging Face...")
-        
-        # Prepare data for Hugging Face dataset
-        dataset_data = {
-            "course": [],
-            "image_filename": [],
-            "caption": [],
-            "processing_server": [],
-            "processing_time": [],
-            "timestamp": []
-        }
-        
-        for metadata in metadata_list:
-            dataset_data["course"].append(course)
-            dataset_data["image_filename"].append(metadata["image"])
-            dataset_data["caption"].append(metadata["caption"])
-            dataset_data["processing_server"].append(metadata["server"])
-            dataset_data["processing_time"].append(metadata["processing_time"])
-            dataset_data["timestamp"].append(metadata["timestamp"])
-        
-        # Login to Hugging Face
-        huggingface_hub.login(token=HF_TOKEN)
-        
-        # Convert to JSON string
-        json_data = json.dumps(dataset_data, indent=2, ensure_ascii=False)
-        
-        # Create filename for the course
-        filename = f"{course.replace('/', '_').replace(' ', '_')}_captions.json"
-        
-        # Upload directly to hub as JSON file
-        huggingface_hub.upload_file(
-            path_or_fileobj=json_data.encode(),
-            path_in_repo=filename,
-            repo_id=HF_DATASET_ID,
-            commit_message=f"Add captions for course {course} - {len(metadata_list)} images"
-        )
-        
-        print(f"✅ Successfully uploaded {len(metadata_list)} captions for {course} to {HF_DATASET_ID}/{filename}")
-        return True
-        
-    except Exception as e:
-        print(f"❌ Error uploading to Hugging Face: {e}")
-        return False
-
-async def process_course(course: str, servers: List[CaptionServer]):
-    """Process all images in a course using available servers with proper retry logic"""
-    # Initialize course tracking
-    if course not in processed_images:
-        processed_images[course] = set()
-    if course not in course_captions:
-        course_captions[course] = load_captions_from_file(course)
-        # Update processed images set from loaded captions
-        for cap in course_captions[course]:
-            processed_images[course].add(cap['image'])
-    if course not in failed_images:
-        failed_images[course] = set()
-    
-    # Get list of images
-    images = await fetch_course_images(course)
-    if not images:
-        print(f"No images found for course {course}")
-        return
-    
-    print(f"\nProcessing {len(images)} images for course {course}")
-    
-    # Track images that need processing with retry count (5 retries)
-    pending_images = {}
-    for img in images:
-        filename = img['filename']
-        if filename not in processed_images[course] and filename not in failed_images[course]:
-            pending_images[filename] = {'image': img, 'retries': 0, 'max_retries': 5}
-    
-    if not pending_images:
-        print(f"All images already processed or failed for course {course}")
-        print(f"- Processed: {len(processed_images[course])}, Failed: {len(failed_images[course])}")
-        
-        # If course is completed, upload to Hugging Face
-        if len(processed_images[course]) + len(failed_images[course]) >= len(images):
-            if course_captions[course]:
-                print(f"📤 Course {course} completed, uploading to Hugging Face...")
-                await upload_to_huggingface(course, course_captions[course])
-        return
-    
-    print(f"Images to process: {len(pending_images)} (already processed: {len(processed_images[course])}, failed: {len(failed_images[course])})")
-    
-    batch_size = len([s for s in servers if not s.busy])
-    processed_in_this_run = 0
-    
-    while pending_images and is_processing:
-        # Create tasks for each available server
-        tasks = []
-        assigned_images = []
-        
-        for server in servers:
-            if not server.busy and pending_images:
-                # Get the next pending image
-                filename, img_data = next(iter(pending_images.items()))
-                img = img_data['image']
-                
-                # Assign this image to the server
-                tasks.append(process_image(server, course, img))
-                assigned_images.append((filename, img, img_data['retries']))
-                # Remove from pending temporarily while it's being processed
-                del pending_images[filename]
-                
-        if not tasks:
-            # If no servers available, wait a bit
-            await asyncio.sleep(0.1)
-            continue
-            
-        # Process images in parallel across servers
-        results = await asyncio.gather(*tasks)
-        
-        # Handle results and retry logic
-        has_new_results = False
-        for (filename, img, current_retries), result in zip(assigned_images, results):
-            if result:
-                # Success - image was processed
-                processed_images[course].add(filename)
-                course_captions[course].append(result)
-                has_new_results = True
-                processed_in_this_run += 1
-                print(f"✓ Successfully processed {filename}")
-            else:
-                # Failure - check if we should retry
-                if current_retries < 5:  # max_retries
-                    # Put back in pending for retry with incremented retry count
-                    pending_images[filename] = {
-                        'image': img, 
-                        'retries': current_retries + 1, 
-                        'max_retries': 5
-                    }
-                    print(f"↻ Retry {current_retries + 1}/5 for {filename}")
-                else:
-                    # Max retries exceeded, mark as failed
-                    failed_images[course].add(filename)
-                    print(f"✗ Failed to process {filename} after 5 retries")
-        
-        # Save progress after each batch with new results
-        if has_new_results:
-            save_captions_to_file(course, course_captions[course])
-        
-        # Show progress
-        total = len(images)
-        done = len(processed_images[course])
-        failed_count = len(failed_images[course])
-        pending_count = len(pending_images)
-        progress_percent = (done / total * 100) if total > 0 else 0
-        
-        print(f"\rProgress: {done}/{total} ({progress_percent:.1f}%) - {pending_count} pending, {failed_count} failed, {processed_in_this_run} new", end="", flush=True)
-        
-        # Small delay to prevent overwhelming the servers
-        await asyncio.sleep(0.5)
-    
-    # Final status for this course
-    total = len(images)
-    done = len(processed_images[course])
-    failed_count = len(failed_images[course])
-    
-    if done + failed_count >= total:
-        if failed_count > 0:
-            print(f"\n✓ Course {course} completed with {failed_count} failed images")
-        else:
-            print(f"\n✓ Course {course} fully completed")
-        
-        # Upload to Hugging Face when course is completed
-        if course_captions[course]:
-            print(f"📤 Uploading {len(course_captions[course])} captions to Hugging Face...")
-            success = await upload_to_huggingface(course, course_captions[course])
-            if success:
-                print(f"✅ Successfully uploaded {course} to Hugging Face")
-            else:
-                print(f"❌ Failed to upload {course} to Hugging Face")
-    else:
-        print(f"\n→ Course {course} partially completed: {done}/{total} processed, {failed_count} failed")
-
-async def processing_loop(specific_courses: Optional[List[str]] = None, continuous: bool = True):
-    """Main processing loop with proper error handling"""
-    global is_processing
-    
-    # Get model information and verify Florence-2-large availability
-    model_info = await get_model_info()
-    print("\nCaption Servers:")
-    available_servers = []
-    for info, server in zip(model_info, servers):
-        server.model = info['model']
-        if MODEL_TYPE in info.get('model', ''):
-            available_servers.append(server)
-            print(f"✓ {server.url} confirmed {MODEL_TYPE}")
-        else:
-            print(f"✗ {server.url} using {server.model} - skipping (requires {MODEL_TYPE})")
-    
-    if not available_servers:
-        print(f"\nError: No servers with {MODEL_TYPE} available!")
-        is_processing = False
-        return
-    
-    # Update servers list to only use those with large model
-    processing_servers = available_servers
-    print(f"\nUsing {len(processing_servers)} servers with {MODEL_TYPE}")
-    
-    # Check for existing caption files and report
-    existing_captions = list(CAPTIONS_DIR.glob("*_captions.json"))
-    if existing_captions:
-        print("\nFound existing caption files:")
-        for cap_file in existing_captions:
-            course = cap_file.stem.replace("_captions", "")
-            try:
-                with open(cap_file, 'r', encoding='utf-8') as f:
-                    captions = json.load(f)
-                    print(f"- {course}: {len(captions)} captions")
-            except Exception as e:
-                print(f"- Error reading {cap_file.name}: {e}")
-        print()
-    
-    start_time = time.time()
-    iteration = 0
-    
-    while is_processing:
-        try:
-            iteration += 1
-            print(f"\n{'='*50}")
-            print(f"Processing Iteration {iteration}")
-            print(f"{'='*50}")
-            
-            # Get available courses
-            if specific_courses:
-                courses = specific_courses
-                print(f"Processing specific courses: {courses}")
-            else:
-                courses = await fetch_courses()
-                print(f"Found {len(courses)} courses")
-                
-            if not courses:
-                print("No courses found, waiting...")
-                if not continuous:
-                    break
-                await asyncio.sleep(10)
-                continue
-            
-            # Process each course with all available servers
-            for course in courses:
-                if not is_processing:
-                    break
-                    
-                print(f"\n--- Processing course: {course} ---")
-                await process_course(course, processing_servers)
-                
-            # Show server stats
-            print("\nServer Stats:")
-            total_processed = sum(s.total_processed for s in processing_servers)
-            elapsed = time.time() - start_time
-            if elapsed > 0:
-                print(f"Total images processed: {total_processed}")
-                print(f"Overall speed: {total_processed/elapsed:.2f} fps")
-                for s in processing_servers:
-                    print(f"- {s.url}: {s.total_processed} images, {s.fps:.2f} fps")
-            print()
-            
-            if not continuous:
-                print("One-time processing completed")
-                break
-                
-            # Wait before next check
-            print("Waiting for new courses...")
-            await asyncio.sleep(5)
-            
-        except asyncio.CancelledError:
-            print("Processing cancelled")
-            break
-        except Exception as e:
-            print(f"Error in processing loop: {str(e)}")
-            import traceback
-            traceback.print_exc()
-            await asyncio.sleep(10)
-    
-    is_processing = False
-    print("Processing loop stopped")
-
-# Startup event
-@app.on_event("startup")
-async def startup_event():
-    """Initialize servers and start processing on startup"""
-    initialize_servers()
-    print("Caption Coordinator API started")
-    print(f"Source server: {SOURCE_SERVER}")
-    print(f"Caption servers: {len(CAPTION_SERVERS)}")
-    print(f"Hugging Face dataset: {HF_DATASET_ID}")
-    print(f"HF Token: {'✅ Set' if HF_TOKEN else '❌ Missing'}")
-    
-    # Start processing automatically (like original main())
-    if auto_start_processing:
-        print("Auto-starting processing loop...")
-        global is_processing, current_processing_task
-        is_processing = True
-        current_processing_task = asyncio.create_task(processing_loop())
-
-
-if __name__ == "__main__":
+import os
+import json
+import time
+import asyncio
+import aiohttp
+from typing import Dict, List, Set, Optional
+from urllib.parse import quote, urljoin
+from datetime import datetime
+from pathlib import Path
+from datasets import Dataset, DatasetDict
+import huggingface_hub
+
+from fastapi import FastAPI, BackgroundTasks, HTTPException, status
+from fastapi.responses import JSONResponse
+from pydantic import BaseModel, Field
+import uvicorn
+
+# Path for storing caption data
+CAPTIONS_DIR = Path("captions_data")
+CAPTIONS_DIR.mkdir(exist_ok=True)
+
+# Hugging Face configuration
+HF_TOKEN = os.getenv("HF_TOKEN")
+HF_DATASET_ID = os.getenv("HF_DATASET_ID", "fred808/helium")
+
+if not HF_TOKEN:
+    raise ValueError("HF_TOKEN environment variable is required")
+
+def get_caption_file_path(course: str) -> Path:
+    """Get the path to the JSON file for storing course captions"""
+    safe_name = quote(course, safe='')
+    return CAPTIONS_DIR / f"{safe_name}_captions.json"
+
+def save_captions_to_file(course: str, captions: List[Dict]) -> None:
+    """Save captions to a JSON file"""
+    try:
+        file_path = get_caption_file_path(course)
+        with open(file_path, 'w', encoding='utf-8') as f:
+            json.dump(captions, f, indent=2, ensure_ascii=False)
+        print(f"✓ Saved {len(captions)} captions for {course}")
+    except Exception as e:
+        print(f"Error saving captions for {course}: {e}")
+
+def load_captions_from_file(course: str) -> List[Dict]:
+    """Load existing captions from JSON file"""
+    try:
+        file_path = get_caption_file_path(course)
+        if file_path.exists():
+            with open(file_path, 'r', encoding='utf-8') as f:
+                captions = json.load(f)
+                print(f"✓ Loaded {len(captions)} existing captions for {course}")
+                return captions
+    except Exception as e:
+        print(f"Error loading captions for {course}: {e}")
+    return []
+
+# Configuration
+SOURCE_SERVER = "https://favoredone-imm.hf.space"
+CAPTION_SERVERS = [
+    "https://favoredone-tv88mp.hf.space",
+  "https://favoredone-7p1dcf.hf.space",
+  "https://favoredone-k7b4mf.hf.space",
+  "https://favoredone-mzlxc7.hf.space",
+  "https://favoredone-aomfwa.hf.space",
+  "https://favoredone-7g6v04.hf.space",
+  "https://favoredone-dk1skh.hf.space",
+  "https://favoredone-z4yo0y.hf.space",
+  "https://favoredone-f6czeq.hf.space",
+  "https://favoredone-5fo8ga.hf.space",
+  "https://favoredone-zde8x6.hf.space",
+  "https://favoredone-r0biih.hf.space",
+  "https://favoredone-ljdzkf.hf.space",
+  "https://favoredone-irrpe5.hf.space",
+  "https://favoredone-bh9rwz.hf.space",
+  "https://favoredone-u8c4dt.hf.space",
+  "https://favoredone-futwyd.hf.space",
+  "https://favoredone-hg2sot.hf.space",
+  "https://favoredone-pvweug.hf.space",
+  "https://favoredone-z6azk2.hf.space",
+  "https://favoredone-4zid9w.hf.space",
+  "https://favoredone-be7a1r.hf.space",
+  "https://favoredone-ayazxa.hf.space",
+  "https://favoredone-6ckj4m.hf.space",
+  "https://favoredone-whn0xu.hf.space",
+  "https://favoredone-t49exm.hf.space",
+  "https://favoredone-cgrh0a.hf.space",
+  "https://favoredone-r1kb5g.hf.space"
+]
+MODEL_TYPE = "Florence-2-large"  # Explicitly request large model
+
+# FastAPI Models
+class CourseInfo(BaseModel):
+    course_folder: str
+
+class ImageInfo(BaseModel):
+    filename: str
+
+class CaptionRequest(BaseModel):
+    image_url: str
+    model_choice: str = MODEL_TYPE
+
+class CaptionResponse(BaseModel):
+    success: bool
+    caption: Optional[str] = None
+    error: Optional[str] = None
+
+class ServerStatus(BaseModel):
+    url: str
+    model: str
+    busy: bool
+    total_processed: int
+    total_time: float
+    fps: float
+
+class ProcessingStatus(BaseModel):
+    course: str
+    total_images: int
+    processed_images: int
+    progress_percent: float
+    status: str
+
+class StartProcessingRequest(BaseModel):
+    courses: Optional[List[str]] = None  # If None, process all courses
+    continuous: bool = True  # Default to continuous like original
+
+# FastAPI App
+app = FastAPI(
+    title="Caption Coordinator API",
+    description="Distributed caption processing coordinator",
+    version="1.0.0"
+)
+
+# Global state
+processed_images: Dict[str, Set[str]] = {}  # {course: set(image_names)}
+course_captions: Dict[str, List[Dict]] = {}  # {course: [{image, caption, metadata}]}
+failed_images: Dict[str, Set[str]] = {}  # {course: set(image_names)}
+servers = []
+is_processing = False
+current_processing_task = None
+auto_start_processing = True  # Set to False if you don't want auto-start
+
+class CaptionServer:
+    def __init__(self, url):
+        self.url = url
+        self.busy = False
+        self.model = "unknown"
+        self.total_processed = 0
+        self.total_time = 0
+
+    @property
+    def fps(self):
+        return self.total_processed / self.total_time if self.total_time > 0 else 0
+
+# Initialize servers
+def initialize_servers():
+    global servers
+    servers = [CaptionServer(url) for url in CAPTION_SERVERS]
+
+# API Routes
+@app.get("/")
+async def root():
+    return {
+        "message": "Caption Coordinator API", 
+        "status": "running",
+        "auto_processing": auto_start_processing,
+        "is_processing": is_processing
+    }
+
+@app.get("/health")
+async def health():
+    return {
+        "status": "healthy",
+        "servers_available": len([s for s in servers if not s.busy]),
+        "total_servers": len(servers),
+        "is_processing": is_processing,
+        "auto_processing": auto_start_processing
+    }
+
+@app.get("/courses")
+async def get_courses():
+    """Fetch available courses from source server"""
+    try:
+        async with aiohttp.ClientSession() as session:
+            async with session.get(f"{SOURCE_SERVER}/courses") as resp:
+                data = await resp.json()
+                if isinstance(data, dict) and 'courses' in data:
+                    return [c['course_folder'] for c in data['courses'] if isinstance(c, dict)]
+                return []
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Error fetching courses: {e}")
+
+@app.get("/courses/{course}/images")
+async def get_course_images(course: str):
+    """Fetch images list for a course"""
+    try:
+        course_frames = f"{course}_frames" if not course.endswith("_frames") else course
+        url = f"{SOURCE_SERVER}/images/{quote(course_frames)}"
+        async with aiohttp.ClientSession() as session:
+            async with session.get(url) as resp:
+                data = await resp.json()
+                if isinstance(data, dict) and 'images' in data:
+                    return data['images']
+                return []
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Error fetching images: {e}")
+
+@app.get("/servers/status")
+async def get_servers_status():
+    """Get status of all caption servers"""
+    server_statuses = []
+    for server in servers:
+        server_statuses.append(ServerStatus(
+            url=server.url,
+            model=server.model,
+            busy=server.busy,
+            total_processed=server.total_processed,
+            total_time=server.total_time,
+            fps=server.fps
+        ))
+    return server_statuses
+
+@app.get("/processing/status")
+async def get_processing_status():
+    """Get current processing status"""
+    status_info = {}
+    for course in processed_images:
+        total = len(processed_images[course])
+        processed = len(course_captions.get(course, []))
+        failed = len(failed_images.get(course, set()))
+        status_info[course] = {
+            "course": course,
+            "total_images": total,
+            "processed_images": processed,
+            "failed_images": failed,
+            "progress_percent": (processed / total * 100) if total > 0 else 0,
+            "status": "completed" if processed + failed >= total else "processing"
+        }
+    return status_info
+
+@app.post("/processing/start")
+async def start_processing(request: StartProcessingRequest = StartProcessingRequest()):
+    """Start caption processing"""
+    global is_processing, current_processing_task
+    
+    if is_processing:
+        raise HTTPException(status_code=400, detail="Processing is already running")
+    
+    is_processing = True
+    current_processing_task = asyncio.create_task(
+        processing_loop(request.courses, request.continuous)
+    )
+    
+    return {
+        "message": "Processing started", 
+        "continuous": request.continuous,
+        "specific_courses": request.courses
+    }
+
+@app.post("/processing/stop")
+async def stop_processing():
+    """Stop caption processing"""
+    global is_processing, current_processing_task
+    
+    if not is_processing:
+        raise HTTPException(status_code=400, detail="Processing is not running")
+    
+    is_processing = False
+    if current_processing_task:
+        current_processing_task.cancel()
+        try:
+            await current_processing_task
+        except asyncio.CancelledError:
+            pass
+        current_processing_task = None
+    
+    return {"message": "Processing stopped"}
+
+@app.get("/captions/{course}")
+async def get_captions(course: str):
+    """Get captions for a specific course"""
+    captions = load_captions_from_file(course)
+    return {
+        "course": course,
+        "total_captions": len(captions),
+        "captions": captions
+    }
+
+@app.delete("/captions/{course}")
+async def delete_captions(course: str):
+    """Delete captions for a specific course"""
+    try:
+        file_path = get_caption_file_path(course)
+        if file_path.exists():
+            file_path.unlink()
+            if course in processed_images:
+                del processed_images[course]
+            if course in course_captions:
+                del course_captions[course]
+            if course in failed_images:
+                del failed_images[course]
+            return {"message": f"Captions for {course} deleted"}
+        else:
+            raise HTTPException(status_code=404, detail=f"No captions found for {course}")
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Error deleting captions: {e}")
+
+# Core processing functions
+async def fetch_courses() -> List[str]:
+    """Fetch available courses from source server"""
+    async with aiohttp.ClientSession() as session:
+        async with session.get(f"{SOURCE_SERVER}/courses") as resp:
+            data = await resp.json()
+            if isinstance(data, dict) and 'courses' in data:
+                return [c['course_folder'] for c in data['courses'] if isinstance(c, dict)]
+            return []
+
+async def fetch_course_images(course: str) -> List[Dict]:
+    """Fetch images list for a course"""
+    course_frames = f"{course}_frames" if not course.endswith("_frames") else course
+    url = f"{SOURCE_SERVER}/images/{quote(course_frames)}"
+    async with aiohttp.ClientSession() as session:
+        async with session.get(url) as resp:
+            data = await resp.json()
+            if isinstance(data, dict) and 'images' in data:
+                return data['images']
+            return []
+
+async def get_caption(server: str, image_url: str) -> Dict:
+    """Get caption from a specific server"""
+    params = {
+        'image_url': image_url,
+        'model_choice': MODEL_TYPE
+    }
+    try:
+        async with aiohttp.ClientSession() as session:
+            async with session.get(server, params=params, timeout=30) as resp:
+                return await resp.json()
+    except Exception as e:
+        print(f"Error from {server}: {e}")
+        return None
+
+async def get_model_info():
+    """Get model information from caption servers"""
+    model_info = []
+    async with aiohttp.ClientSession() as session:
+        for server in CAPTION_SERVERS:
+            try:
+                health_url = server.rsplit('/analyze', 1)[0] + '/health'
+                async with session.get(health_url) as resp:
+                    info = await resp.json()
+                    model_info.append({
+                        'url': server,
+                        'model': info.get('model_choice', 'unknown')
+                    })
+            except Exception as e:
+                print(f"Couldn't get model info from {server}: {e}")
+    return model_info
+
+async def process_image(server: CaptionServer, course: str, image: Dict) -> Dict:
+    """Process single image through one caption server with better error handling"""
+    if server.busy:
+        return None
+        
+    server.busy = True
+    start_time = time.time()
+    
+    try:
+        # Structure URL correctly: /images/COURSE_NAME_frames/IMAGE.png
+        course_frames = f"{course}_frames" if not course.endswith("_frames") else course
+        image_url = urljoin(SOURCE_SERVER, f"/images/{quote(course_frames)}/{quote(image['filename'])}")
+        result = await get_caption(server.url, image_url)
+        
+        processing_time = time.time() - start_time
+        server.total_time += processing_time
+        
+        if result and result.get('success') and result.get('caption'):
+            server.total_processed += 1
+            metadata = {
+                "image": image['filename'],
+                "caption": result['caption'],
+                "server": server.url,
+                "processing_time": processing_time,
+                "timestamp": datetime.now().isoformat()
+            }
+            print(f"Server {server.url} processed {image['filename']} in {processing_time:.2f}s ({server.fps:.2f} fps)")
+            return metadata
+        else:
+            # Server responded but no caption (might be error or empty response)
+            error_msg = result.get('error', 'Unknown error') if result else 'No response'
+            print(f"Server {server.url} failed for {image['filename']}: {error_msg}")
+            return None
+            
+    except asyncio.TimeoutError:
+        print(f"Server {server.url} timeout for {image['filename']}")
+        return None
+    except Exception as e:
+        print(f"Error processing {image['filename']} on {server.url}: {e}")
+        return None
+        
+    finally:
+        server.busy = False
+
+async def upload_to_huggingface(course: str, metadata_list: List[Dict]):
+    """Upload course captions to Hugging Face dataset"""
+    try:
+        print(f"📤 Uploading {len(metadata_list)} captions for {course} to Hugging Face...")
+        
+        # Prepare data for Hugging Face dataset
+        dataset_data = {
+            "course": [],
+            "image_filename": [],
+            "caption": [],
+            "processing_server": [],
+            "processing_time": [],
+            "timestamp": []
+        }
+        
+        for metadata in metadata_list:
+            dataset_data["course"].append(course)
+            dataset_data["image_filename"].append(metadata["image"])
+            dataset_data["caption"].append(metadata["caption"])
+            dataset_data["processing_server"].append(metadata["server"])
+            dataset_data["processing_time"].append(metadata["processing_time"])
+            dataset_data["timestamp"].append(metadata["timestamp"])
+        
+        # Login to Hugging Face
+        huggingface_hub.login(token=HF_TOKEN)
+        
+        # Convert to JSON string
+        json_data = json.dumps(dataset_data, indent=2, ensure_ascii=False)
+        
+        # Create filename for the course
+        filename = f"{course.replace('/', '_').replace(' ', '_')}_captions.json"
+        
+        # Upload directly to hub as JSON file
+        huggingface_hub.upload_file(
+            path_or_fileobj=json_data.encode(),
+            path_in_repo=filename,
+            repo_id=HF_DATASET_ID,
+            commit_message=f"Add captions for course {course} - {len(metadata_list)} images"
+        )
+        
+        print(f"✅ Successfully uploaded {len(metadata_list)} captions for {course} to {HF_DATASET_ID}/{filename}")
+        return True
+        
+    except Exception as e:
+        print(f"❌ Error uploading to Hugging Face: {e}")
+        return False
+
+async def process_course(course: str, servers: List[CaptionServer]):
+    """Process all images in a course using available servers with proper retry logic"""
+    # Initialize course tracking
+    if course not in processed_images:
+        processed_images[course] = set()
+    if course not in course_captions:
+        course_captions[course] = load_captions_from_file(course)
+        # Update processed images set from loaded captions
+        for cap in course_captions[course]:
+            processed_images[course].add(cap['image'])
+    if course not in failed_images:
+        failed_images[course] = set()
+    
+    # Get list of images
+    images = await fetch_course_images(course)
+    if not images:
+        print(f"No images found for course {course}")
+        return
+    
+    print(f"\nProcessing {len(images)} images for course {course}")
+    
+    # Track images that need processing with retry count (5 retries)
+    pending_images = {}
+    for img in images:
+        filename = img['filename']
+        if filename not in processed_images[course] and filename not in failed_images[course]:
+            pending_images[filename] = {'image': img, 'retries': 0, 'max_retries': 5}
+    
+    if not pending_images:
+        print(f"All images already processed or failed for course {course}")
+        print(f"- Processed: {len(processed_images[course])}, Failed: {len(failed_images[course])}")
+        
+        # If course is completed, upload to Hugging Face
+        if len(processed_images[course]) + len(failed_images[course]) >= len(images):
+            if course_captions[course]:
+                print(f"📤 Course {course} completed, uploading to Hugging Face...")
+                await upload_to_huggingface(course, course_captions[course])
+        return
+    
+    print(f"Images to process: {len(pending_images)} (already processed: {len(processed_images[course])}, failed: {len(failed_images[course])})")
+    
+    batch_size = len([s for s in servers if not s.busy])
+    processed_in_this_run = 0
+    
+    while pending_images and is_processing:
+        # Create tasks for each available server
+        tasks = []
+        assigned_images = []
+        
+        for server in servers:
+            if not server.busy and pending_images:
+                # Get the next pending image
+                filename, img_data = next(iter(pending_images.items()))
+                img = img_data['image']
+                
+                # Assign this image to the server
+                tasks.append(process_image(server, course, img))
+                assigned_images.append((filename, img, img_data['retries']))
+                # Remove from pending temporarily while it's being processed
+                del pending_images[filename]
+                
+        if not tasks:
+            # If no servers available, wait a bit
+            await asyncio.sleep(0.1)
+            continue
+            
+        # Process images in parallel across servers
+        results = await asyncio.gather(*tasks)
+        
+        # Handle results and retry logic
+        has_new_results = False
+        for (filename, img, current_retries), result in zip(assigned_images, results):
+            if result:
+                # Success - image was processed
+                processed_images[course].add(filename)
+                course_captions[course].append(result)
+                has_new_results = True
+                processed_in_this_run += 1
+                print(f"✓ Successfully processed {filename}")
+            else:
+                # Failure - check if we should retry
+                if current_retries < 5:  # max_retries
+                    # Put back in pending for retry with incremented retry count
+                    pending_images[filename] = {
+                        'image': img, 
+                        'retries': current_retries + 1, 
+                        'max_retries': 5
+                    }
+                    print(f"↻ Retry {current_retries + 1}/5 for {filename}")
+                else:
+                    # Max retries exceeded, mark as failed
+                    failed_images[course].add(filename)
+                    print(f"✗ Failed to process {filename} after 5 retries")
+        
+        # Save progress after each batch with new results
+        if has_new_results:
+            save_captions_to_file(course, course_captions[course])
+        
+        # Show progress
+        total = len(images)
+        done = len(processed_images[course])
+        failed_count = len(failed_images[course])
+        pending_count = len(pending_images)
+        progress_percent = (done / total * 100) if total > 0 else 0
+        
+        print(f"\rProgress: {done}/{total} ({progress_percent:.1f}%) - {pending_count} pending, {failed_count} failed, {processed_in_this_run} new", end="", flush=True)
+        
+        # Small delay to prevent overwhelming the servers
+        await asyncio.sleep(0.5)
+    
+    # Final status for this course
+    total = len(images)
+    done = len(processed_images[course])
+    failed_count = len(failed_images[course])
+    
+    if done + failed_count >= total:
+        if failed_count > 0:
+            print(f"\n✓ Course {course} completed with {failed_count} failed images")
+        else:
+            print(f"\n✓ Course {course} fully completed")
+        
+        # Upload to Hugging Face when course is completed
+        if course_captions[course]:
+            print(f"📤 Uploading {len(course_captions[course])} captions to Hugging Face...")
+            success = await upload_to_huggingface(course, course_captions[course])
+            if success:
+                print(f"✅ Successfully uploaded {course} to Hugging Face")
+            else:
+                print(f"❌ Failed to upload {course} to Hugging Face")
+    else:
+        print(f"\n→ Course {course} partially completed: {done}/{total} processed, {failed_count} failed")
+
+async def processing_loop(specific_courses: Optional[List[str]] = None, continuous: bool = True):
+    """Main processing loop with proper error handling"""
+    global is_processing
+    
+    # Get model information and verify Florence-2-large availability
+    model_info = await get_model_info()
+    print("\nCaption Servers:")
+    available_servers = []
+    for info, server in zip(model_info, servers):
+        server.model = info['model']
+        if MODEL_TYPE in info.get('model', ''):
+            available_servers.append(server)
+            print(f"✓ {server.url} confirmed {MODEL_TYPE}")
+        else:
+            print(f"✗ {server.url} using {server.model} - skipping (requires {MODEL_TYPE})")
+    
+    if not available_servers:
+        print(f"\nError: No servers with {MODEL_TYPE} available!")
+        is_processing = False
+        return
+    
+    # Update servers list to only use those with large model
+    processing_servers = available_servers
+    print(f"\nUsing {len(processing_servers)} servers with {MODEL_TYPE}")
+    
+    # Check for existing caption files and report
+    existing_captions = list(CAPTIONS_DIR.glob("*_captions.json"))
+    if existing_captions:
+        print("\nFound existing caption files:")
+        for cap_file in existing_captions:
+            course = cap_file.stem.replace("_captions", "")
+            try:
+                with open(cap_file, 'r', encoding='utf-8') as f:
+                    captions = json.load(f)
+                    print(f"- {course}: {len(captions)} captions")
+            except Exception as e:
+                print(f"- Error reading {cap_file.name}: {e}")
+        print()
+    
+    start_time = time.time()
+    iteration = 0
+    
+    while is_processing:
+        try:
+            iteration += 1
+            print(f"\n{'='*50}")
+            print(f"Processing Iteration {iteration}")
+            print(f"{'='*50}")
+            
+            # Get available courses
+            if specific_courses:
+                courses = specific_courses
+                print(f"Processing specific courses: {courses}")
+            else:
+                courses = await fetch_courses()
+                print(f"Found {len(courses)} courses")
+                
+            if not courses:
+                print("No courses found, waiting...")
+                if not continuous:
+                    break
+                await asyncio.sleep(10)
+                continue
+            
+            # Process each course with all available servers
+            for course in courses:
+                if not is_processing:
+                    break
+                    
+                print(f"\n--- Processing course: {course} ---")
+                await process_course(course, processing_servers)
+                
+            # Show server stats
+            print("\nServer Stats:")
+            total_processed = sum(s.total_processed for s in processing_servers)
+            elapsed = time.time() - start_time
+            if elapsed > 0:
+                print(f"Total images processed: {total_processed}")
+                print(f"Overall speed: {total_processed/elapsed:.2f} fps")
+                for s in processing_servers:
+                    print(f"- {s.url}: {s.total_processed} images, {s.fps:.2f} fps")
+            print()
+            
+            if not continuous:
+                print("One-time processing completed")
+                break
+                
+            # Wait before next check
+            print("Waiting for new courses...")
+            await asyncio.sleep(5)
+            
+        except asyncio.CancelledError:
+            print("Processing cancelled")
+            break
+        except Exception as e:
+            print(f"Error in processing loop: {str(e)}")
+            import traceback
+            traceback.print_exc()
+            await asyncio.sleep(10)
+    
+    is_processing = False
+    print("Processing loop stopped")
+
+# Startup event
+@app.on_event("startup")
+async def startup_event():
+    """Initialize servers and start processing on startup"""
+    initialize_servers()
+    print("Caption Coordinator API started")
+    print(f"Source server: {SOURCE_SERVER}")
+    print(f"Caption servers: {len(CAPTION_SERVERS)}")
+    print(f"Hugging Face dataset: {HF_DATASET_ID}")
+    print(f"HF Token: {'✅ Set' if HF_TOKEN else '❌ Missing'}")
+    
+    # Start processing automatically (like original main())
+    if auto_start_processing:
+        print("Auto-starting processing loop...")
+        global is_processing, current_processing_task
+        is_processing = True
+        current_processing_task = asyncio.create_task(processing_loop())
+
+
+if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=8000, reload=True)