Spaces:
Sleeping
Sleeping
Commit History
-
fixing some issues including browse section disappear and whenever the image is replaced or cleared the vlm_usage_state because that usage belongs to the previous image and VLM caption. When the replacement image gets its new caption, vlm_usage_state will be populated with the new API call’s usage.
8aa9bea
-
Require VLM feedback before submission, localize feedback validation, and track OpenAI token usage in dataset JSON
6e56c0c -
Merge branch 'main' of hf.co:spaces/team-wonders/ema-dev
# Conflicts: # ui/layout.py62266f8 -
Fix VLM caption generation variable, remove high-quality wording from examples, and enforce instant scroll-to-top on step transitions
- Restore API response variable 'raw' before polishing inside generate_caption(). - Remove 'of high-quality submissions' / 'de alta calidad' from Examples_Intro across metadata.json and layout.py. - Update scroll_to_top_js to force instant 0,0 scroll across all window and container elements, running on click and via post-render timeouts in step navigation chains.12af6d2 -
Fix VLM caption generation variable, remove high-quality wording from examples, and enforce instant scroll-to-top on step transitions
- Restore API response variable 'raw' before polishing inside generate_caption(). - Remove 'of high-quality submissions' / 'de alta calidad' from Examples_Intro across metadata.json and layout.py. - Update scroll_to_top_js to force instant 0,0 scroll across all window and container elements, running on click and via post-render timeouts in step navigation chains.d502924 -
Ensure step navigation smoothly scrolls up to the top of the page
- Set scroll_to_output=False and attach js=scroll_to_top_js on step1_next_btn, step2_next_btn, step2_back_btn, and step3_back_btn click handlers. - Prevents Gradio from overriding smooth scroll-to-top behavior when transitioning between wizard steps.17beff5 -
Fix VLM prompt version and model assignment recording & clean up word cap post-processing
- Remove unused caption_word_target and caption_word_max settings from data/vlm_prompts.yaml and logic/openai_vlm.py. - Simplify polish_caption() to avoid truncating valid multi-sentence v1/v2 cultural descriptions by word count while preserving sentence fragment cleanup. - Remove vlm_model_dropdown.change passthru trigger in ui/layout.py so vlm_model_state retains the exact model name and prompt version tag (e.g. 'gpt-5-mini (v1)') set during caption generation.09619f6
-
Fix duplicate box borders by resetting nested step-description-box styles
e8ce9f9 -
Wrap step descriptions in styled box containers
4a1bba3 -
Support versioned overrides (v0, v1, v2) for localized prompts
9f52beb -
Support random selection of active prompt versions and save chosen version
3b232d2 -
Rename default, default_step3, and default_step3_detailed to v0, v1, and v2
f1a7bc7 -
Increase word count limits for step 3 prompts to 80-100 words
76ad6a3 -
Add default_step3 and default_step3_detailed templates to vlm_prompts.yaml
ee8c49e -
Update instructions.
d3ecac1 -
Add is_positive field to USA and Spain examples JSON files
6a5b90c -
Fallback to default VLM model if loaded model is obsolete or empty
9792065 -
Remove Image and Description tip cards from examples modal
4573f14 -
Move locale selection to first page and position dropdowns next to each other
f0ea1f2
-
Replacing the image in a loaded sample now triggers a fresh VLM caption without clearing the existing annotation fields.
34ce212 -
several fixes - see slack
555a9ea
-
fixing categorical value choices
450579b -
solving merging conflicts
bbec658 -
scroll to top after next
8d7c479
-
Fix examples image load
Lea3808104 -
Fix load image
Lea0feeada -
Upload example1.png
Verified signatureb6f3d10 -
Add files for per country examples, update Spanish example and fix vlm voting language
Leaa10d356
-
Add draft of negative example
Leabc6047e -
add black forest cake example
Lea9d81dc1 -
Multilingual table headers
Lea7cc7939 -
Multilingual stats tables on user page
Leae8933f3 -
Add multilingual step instructions
Lea1cd24e1 -
Multilingual user page and other language fixes
Lea1dc92be -
Spanish translations
Lea19cbbf3
-
Enhance input validations, add step 3 image preview, and style user back button
- Disable Submit button until required inputs (including description) are filled unless Quick Submit is enabled. - Add image preview component to step 3 layout, keeping it synchronized with the step 1 loaded image. - Set user page back button variant to primary. - Update selection page Annotate button to variant primary.f83cc97 -
Update AI Description layout, instructions, and translation values
- Remove empty VLM Caption Accordion wrapper and nest elements directly inside the AI Description Group block. - Update language configuration handler to refresh the Markdown title value instead of updating an Accordion label. - Update VLM_Description translation value across all locales in metadata.json to describe state-of-the-art AI models.a5e00c4 -
Configure back buttons layout, colors and update metadata tab translations
- Set variant of global Back button to huggingface (orange). - Hide step-specific full-width Back buttons on step 2 and step 3 layout sections. - Update Tab_Examples metadata key translation across all locales to 🖼️ Examples.973f8b7 -
Update guidelines layout, metadata translation structure, and USA English guidelines
- Integrate dynamic Image and Description instructions accordions into main page UI layout and handler state. - Update all 66 country/language metadata configurations in data/metadata.json with new Image and Description detailed instructions. - Ensure proper JSON escaping for unescaped nested double quotes in metadata accordion descriptions. - Revise USA English guidelines (instructions.md, step2.md, step3.md) to define new category structure and Visual Evidence requirements.6b04780
-
More minor formatting
Leaea63d83 -
Minor formatting
Lea6c78a13 -
Soften wording for steps 1&3
Lea15ab7e0 -
metadata fix
Lea42dfd5e -
More instruction changes
Leab3501bf -
Updated instructions
Lea304a47e
-
fix: replace component instantiation with gr.update() in all event handlers
- Replace all Modal(visible=True/False) lambdas with gr.update(visible=...) - Replace gr.Markdown() / gr.Dropdown() / gr.Column() / gr.Textbox() returns in callbacks with gr.update() - Guard update_language against None country/language and fix KeyError - Clean up language_outputs by removing deleted components, resolvingNameError - Ignore local_data/ in .gitignore40eae7a -
generate captions automatically and country/language selection skip
35e91ca -
Update data/guidelines/USA/English/intro.md
Verified signature8fcf911 -
Update data/guidelines/USA/English/intro.md
Verified signaturebe3b9f3
-
Update with new intructions
Lead8db908