Anil Turaga

Fine-tuning and Distillation of Large Language Models

Anil Turaga • June 5, 2025

Video and reference links for a one hour talk I gave as part of the Dynamic Talk series at Grid Dynamics

Resources for fine-tuning LLMs

Author: Anil Turaga's LinkedIn

Slides: Dynamic Talk on Fine-tuning LLMs – PDF

Libraries for finetuning
Infrastructure
  1. Colab by Google – Free T4 for up to 4 hours/day. colab.research.google.com
  2. RunPod – Wide array of GPU options. runpod.io
  3. Hyperscalers – GPU-enabled ML services on major clouds.
  4. Local – Nvidia GPU equipped machines. macOS support
Example use cases covered in the talk
  1. Continued pre-training for another language – Notebook, Datasets: Wikipedia, Alpaca Korean chat
  2. Supervised fine-tuning for code edit prediction – SFT Notebook, DPO Notebook, Dataset: Zeta
  3. Reinforcement learning with GRPO for writing style – Training Script, Datasets: Style Classification, Chat Questions
OpenAI fine-tuning as a service
Open source models
  1. Vision: Qwen-VL
  2. Reasoning: Qwen-3
  3. Coding: Qwen-Coder
Miscellaneous