fix: cap token×beam product so generation fits its GPU reservation

fix: harden generation params, fix swap RTL, polish from review

feat: expose advanced generation params on the public /translate API

feat: add honesty captions, RTL output, empty guard, advanced params

fix: greedy-decode by default; apply review cleanups

fix: use bfloat16 instead of float16 on CUDA for T5 stability

docs: record ZeroGPU pickle and T5 attention constraints in app.py

perf: engage ZeroGPU pack/stream, tune GPU duration, fix requirements

feat: expose stable /translate API endpoint and tidy UI

refactor: use concrete T5 types and run ty at full strictness

fix: remove invalid Textbox clear button and tighten model/tokenizer types

fix: address code review findings

feat: add symmetric language dropdowns with swap button

feat: update language mappings for new dict shape with region sorting

feat: update browser tab title to MADLAD-400 Translate

feat: switch theme from Soft to Ocean

refactor: move translate button below both columns

refactor: remove hint and character count from UI

fix: comma-format character count for consistency

fix: use Gradio CSS variables for info text styling

feat: move translate button into input column

refactor: move hint and char count from textbox info to gr.HTML element

feat: enforce 2000 character limit on input

refactor: extract _format_info helper for info string

feat: move hint and char count into input textbox info

fix: use change event for char count to handle inline clear

feat: add loading feedback on Translate button

feat: replace standalone Clear button with inline clear on input

feat: add character limit display and keyboard shortcut hint

feat: increase textbox height to 6 lines and remove input placeholder

feat: replace English label with static dropdown for visual symmetry

feat: add UI improvements for usability

feat: rewrite UI with Google Translate-style layout

refactor: remove batch translation code

feat: redesign Batch tab with two-column layout matching Single tab

fix: replace deprecated torch_dtype with dtype in model loading

feat: add preview table to batch tab UI

feat: translate_batch returns table data alongside file path

feat: rebuild UI with sidebar and tabbed Single/Batch layout

feat: add translate_batch function for file-based translation

feat: add batch file parsing and output helpers

feat: add generation parameters to translate function

fix: update app UI and examples for Tier 1 languages

refactor: cache language mappings and use model.device

fix: remove MPS support and use float32 on CPU

fix: add space between language token and input text

fix: expose demo at module level for Gradio hot reload

fix: move theme parameter from Blocks() to launch()

feat: replace gr.Interface with gr.Blocks for improved UI

Update app for evaluated-only language set