PyTorch
Deep learning framework I use for real model work — fine-tuning a time-series foundation model and PPO implemented from scratch.
PyTorch is an open-source deep learning framework in Python with tensors, automatic differentiation and GPU support; most open LLM training and fine-tuning tooling is built on it.
How I use it
- Fine-tuning a time-series foundation model. In my Real-time Data Systems work I fine-tuned IBM Granite TSPulse with PyTorch on GPU (Google Colab), with hyperparameter grids and evaluation against baselines.
- Reinforcement learning from scratch. I implemented PPO directly in PyTorch — no RL framework — against a purpose-built simulation environment, and retrained the same policy with Evolution Strategies when per-step rewards were too noisy.
- Foundations. PyTorch is also part of my Deep Learning grounding from the IBM Generative AI Engineering Professional Certificate professional certificate, where it is used to build data loaders, train language models and implement Transformers — notably in Generative AI and LLMs: Architecture and Data Preparation, Generative AI Foundational Models for NLP & Language Understanding and Generative AI Language Modeling with Transformers.
Why it matters
Knowing how tensors, loss functions and training loops work makes the behaviour of hosted models easier to reason about — from tokenization limits to why Fine-Tuning changes what a model will and will not do.