initial commit

app.py

@@ -0,0 +1,100 @@
+# -*- coding: utf-8 -*-
+"""app.py.ipynb
+
+Automatically generated by Colab.
+
+Original file is located at
+    https://colab.research.google.com/drive/16d-Y-lJ3S-vY04H28dnBtHqjBTz3_lQh
+"""
+
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
+import os
+import pandas as pd
+
+
+# Load DeepSeek-Math 7B Instruct model and tokenizer
+model_name = "deepseek-ai/deepseek-math-7b-instruct"
+try:
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
+    model.generation_config.pad_token_id = tokenizer.eos_token_id
+    print("DeepSeek-Math-7B-Instruct loaded successfully!")
+except Exception as e:
+    print(f"Error loading model: {e}")
+    model = None
+    tokenizer = None
+
+# Load training data from math_meme_training.txt
+training_file = "math_meme_training.txt"
+few_shot_prompt = "Below are examples of correcting incorrect math memes with step-by-step reasoning:\n\n"
+if os.path.exists(training_file):
+    with open(training_file, 'r') as f:
+        lines = f.readlines()
+        # Take up to 3 examples to keep prompt manageable
+        examples = []
+        for i in range(0, min(6, len(lines)), 2):  # Step by 2 for Prompt/Response pairs
+            if lines[i].startswith("Prompt: ") and i+1 < len(lines) and lines[i+1].startswith("Response: "):
+                prompt = lines[i].replace("Prompt: ", "").strip()
+                response = lines[i+1].replace("Response: ", "").strip()
+                examples.append(f"{prompt}\nResponse: {response}")
+        few_shot_prompt += "\n".join(examples[:3]) + "\n\nNow, correct the following incorrect math meme:\n"
+else:
+    few_shot_prompt += "No training data found. Using default behavior.\n\nNow, correct the following incorrect math meme:\n"
+    print("Warning: math_meme_training.txt not found. Using minimal prompt.")
+
+# Function to generate correction
+def generate_deepseek_correction(user_input):
+    if model is None or tokenizer is None:
+        return "Error: Model not loaded. Please check the Space logs and ensure a GPU is enabled."
+
+    # Validate input format
+    if '=' not in user_input or len(user_input.split('=')) != 2:
+        return "Invalid input. Please enter in the format 'expression = wrong_answer' (e.g., '8 ÷ 2(2+2) = 1')."
+
+    # Construct prompt with training data
+    meme_prompt = f"Correct this incorrect math meme: '{user_input}'\nPlease reason step by step and provide the correct answer with an explanation."
+    full_prompt = few_shot_prompt + meme_prompt
+
+    try:
+        # Generate response
+        inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=150,
+            temperature=0.7,
+            do_sample=True,
+            eos_token_id=tokenizer.eos_token_id
+        )
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        correction = response[len(full_prompt):].strip()
+
+        if not correction:
+            correction = "DeepSeek-Math failed to generate a meaningful correction."
+
+        # Add funny error rating
+        error_rating = "Funny Error Rating: '90% fixing memes, 10% wondering why fractions are so hard'"
+        return f"{correction}\n\n{error_rating}"
+    except Exception as e:
+        return f"Error processing input: {str(e)}"
+
+# Gradio interface
+interface = gr.Interface(
+    fn=generate_deepseek_correction,
+    inputs=gr.Textbox(
+        label="Enter an Incorrect Math Meme",
+        placeholder="e.g., '7 + 2 × 3 = 27' or '1/2 + 1/2 = 0.25'"
+    ),
+    outputs=gr.Textbox(label="Corrected Answer and Explanation"),
+    title="Math Meme Repair with DeepSeek-Math",
+    description="Input an incorrect math meme (e.g., '8 ÷ 2(2+2) = 1') and get the correct answer with an explanation!",
+    examples=[
+        ["7 + 2 × 3 = 27"],
+        ["4 ÷ 2(1+1) = 1"],
+        ["1/2 + 1/2 = 0.25"]
+    ]
+)
+
+# Launch the interface
+interface.launch(server_name="0.0.0.0", server_port=7860)

math_meme_training.txt

@@ -0,0 +1,80 @@
+Prompt: Correct this incorrect math meme: '6 ÷ 2(1+2) = 1.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 9.0. Explanation: PEMDAS: (1+2) = 3, then 6 ÷ 2 × 3 = 3 × 3 = 9, left-to-right.
+
+Prompt: Correct this incorrect math meme: '8 ÷ 2(2+2) = 1.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 16.0. Explanation: PEMDAS: (2+2) = 4, then 8 ÷ 2 × 4 = 4 × 4 = 16.
+
+Prompt: Correct this incorrect math meme: '5 + 3 × 2 = 16.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 11.0. Explanation: PEMDAS: 3 × 2 = 6, then 5 + 6 = 11. Multiply first.
+
+Prompt: Correct this incorrect math meme: '2³ = 6.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 8.0. Explanation: Exponents mean 2 × 2 × 2 = 8, not 2 × 3.
+
+Prompt: Correct this incorrect math meme: '1/2 + 1/2 = 0.25'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 1.0. Explanation: Add numerators: 1+1 = 2, keep denominator 2, so 2/2 = 1.
+
+Prompt: Correct this incorrect math meme: '9 - 3 ÷ 1/3 = 0.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 0.0. Explanation: 3 ÷ 1/3 = 9, then 9 - 9 = 0. Division by fraction flips it.
+
+Prompt: Correct this incorrect math meme: '4 + 2² = 36.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 8.0. Explanation: PEMDAS: 2² = 4, then 4 + 4 = 8. Exponents first.
+
+Prompt: Correct this incorrect math meme: '10 ÷ 5 × 2 = 1.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 4.0. Explanation: PEMDAS: 10 ÷ 5 = 2, then 2 × 2 = 4, left-to-right.
+
+Prompt: Correct this incorrect math meme: '3 + 4 × 5 = 35.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 23.0. Explanation: PEMDAS: 4 × 5 = 20, then 3 + 20 = 23.
+
+Prompt: Correct this incorrect math meme: '5² = 10.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 25.0. Explanation: 5² = 5 × 5 = 25, not 5 × 2.
+
+Prompt: Correct this incorrect math meme: '6 ÷ 3 + 1 = 3.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 3.0. Explanation: 6 ÷ 3 = 2, then 2 + 1 = 3. Division first.
+
+Prompt: Correct this incorrect math meme: '2 + 2 × 2 = 8.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 6.0. Explanation: PEMDAS: 2 × 2 = 4, then 2 + 4 = 6.
+
+Prompt: Correct this incorrect math meme: '1/3 + 1/3 = 0.333'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 0.667. Explanation: 1/3 + 1/3 = 2/3 ≈ 0.667, not 2/6 without simplifying.
+
+Prompt: Correct this incorrect math meme: '7 - 2 × 3 = 15.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 1.0. Explanation: PEMDAS: 2 × 3 = 6, then 7 - 6 = 1.
+
+Prompt: Correct this incorrect math meme: '4 ÷ 2 + 2 = 3.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 4.0. Explanation: PEMDAS: 4 ÷ 2 = 2, then 2 + 2 = 4.
+
+Prompt: Correct this incorrect math meme: '3² + 1 = 10.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 10.0. Explanation: 3² = 9, then 9 + 1 = 10. Exponents first.
+
+Prompt: Correct this incorrect math meme: '8 ÷ 4 × 2 = 1.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 4.0. Explanation: PEMDAS: 8 ÷ 4 = 2, then 2 × 2 = 4, left-to-right.
+
+Prompt: Correct this incorrect math meme: '5 × 2 - 3 = 13.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 7.0. Explanation: PEMDAS: 5 × 2 = 10, then 10 - 3 = 7.
+
+Prompt: Correct this incorrect math meme: '1 + 1 ÷ 2 = 1.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 1.5. Explanation: PEMDAS: 1 ÷ 2 = 0.5, then 1 + 0.5 = 1.5.
+
+Prompt: Correct this incorrect math meme: '9 ÷ 3(3) = 1.0'
+Please reason step by step and provide the correct answer with an explanation.
+Response: Correct answer: 9.0. Explanation: PEMDAS: 9 ÷ 3 × 3 = 3 × 3 = 9, left-to-right.
+

requirements.txt

@@ -0,0 +1,3 @@
+gradio
+transformers
+torch