Parameter Scale
The total number of trainable weights in a model, a basic measure of model capacity and complexity that is generally related to model capability and generalization performance.
Core Facts
Timeline (last 90 days)
Parameter scale refers to the number of trainable weights in a model and is a rough indicator of model capacity.
Generally speaking, models with more parameters can extract more general representations from the same data, and thus are more likely to generalize hidden skills from unrelated data.
Parameter scale and problem-setting ability exhibit a non-monotonic relationship.
Parameter scale mainly affects a model's capacity and knowledge density, but engineering deployment capability depends more on the training data mix, post-training (instruction fine-tuning, RLHF), and context management at inference time.
In video-language models, parameter scale and proactive capability are not simply positively correlated: larger models are usually stronger at perception and commonsense reasoning, but 'when to respond' is a temporal decision-making problem that is often influenced more by training objectives and data distribution.
The parameter scale of a model is expressed in B (billions); for example, 7B represents 7 billion parameters.
All Facts (6)
The parameter scale of a model is expressed in B (billions); for example, 7B represents 7 billion parameters.
75%UnverifiedParameter scale refers to the number of trainable weights in a model and is a rough indicator of model capacity.
50%UnverifiedGenerally speaking, models with more parameters can extract more general representations from the same data, and thus are more likely to generalize hidden skills from unrelated data.
50%UnverifiedParameter scale and problem-setting ability exhibit a non-monotonic relationship.
50%UnverifiedParameter scale mainly affects a model's capacity and knowledge density, but engineering deployment capability depends more on the training data mix, post-training (instruction fine-tuning, RLHF), and context management at inference time.
50%UnverifiedIn video-language models, parameter scale and proactive capability are not simply positively correlated: larger models are usually stronger at perception and commonsense reasoning, but 'when to respond' is a temporal decision-making problem that is often influenced more by training objectives and data distribution.
50%