gurumurthy3/cooking-recipe

🍪 Cooking Recipe GPT — Small Causal Language Model

This is a small GPT-style causal language model trained on structured cooking recipe data. It follows a standard GPT-2 architecture with minimal custom tweaks. The model can generate complete ingredient lists and directions from a given title or partial prompt like "No-Bake Nut Cookies".

---

🔧 Model Architecture

• Base: GPT-2 like architecture
• Total Parameters: ~3.1M
• Number of Layers (Blocks): 2
• vocab_size:4,000
• Embedding Dimension: 384
• Hidden Dimension: 512
• Attention Heads: 4
• Max Sequence Length: 512
• Position Encoding: Sinusoidal
• Normalization: LayerNorm
• Activation Function: ReLU
• Dropout: 0.1

---

📚 Training Details

🧠 Pretraining

• Training Data: 180,000 structured recipes
• Validation Data: 20,000 recipes
• Epochs: 4
• Training Loss: 1.7928
• Validation Loss: 1.7219

🔄 SFT (Supervised Fine-Tuning)

• SFT Training Data: 50,000 recipes
• SFT Validation Data: 5,000 recipes
• SFT Training Loss: 1.6551
• SFT Validation Loss: 1.6066

---

🚀 Example Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("gurumurthy3/cooking-recipe", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("gurumurthy3/cooking-recipe")

prompt = "No-Bake Nut Cookies"
input_ids = tokenizer(prompt, return_tensors="pt").input_ids

output = model.generate(
    input_ids,
    tokenizer,
    max_new_tokens=250,
    temperature=1.0,
    top_k=50,
    eos_token_id=tokenizer.eos_token_id
)

print(output)