Upload 3 files

README.md

@@ -1,5 +1,5 @@
 ---
-title: Gemini Distractor Generator
+title: Gemini + Claude Distractor Generator
 emoji: 🧠
 colorFrom: indigo
 colorTo: pink
@@ -9,43 +9,51 @@ app_file: app.py
 pinned: false
 ---
 
+# Gemini + Claude Distractor Generator
 
-# Gemini Distractor Generator
-
-This is a simple web app that uses Gemini 1.5 Flash (via Gemini API) to generate multiple-choice distractors for a given question. It automatically identifies the correct answer and returns three plausible distractors.
+This is a web app that uses **Gemini 1.5 Flash (Free Tier)** to generate multiple-choice distractors and then uses **Claude 3.5 Sonnet (via OpenRouter)** to **rank the distractors** based on how confusing they are.
 
 ## Features
-- Uses Gemini 1.5 Flash (Free Tier) via API
-- Auto-generates distractors based on question context
-- Also attempts to identify the correct answer
-- Includes minimal Gradio UI
+
+- Generates distractors using Gemini 1.5 Flash
+- Uses Claude Sonnet 3.5 to evaluate and rank distractors (LLM-as-Judge)
+- Displays the correct answer, distractors, and Claude's ranking
+- Simple and clean Gradio UI
 
 ## Files
-- `app.py`: Main Gradio application.
-- `requirements.txt`: Required Python libraries.
-- `README.md`: This file.
+
+- `app.py`: Main Gradio application logic
+- `requirements.txt`: Dependencies
+- `README.md`: This file
 
 ## How to Run
 
 ### Locally
-1. Install the dependencies:
+
+1. Install dependencies:
    ```bash
    pip install -r requirements.txt
    ```
 
-2. Run the app:
+2. Set your API keys inside `app.py`:
+   - Gemini API key (for generation)
+   - OpenRouter API key (for ranking via Claude)
+
+3. Run:
    ```bash
    python app.py
    ```
 
-3. Visit the local URL shown by Gradio.
+4. Open the Gradio link shown in your terminal.
 
 ### On Hugging Face Spaces
-1. Upload all three files to your Hugging Face Space.
-2. Choose `Gradio` as the SDK.
-3. Make sure to set your Gemini API key in a secure way (e.g., using secrets or hardcoded for testing).
-4. Set `app.py` as the entry point.
+
+1. Upload `app.py`, `requirements.txt`, and `README.md`
+2. Choose `Gradio` as the SDK
+3. Make sure to set your **Gemini API key** and **OpenRouter API key** via HF secrets or hardcode for testing
+4. Set `app.py` as your main file (`app_file`)
 
 ## Notes
-- Gemini 1.5 Flash is selected to stay within free-tier quotas.
-- You may need to set your API key inside `app.py` manually.
+
+- Claude is used only to rank distractors; Gemini handles generation.
+- Free-tier Gemini is slower and limited — ranking may take longer due to OpenRouter limits.

app.py.py

@@ -0,0 +1,106 @@
+
+import gradio as gr
+import google.generativeai as genai
+import requests
+import json
+import time
+
+genai.configure(api_key="GEMINI_API_KEY")
+gemini_model = genai.GenerativeModel("gemini-2.0-flash")
+
+class OpenRouter:
+    def __init__(self, model_name, key, role="user"):
+        self.model_name = model_name
+        self.key = key
+        self.api_url = "https://openrouter.ai/api/v1/chat/completions"
+        self.role = role
+
+    def get_response(self, prompt):
+        headers = {
+            "Authorization": f"Bearer {self.key}",
+            "Content-Type": "application/json"
+        }
+        payload = {
+            "model": self.model_name,
+            "messages": [{"role": self.role, "content": prompt}]
+        }
+        response = requests.post(self.api_url, headers=headers, data=json.dumps(payload))
+        if response.status_code == 200:
+            return response.json()["choices"][0]["message"]["content"]
+        else:
+            raise Exception(f"Request failed: {response.status_code} - {response.text}")
+
+def call_openrouter(llm, prompt, retries=5, delay=5):
+    for attempt in range(retries):
+        try:
+            return llm.get_response(prompt)
+        except Exception as e:
+            print(f"Retry {attempt+1}: {e}")
+            time.sleep(delay)
+    return "ERROR"
+
+def prepare_claude_ranking_prompt(question, correct, distractors):
+    distractor_text = "\n".join([f"Choice {i+1}: {d}" for i, d in enumerate(distractors)])
+    return f"""You are a helpful assistant evaluating multiple-choice distractors.
+
+Question: {question}
+Correct Answer: {correct}
+
+Here are the distractors:
+{distractor_text}
+
+Your task: Rank the distractors from most to least confusing.
+
+Only respond with one line in this format: 2 > 1 > 3
+"""
+
+
+def generate_and_rank(question):
+    gemini_prompt = f"""You are an assistant generating multiple-choice questions.
+
+Given a question, generate:
+- One correct answer
+- Three plausible but incorrect distractors
+
+Format:
+Correct Answer: ...
+Distractor 1: ...
+Distractor 2: ...
+Distractor 3: ...
+
+Question: {question}
+""" 
+    gemini_response = gemini_model.generate_content(gemini_prompt).text.strip()
+
+    lines = [line.strip() for line in gemini_response.splitlines() if line.strip()]
+    correct = next((line.split(":", 1)[1].strip() for line in lines if line.startswith("Correct Answer")), "N/A")
+    distractors = [line.split(":", 1)[1].strip() for line in lines if line.startswith("Distractor")]
+
+    if len(distractors) != 3:
+        return "Error: Could not extract exactly 3 distractors.", gemini_response, ""
+
+    claude_llm = OpenRouter("anthropic/claude-3.5-sonnet-20240612", "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")
+    claude_prompt = prepare_claude_ranking_prompt(question, correct, distractors)
+    ranking = call_openrouter(claude_llm, claude_prompt)
+
+    output = f"**Question:** {question}\n\n"
+    output += f"**Correct Answer:** {correct}\n"
+    for i, d in enumerate(distractors, 1):
+        output += f"**Distractor {i}:** {d}\n"
+    output += f"\n**Claude Ranking:** {ranking.strip()}"
+
+    return output, gemini_response, claude_prompt
+
+
+iface = gr.Interface(
+    fn=generate_and_rank,
+    inputs=gr.Textbox(label="Enter your MCQ Question"),
+    outputs=[
+        gr.Textbox(label="Final Output"),
+        gr.Textbox(label="Raw Gemini Output"),
+        gr.Textbox(label="Claude Prompt")
+    ],
+    title="Confusing Distractor Generator + Claude Ranking"
+)
+
+iface.launch()

requirements.txt

@@ -0,0 +1,4 @@
+google-generativeai>=0.3.2
+gradio>=4.28.3
+python-dotenv
+requests