LLM Training Time and Cost Calculator

Calculate both the training time and cost for large language models (LLM) with parallel computing support.

Input Parameters:

  • GPU Selection: Choose from various GPU models with different compute capabilities
  • Number of GPUs: Specify how many GPUs to use in parallel
  • Model Size: Number of parameters in billions
  • Dataset Size: Number of tokens in your dataset in billions
  • Training Epochs: Number of times to iterate over the dataset
  • Utilization Rate: Expected GPU utilization (typically 0.4-0.7)
  • Overhead: Additional time/cost factor for data loading, checkpointing, etc.
Ouputs:
  • Estimated Training Time: Total days and hours required for training
  • Estimated Training Cost: Total cost in dollars based on GPU hours
Modified from this Hf Space.
Select GPU
Select Precision
0.1 1
1 2

Improved with good intentions by ghost.