AI glossary · Models
Model parameters
In one sentence
A model's parameters are its internal numeric values, mainly the weights of a neural network, which are adjusted during training and store what it has learned. Their count is used as a rough measure of model size.
Also known as: AI parameters · model weights · weights · parameter count
What are model parameters?
Picture a mixing desk with billions of dials. At the start they’re all set at random and all you hear is noise. Training means nudging each dial a fraction, millions of times, until the music sounds right. Those dials are the parameters, and where they end up is everything the model knows.
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingIn a neural network, the parameters are mostly the weights on the connections between neurons, plus some offset values called biases (a technical term that has nothing to do with AI bias). The model doesn’t store sentences or documents; it stores these numbers.
Related to this term
UTIO agents
- Available 24/7 in 26 languages
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
- During training, the parameters change. Each example in the dataset nudges their values to reduce the error. See model training.
- During use, the parameters stay fixed. Every answer is the result of running the input through them. See inference.
Parameter count has become shorthand for model size. For a historical reference point, GPT-3 (2020) had 175 billion. Many developers no longer publish the figure for their latest models, and size no longer tells the whole story: architectures such as mixture of experts only switch on part of their parameters for each response.
Size has practical consequences. Every parameter takes up memory: at 16 bits each, a 7-billion-parameter model needs around 14 GB just to load. Quantisation lowers that precision so the model fits on more modest hardware, at the cost of a little quality.
Example
A vet practice in York wants an assistant to sort client messages into emergency, appointment, repeat prescription or other. It tries a large general-purpose model and a small one. For this particular job the small one is almost as accurate, answers faster and costs less per message.
For writing long, careful replies to questions about treatment, though, the large model makes a difference. The lesson: parameter count is a useful indicator, not a quality score.
What it means for your business
You don’t need to compare models by parameter count to serve your customers. What matters is an answer that’s correct, quick and based on your information.
With UTIO’s agents, such as the voice agent or the WhatsApp agent, the model is included and there’s nothing to compare. If you build with the AI API, you choose the model tier (basic, standard or superior): a practical way to decide how much capability each task needs without getting into billions of parameters.
Frequently asked questions
What does 7B or 70B mean for an AI model?
It's the number of parameters in billions. A 7B model has around 7 billion parameters; a 70B model, around 70 billion.
Do more parameters mean a better model?
Not necessarily. More parameters give a model more capacity, but quality also depends on the data and how it was trained. A small, well-trained model can beat a large one at a specific task, and it's faster and cheaper to run.
What's the difference between parameters and hyperparameters?
Parameters are learned by the model itself during training. Hyperparameters are set by a person before training, such as the number of layers or the learning rate. Settings like the temperature of a response aren't model parameters either; they're options you choose when using it.
Related terms
- AI model
An AI model is the program you get by training an algorithm on data: a set of learned parameters that, given a new input, produces an output such as a prediction, a label, a piece of text or an image.
- LLM
An LLM (large language model) is an AI model trained on vast amounts of text to predict the next word. That single skill lets it draft, summarise, translate, classify and hold a conversation.
- Quantization
Quantization is a technique that stores an AI model's parameters as lower-precision numbers, for example 8 or 4 bits instead of 16. The model takes up less memory and often runs faster, in exchange for a loss of quality that can be small.
- Mixture of experts (MoE)
Mixture of experts (MoE) is an AI model architecture split into several sub-networks called experts. For each token, a router switches on only a few of them, so the model can be very large without using all of its compute at every step.
- Small language model (SLM)
A small language model (SLM) is a language model with far fewer parameters than an LLM. It knows less about everything, but it replies faster, costs less and can run on a modest server, a laptop or even a phone.
Updated on