Decoder-Only
Decoder-only models (GPT-style) apply a causal mask so tokens attend only leftward. The same stack handles pretraining and generation.
They dominate open and closed LLM deployments due to scaling simplicity and unified training.
Decoder-only models (GPT-style) apply a causal mask so tokens attend only leftward. The same stack handles pretraining and generation.
They dominate open and closed LLM deployments due to scaling simplicity and unified training.