Model architecture
Parameter count
What is parameter count?
Parameter count is the number of learned weights in a model, usually quoted in billions. It is a rough proxy for capability and a good proxy for memory requirement, but a poor predictor of performance on any specific task.
Training data quality, architecture and post-training all move the relationship substantially. Models that outperform others twice their size are common enough that size alone is a weak argument.
For a task-specific fine-tune, the right size is the smallest one that clears your accuracy bar. Everything above that is a recurring cost with no return.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.