Set fallback chat template on both processor and tokenizer to resolve apply_chat_template error

Use AutoModelForImageTextToText and MODEL_NAME google/gemma-4-E2B per user request

Add sentencepiece and protobuf tokenizer dependencies and debug logging in app.py

Optimize RAM usage on CPU load by enabling low_cpu_mem_usage

Migrate architecture from Ollama to native Hugging Face transformers with CPU optimization

Add live chat UI with request/response log panel

Optimize: reduce context to 2048 + pre-warm model at startup

Add /chat and /vision SSE streaming endpoints for probar_ia.html

Convert roots and routes to pure JSON API

Install Ollama and run gemma4:e2b with FastAPI wrapper

Add application files