A deep learning project that generates text sequences using a Gated Recurrent Unit (GRU) based Recurrent Neural Network (RNN).This project demonstrates Text Generation using a GRU-based Recurrent Neural Network. It learns from a given text corpus and generates new text word-by-word. The project includes data preprocessing, model training, text generation, and deployment via a Streamlit web app. 🚀

---

🔗 Live Demo

Try the deployed Streamlit app here:
Text Generation using GRU Model — 🌐 Live Demo

---

🚀 Project Overview

(https://github.com/shivareddy2002/GRU-Text-Generation/blob/main/UI_Galary/text_generation_flow.png).

The workflow involves the following key steps:

1️⃣ Importing Dependencies

Load essential libraries such as TensorFlow/Keras, NumPy, Matplotlib and other utilities for data processing and model building.

2️⃣ Loading the Text Corpus

Provide a custom text file as input for the model to learn from.

3️⃣ Preprocessing the Data

• Cleaning the text (lowercasing, removing unwanted characters)
• Tokenizing words and converting them into numerical sequences
• Creating training sequences and applying padding for uniform input length

4️⃣ Building the GRU Model

• Embedding layer for word representations
• GRU (Gated Recurrent Unit) layers to capture sequential patterns
• Dense output layer with softmax activation for word prediction

5️⃣ Training the Model

• Loss Function: Categorical Crossentropy
• Optimizer: Adam
• The model is trained to predict the next word given a sequence of previous words.

6️⃣ Generating Text

• Provide a seed text to the trained model
• The model predicts and appends words iteratively to create meaningful text.

7️⃣ Deploying with Streamlit

A simple web interface built with Streamlit allows users to input a seed text and instantly generate new text outputs.

---

🚀 Key Features

• Efficient GRU architecture for sequence learning.
• Beam Search for better text coherence.
• Customizable text corpus and hyperparameters.
• Interactive Streamlit web app for real-time text generation.
• Lightweight and fast training compared to LSTMs.
• Supports checkpoint saving and model reuse.

---

🧰 Dependencies

• Python
• TensorFlow / Keras
• NumPy
• Pandas
• Matplotlib
• Streamlit

---

🔄 Customization

• Dataset: Replace data/input.txt with your own text corpus.
• Model Hyperparameters: Modify embedding size, GRU units, and number of layers in model.py.
• Generation Settings: Change beam_width or temperature for different text creativity.

---

📈 Training & Evaluation

• Model is trained using categorical crossentropy and Adam optimizer.
• Can evaluate model loss and perplexity during training.
• Supports checkpoint saving and loading to avoid retraining.

---

💡 Tips for Better Text Generation

• Provide longer seed text for more coherent output.
• Adjust temperature:
  • temperature < 1 → more predictable text
  • temperature > 1 → more creative text
• Use beam search for improved sequence generation.

---

📊 Model Details

• Architecture: GRU layer(s) + Dense output layer
• Input: Tokenized and padded sequences of text
• Output: Probability distribution over the vocabulary for next word prediction
• Loss: Categorical Crossentropy
• Optimizer: Adam

Why GRU?

• Fewer parameters than LSTM → faster training ⚡
• Retains capability to capture long-term dependencies ✅
• Ideal for lightweight text generation

---

💾 Dataset

• Any text corpus can be used (books, articles, plays, custom corpora). 📚
• Preprocessing: tokenization, sequence creation, padding.
• Replace the dataset to generate text in different styles/tones.

---

⚙️ Usage

1. Open the Streamlit web application.
2. Enter a seed text (starting phrase) in the input box. 📝
3. Specify the number of words you want the model to generate. 🔢
4. Adjust the temperature value to control the creativity of the generated text.
   • temperature < 1 → more predictable text
   • temperature > 1 → more creative text
5. Click the Generate button to produce the text. ▶️
6. The model will generate text word-by-word based on the seed text.
7. You can copy or download the generated text for further use.

---

Example Output

• Seed Text: "Once upon a time"
• Number of Words: 10
• Temperature: 0.8

Generated Text:
"Once upon a time in a land far away there lived a wise king..."

---

🖼️ Visual Workflow

   flowchart LR
    %% --- Data Preparation Stage ---
    subgraph DP[📂 Data Preparation]
        A["📦 Importing Required Libraries"]
        B["📚 Loading Input Text Corpus"]
        C["✂️ Preprocessing\n(Cleaning + Tokenization + Creating Sequences + Padding)"]
    end

    %% --- Modeling & Training Stage ---
    subgraph MT[🤖 Modeling & Training]
        D["🏗️ GRU Model\n(RNN Layers + Dense)"]
        E["⚡ Training\n(Categorical Crossentropy + Adam)"]
    end

    %% --- Text Generation Stage ---
    subgraph TG[🚀 Text Generation ]
        F["✍️ Text Generation\n(Seed + Predicted Words)"]
    end

    %% --- Deployment Stage ---
    subgraph DS[🚀  Deployment]
        G["🌐 Streamlit Deployment\n(Interactive Web App)"]
    end

    %% --- Flow Connections ---
    A --> B --> C --> D --> E --> F --> G
    %% Styles
    style A fill:#FFD54F,stroke:#F57F17,stroke-width:2px,color:#000;
    style B fill:#4FC3F7,stroke:#0277BD,stroke-width:2px,color:#fff;
    style C fill:#AED581,stroke:#33691E,stroke-width:2px,color:#000;
    style D fill:#BA68C8,stroke:#4A148C,stroke-width:2px,color:#fff;
    style E fill:#FF8A65,stroke:#BF360C,stroke-width:2px,color:#fff;
    style F fill:#90CAF9,stroke:#0D47A1,stroke-width:2px,color:#000;
    style G fill:#F44336,stroke:#B71C1C,stroke-width:2px,color:#fff

Loading

👨‍💻 Author

Lomada Siva Gangi Reddy

• 🎓 B.Tech CSE (Data Science), RGMCET (2021–2025)
• 💡 Interests: Python | Machine Learning | Deep Learning | Data Science
• 📍 Open to Internships & Job Offers

Contact Me:

• 📧 Email: lomadasivagangireddy3@gmail.com
• 📞 Phone: 9346493592
• 💼 LinkedIn 🌐 GitHub 🚀 Portfolio

---