llama-3.2-1b-aiml-assistant

🤖 Llama 3.2 1B — AI/ML Assistant

A domain-specific AI/ML assistant fine-tuned on the Open-Platypus dataset using QLoRA on Google Colab (T4 GPU).

Built by Aryan Tyagi — AI/ML Engineer specializing in LangChain, LangGraph, and Agentic AI.

---

🧠 What it does

Answers questions about Artificial Intelligence and Machine Learning clearly and accurately — including topics like:

• Supervised vs Unsupervised Learning
• Neural Networks and Backpropagation
• Overfitting, Regularization, Dropout
• Optimization algorithms (SGD, Adam, etc.)
• Transformers, Attention Mechanisms
• RAG, LLMs, Fine-tuning concepts

---

⚙️ Training Details

---

🚀 How to Use

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "unsloth/Llama-3.2-1B-Instruct",
    load_in_4bit=True
)

# Load fine-tuned adapter
model = PeftModel.from_pretrained(
    base_model,
    "aryantyagi0/llama-3.2-1b-aiml-assistant"
)
tokenizer = AutoTokenizer.from_pretrained(
    "aryantyagi0/llama-3.2-1b-aiml-assistant"
)

prompt = """Below is a question about Artificial Intelligence or Machine Learning.
Provide a clear, accurate and detailed answer.

### Question:
{}

### Answer:
{}"""

inputs = tokenizer(
    [prompt.format("What is backpropagation?", "")],
    return_tensors="pt"
)

outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

📊 Training Loss Curve

Loss converged from 1.02 → 0.63 — indicating strong domain adaptation.

---

👤 Author

Aryan Tyagi — AI/ML Engineer
🔗 GitHub · LinkedIn

Specializes in LangChain · LangGraph · RAG · Agentic AI · LLM Fine-tuning

---

📄 License

This model is fine-tuned from Llama 3.2 and follows the Llama 3.2 Community License.