Text Generation Demos (PyTorch + HF Transformers)

This notebook bundles four focused text-generation experiments:

1. Char-RNN from scratch (PyTorch)

   • Custom RNNmodel using nn.RNN → nn.Linear
   • Trains on a short text snippet (e.g., Alice-like sample)
   • Next-character prediction with cross-entropy

2. Toy Character GAN

   • Generator and Discriminator are small nn.Sequential MLPs
   • Operate on the character vocabulary vector
   • Prints a few generated characters (illustrative, not SOTA)

3. GPT-2 Free-form Sampling

   • GPT2Tokenizer, GPT2LMHeadModel
   • Generate with controls like temperature, top_k, top_p, no_repeat_ngram_size, max_length

4. T5-Small Translation

   • T5Tokenizer, T5ForConditionalGeneration
   • Example prompt: translate English to French: 'Hello, ...'

5. BLEU & ROUGE with torchmetrics

   • Quick, reference vs generated comparison to sanity-check outputs

---

Quickstart

# 1) Create & activate a virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS/Linux
source .venv/bin/activate

# 2) Install deps
pip install -r requirements.txt

# 3) Launch
jupyter lab  # or: jupyter notebook

4) Git clone
git clone https://github.com/Joe-Naz01/pytorch_text_gem.git
cd pytorch_text_gem