AI Model Training & Fine-Tuning Budgets: Tokens, Compute, GPU-Hours, Experiment Grids & Scaling Curves
Budget an AI training run before you start it: dataset tokens, compute, GPU-hours, fine-tuning costs, experiment grids and scaling curves, with every cell checked as you go.
One-time payment · lifetime access · lessons, labs, exam and certificate included · no subscription.
About this course
This course turns an AI training project into spreadsheet work you can check. For a fictional lab you count dataset tokens per source with deduplication and a validation split, estimate training compute with the 6 × parameters × tokens rule of thumb in scientific notation, and convert FLOPs into GPU-hours, wall-clock days and a cluster size for a deadline. You price the run on on-demand, reserved and interruptible plans, budget a fine-tuning job with low-rank adapters and memory per parameter, size an experiment grid with early stopping, fit a power-law scaling curve with LN and SLOPE to predict loss, and close with a full project budget and chart.
Lessons
- Dataset Tokens: Documents, Words, Deduplication & the Validation Split Pro
- Training Compute: The 6 × Parameters × Tokens Rule & Scientific Notation Pro
- GPU-Hours, Utilisation & Cluster Size for a Deadline Pro
- On-Demand, Reserved or Interruptible: Pricing the GPU-Hours Pro
- Fine-Tuning Budgets: Epochs, Low-Rank Adapters & Memory per Parameter Pro
- Experiment Grids: Learning Rates × Batch Sizes × Seeds and Early Stopping Pro
- Scaling Curves: Fitting a Power Law and Predicting Loss at a Bigger Budget Pro
- The Project Budget: Checkpoints, Storage, Evaluations, Contingency & the Chart Pro