Fix: Use Qwen3-0.6B (correct model) with proper PEFT adapter switching via set_adapter()

Add lightweight character manager - uses one base model with adapter swapping for HF Spaces

🔧 Fixed LoRA adapter loading and Gradio interface

Fix character isolation: create separate model instances to prevent adapter interference

Implement proper adapter switching instead of multiple models

Fix adapter interference: create dedicated model instances per character

Fix syntax error: add missing except clause and fix indentation

Comprehensive fix: Update transformers for Qwen3 support and improve LoRA adapter compatibility

CRITICAL FIX: Use Qwen/Qwen3-0.6B to match LoRA adapter training base model

Fix base model to Qwen2.5-0.5B and clean LoRA adapter config compatibility

Fix LoRA adapter compatibility issues and add robust fallbacks

Add complete LoRA adapter files and base model configuration for proper character loading

Deploy Roleplay Chat Box - optimized version without large files