Fix HF Space: update SDK version, add error handling, improve README

README.md

@@ -4,7 +4,7 @@ emoji: 🎭
 colorFrom: purple
 colorTo: pink
 sdk: gradio
-sdk_version: 4.44.1
+sdk_version: 3.35.0
 app_file: app.py
 pinned: false
 license: mit
@@ -17,72 +17,30 @@ tags:
 - chatbot
 - transformers
 - qwen
-models:
-- Qwen/Qwen2.5-1.5B
 ---
 
 # Roleplay Chat Box 🎭
 
-An intelligent conversational AI system featuring multiple character personalities powered by LoRA (Low-Rank Adaptation) fine-tuned language models. Each character has been trained with specific datasets to maintain unique personalities, speaking patterns, and areas of expertise.
+An intelligent conversational AI system featuring multiple character personalities powered by LoRA (Low-Rank Adaptation) fine-tuned language models.
 
-## Features
+## ⚠️ Important Notes for HuggingFace Spaces
 
-- **Multi-Character Support**: Switch between different character personalities using LoRA adapters
-- **Voice Synthesis**: Text-to-speech with character-specific voices using Microsoft's VibeVoice
-- **Modern Interface**: ChatGPT-style web interface with character selection sidebar
-- **Real-time Chat**: WebSocket-based real-time communication
-- **Audio Toggle**: Optional voice output for responses
+This app requires significant resources:
+- **Model**: Qwen/Qwen3-0.6B (will be downloaded from HuggingFace Hub)
+- **Memory**: ~4GB RAM minimum for model loading
+- **Storage**: LoRA adapters are included in the repo
 
-## Architecture
-
-```
-┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
-│   Frontend      │    │   Backend API   │    │   ML Models     │
-│   (React/HTML)  │◄──►│   (FastAPI)     │◄──►│   Base + LoRA   │
-│                 │    │                 │    │   + VibeVoice   │
-│ - Chat UI       │    │ - Model Loading │    │                 │
-│ - Character     │    │ - LoRA Switching│    │ - Qwen2 Base    │
-│   Selection     │    │ - TTS Service   │    │ - Character     │
-│ - Audio Toggle  │    │ - WebSocket     │    │   Adapters      │
-└─────────────────┘    └─────────────────┘    └─────────────────┘
-```
+The app will automatically download the base model from HuggingFace on first run. This may take a few minutes.
 
 ## Characters
 
 1. **Moses** - Biblical prophet with wisdom and authority
-2. **Samsung Employee** - Tech-savvy corporate representative
+2. **Samsung Employee** - Tech-savvy corporate representative  
 3. **Jinx** - Complex, chaotic character from Arcane
 
-## Setup
-
-1. Install dependencies:
-```bash
-pip install -r requirements.txt
-```
-
-2. Download models (see Resources section)
-
-3. Start backend:
-```bash
-cd backend
-python main.py
-```
-
-4. Start frontend:
-```bash
-cd frontend
-# Instructions for your chosen frontend framework
-```
-
-## Requirements
-
-- Python 3.8+
-- CUDA-compatible GPU (recommended)
-- ~8GB RAM minimum
-- ~20GB storage for models
-
-## Models Used
+## How It Works
 
-- Base: Qwen2 (via Polarium/qwen2-yoda-lora)
-- Voice: Microsoft VibeVoice-1.5B
-- Custom LoRA adapters for each character
+- Uses a shared base model (Qwen3-0.6B) from HuggingFace
+- Switches between character-specific LoRA adapters
+- LoRA adapters are tiny (few MB each) vs full models (GB each)
+- Efficient memory usage by sharing one base model

app.py

@@ -17,12 +17,18 @@ def initialize_models():
     global character_manager
     if character_manager is None:
         print("🔄 Initializing character manager...")
-        character_manager = CharacterManager()
-        loop = asyncio.new_event_loop()
-        asyncio.set_event_loop(loop)
-        loop.run_until_complete(character_manager.initialize())
-        loop.close()
-        print("✅ Models loaded successfully!")
+        try:
+            character_manager = CharacterManager()
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            loop.run_until_complete(character_manager.initialize())
+            loop.close()
+            print("✅ Models loaded successfully!")
+        except Exception as e:
+            print(f"❌ Error initializing models: {e}")
+            import traceback
+            traceback.print_exc()
+            raise
 
 def chat_function(message, character_id, history):
     """Handle chat with character - returns updated history and empty message"""