Disable Tesseract by default for GPU-first benchmark runs

Deploy CUDA-enforced runtime changes for OCR

Enforce strict GPU startup + runtime checks

Hard-fail non-CUDA model placements in strict mode

Fix startup indentation and keep strict CUDA policy in sync

Enforce strict CUDA path for local OCR inference

Force ZeroGPU CUDA inference path and direct-model device hardening

Harden image/model input parsing for stable Space runs

Fix comparison runtime robustness for zero-gpu and API route

Hardened precheck and runtime recovery

Fix local transformer bootstrap and dependency baseline

Fix local inference runner and result collection

Patch: fix dependency/runtime fallbacks for OCR models

Fix syntax error in pipeline dtype loop

Compatibility fixes for local OCR models

Use local ZeroGPU offline inference only

Add PaddleOCR + DeepSeek OCR models and raise default output token budget

Scope GPU lease to active inference runs

Remove persistent GPU warmup; rely on sleep timeout for idle GPU control

Treat tesseract as low VRAM workload in scheduler

ZeroGPU compute-mode scheduling + safety caps

Handle starlette request-first TemplateResponse for gradio root

Fix gradio template response argument handling

Simplify TemplateResponse shim context normalization to plain dict

Harden TemplateResponse shim for mapping-context calls

Fix template context injection in Starlette Jinja shim

Re-add scoped safe Jinja2 get_template shim to fix dict template name

Remove gradio template monkeypatches to avoid root template runtime errors

Keep safe get_template only; remove safe TemplateResponse monkeypatch

Remove Jinja template monkeypatches, keep chainable undefined fallback

Debug template context payload once

Remove debug prints after template fix

Fix template debug patch indentation

Add one-time template context debug logging

Set Gradio Jinja undefined to ChainableUndefined

Fix safe template context assignment precedence

Fix gradio template context injection for zero gpu space

Add fallback template context defaults to avoid config undefined at startup

Add @spaces.GPU warmup marker for zero-GPU startup

Fix Gradio template path handling and localhost check for Zero-GPU launch

Restore Gradio runtime with GRADIO_SHARE-safe launch and template compatibility patch

Switch Space UI to Streamlit to avoid gradio runtime errors on zero-a10g

Fix gradio launch path and gradio version for HF runtime

Patch gradio template handling for dict name bug in runtime

Use share=True only to satisfy ZeroGPU launch requirement

Remove unsupported share=True launch on HF space

Reupload app.py after accidental runtime mismatch

Try gradio 4.43.0 to avoid runtime API/template exceptions

Hide event endpoint from API schema to avoid bool-schema crash

Pin run_comparison return annotation to avoid API schema inference crash