Qwen/Qwen3-14B-Base

Qwen3-14B-Base is a 14.8 billion parameter causal language model developed by Qwen, pre-trained on 36 trillion tokens across 119 languages. This model features a three-stage pre-training process focusing on broad language modeling, reasoning skills (STEM, coding), and long-context comprehension up to 32k tokens. It incorporates architectural refinements like qk layernorm and scaling law-guided hyperparameter tuning for improved stability and performance. Qwen3-14B-Base is designed for general knowledge acquisition and advanced reasoning tasks.