Parameters
Parameters are the internal values of an AI model and are often used as a rough indicator of model size.
Parameters are numerous numeric values that are adjusted through learning within an AI model. It can be compared to each dial on a huge mixing console, and learning is the process of finely tuning billions of these dials to match the data. The commonly seen expression 7B model means that there are 7 billion such adjustment values.
The number of parameters is used as a representative indicator of the scale and potential of the model. In general, the more parameters there are, the greater the expressive power, but the memory and computational costs required for learning and execution also increase, so it is a reference point when choosing a model.
However, the number of parameters does not directly determine the performance ranking. There are increasing cases of small models beating large models depending on the quality and quantity of learning data and training method, making it dangerous to evaluate models based solely on numbers.
✅ Why it matters
- It serves as a standard for estimating the scale of the model and the required hardware specifications
- Allows you to read model names and specifications such as 7B and 70B
- It is a basic concept for understanding the relationship between model performance and cost.
⚠️ Limits and debates
- A larger number of parameters does not necessarily mean better performance
- The impact of data quality and learning method is not revealed in the numbers
- Major commercial models do not disclose the number of parameters, making direct comparison difficult