manishw10/devgen-trocr-devanagari-lora

Stability: Reverted to Gradio 3.50.2 with Premium CSS Styling

Final: Stable All-In-One Suite (UI + Debug + Scoring + Fixes)

Fix: Event context and re-enable premium Gradio 4.x theme

Layout: Move visual diagnostic to main screen and preserve all logic

Feature: Premium SVG Confidence Gauge and math-backed scoring

Design: Final Premium Suite with Pro Mode Toggle

Fix: Restore 1.75 AR threshold for character routing

Diagnostic: Add visual before/after comparison to UI

Final Fidelity: Mirror local Preprocess-Route-Recognize pipeline

Deep Alignment: Merged LoRA weights and synced token configs

Restore: 1:1 copy of original backend routing and preprocessing logic

Fix: Corrected routing logic and added manual mode override

Premium Update: Glassmorphism UI and training-accurate preprocessing

Feature: Full Smart Pipeline with Auto-Routing and Preprocessing

Feature: Integrated CNN model and optimized TrOCR with Beam Search

Robust: Bypass PEFT generate wrapper to fix inference crash

Fix: Upgrade PEFT and use explicit pixel_values for inference

Universal: Fix Peft error and add local/HF detection

Fix: Quote python_version to avoid 3.1 error

Final: Stable Gradio 3.x revert

Fix: Surgical patch to enable Recognize button

Robust: Monkey-patch Gradio API schema bug

Fix: Use quoted version strings for HF config

Robust Fix: Pin env to Python 3.9 and Gradio 4.30

Fix: Strip examples to bypass schema bug

Upgrade: Use modern Gradio with stable Interface

Fix: Switch to stable Gradio 3.x Interface

Fix: Disable show_api to bypass Gradio bug

Fix: Stable Gradio 4.40.0 and server binding

Fix: Sync SDK version with requirements

Fix: Add fn and outputs to gr.Examples

Fix: Pin huggingface_hub to resolve import error

Fix python version quotes

Fix: Pin python version and update model link

Initial upload: TrOCR LoRA adapter and Gradio demo