Spaces:
Running on Zero
Running on Zero
| """Plain-language question editing and handoff to an app builder.""" | |
| import html | |
| import json | |
| from bongard.data import DataError, strict_loads, validate_question | |
| KIND_LABELS = {"choice": "Choose one option", "noul": "Yes or no", "score": "Rate on a scale"} | |
| def label(key): | |
| text = key.replace("_", " ").strip() | |
| return text[:1].upper() + text[1:] | |
| def text_value(value): | |
| return value if isinstance(value, str) else json.dumps(value, ensure_ascii=False) | |
| def load_questions(text): | |
| try: | |
| questions = strict_loads(text or "") | |
| except (ValueError, DataError) as exc: | |
| raise DataError(f"The question JSON could not be read: {exc}") from None | |
| if not isinstance(questions, dict) or not 1 <= len(questions) <= 8: | |
| raise DataError("Include 1–8 questions in this request.") | |
| for key, question in questions.items(): | |
| validate_question(question, key) | |
| if question["type"] == "choice" and not 2 <= len(question["criteria"]) <= 32: | |
| raise DataError("A choice question needs 2–32 answer options.") | |
| return questions | |
| def question_preview(text): | |
| try: | |
| questions = load_questions(text) | |
| except DataError as exc: | |
| if (text or "").strip() == "{}": | |
| return '<div class="question-empty">Include at least one question in the JSON.</div>' | |
| return f'<div class="inline-error" role="alert">{html.escape(str(exc))}</div>' | |
| rows = [] | |
| for question in questions.values(): | |
| kind = question["type"] | |
| detail = KIND_LABELS[kind] | |
| if kind != "noul": | |
| count = len(question["criteria"]) | |
| detail += f" · {count} {'options' if kind == 'choice' else 'levels'}" | |
| rows.append( | |
| f'<div class="question-summary"><span>{html.escape(detail)}</span>' | |
| f"<p>{html.escape(text_value(question.get('instructions')))}</p></div>" | |
| ) | |
| return ( | |
| f'<div class="question-list"><div class="questions-count">{len(questions)} questions · one request</div>' | |
| + "".join(rows) | |
| + "</div>" | |
| ) | |
| def handoff(state_text, questions_text, has_image=False, calibrated=True, rotations=False): | |
| questions = load_questions(questions_text) | |
| image = 'handle_file("your-photo.png")' if has_image else "None" | |
| code = ( | |
| "# Install once: pip install gradio_client\n" | |
| "from gradio_client import Client, handle_file\n\n" | |
| 'client = Client("AgentBull/bongard-mini")\n' | |
| f"state_text = {state_text!r}\n" | |
| f"questions_text = {json.dumps(questions, ensure_ascii=False)!r}\n\n" | |
| "_, _, response = client.predict(\n" | |
| f" state_text, questions_text, {image}, {calibrated}, {rotations},\n" | |
| ' api_name="/decide",\n' | |
| ")\n" | |
| 'print(response["answers"])\n' | |
| ) | |
| prompt = ( | |
| "Add Bongard-mini decision judgments to my app using the tested request below. " | |
| "For a prototype, call the public Hugging Face Space AgentBull/bongard-mini with " | |
| "the official Gradio client for my app's stack: gradio_client for Python or " | |
| "@gradio/client for JavaScript. Use the '/decide' endpoint. The arguments, in order, " | |
| "are state_text, questions_text " | |
| f"(JSON string), image (optional), calibrated={calibrated}, rotations={rotations}. The third return " | |
| "value (Python prediction[2], JavaScript result.data[2]) is the model response; " | |
| "use its 'answers' dictionary and preserve its full " | |
| "probability distributions. Handle response['error'] and waiting for a GPU. " | |
| "Keep credentials on the server and do not bake a Hugging Face token into the frontend. " | |
| "For deployment, use the downloadable model and public runtime at " | |
| "https://huggingface.co/AgentBull/bongard-mini and https://github.com/AgentBull/bongard.\n\n" | |
| f"Context:\n{state_text}\n\nQuestions:\n{json.dumps(questions, ensure_ascii=False, indent=2)}" | |
| ) | |
| if has_image: | |
| prompt += "\n\nThis request also requires an image. Let the user attach it as image 1." | |
| return prompt, code | |