Qwen2 72B Instruct
Alibaba · 2024-06-07 · 72.7B parameters
Qwen2-72B-Instruct is Alibaba's 72-billion-parameter instruction- and preference-aligned language model from the Qwen2 series, released June 7, 2024. It is a dense Transformer decoder with GQA, SwiGLU, RoPE, and RMSNorm; it has a 32,768-token native context and is documented for extrapolation to 131,072 tokens with YaRN. It was post-trained with supervised fine-tuning and Direct Preference Optimization and supports multilingual understanding and generation, coding, and mathematics.