Fix: Align Gradio sdk_version in README.md

Add health check endpoint

Fix: Add root_path to Gradio mount for asset resolution

Add gradio to requirements.txt

Revert: Mount Gradio back to root path

Fix: Mount Gradio at /gradio and remove redundant print

Fix: Use CPU if CUDA is not available

Use ZeroGPU with lazy model loading for free tier

Restore TranslateGemma-4B-IT with T4 GPU hardware

Switch to NLLB-200-distilled-600M for fast CPU inference

Switch to HTTP streaming transport for MCP

Add MCP server with ZeroGPU support for TranslateGemma

Lazy load model to allow Gradio to start immediately

Simplify to single Interface - text translation only

Use gr.Interface instead of gr.Blocks to fix schema bug

Remove Examples component to fix Gradio schema bug

Use Python 3.11 and Gradio 5.9.0 for compatibility

CPU-only version for 16GB RAM Space

Add @spaces.GPU decorator for ZeroGPU with duration=120

Use Gradio 6.3.0 with dedicated A10G hardware

Use Gradio 4.31.0 with @spaces.GPU decorator

Use Gradio 5.9.0 for stability

Pin Gradio 4.44.0 for stability

Disable SSR mode to fix HTTP response errors

Fix ZeroGPU: load model on CPU, move to CUDA in GPU-decorated functions

Add @spaces.GPU decorator for ZeroGPU support

Add HF token authentication for gated model access

Initial commit: TranslateGemma-4B-IT Gradio demo

initial commit